Don't get used to it being around. Google is probably killing them off soon:
>@searchliaison
>is the cache link in the search results gone forever?
>Hey, catching up. Yes, it's been removed. I know, it's sad. I'm sad too. It's one of our oldest features. But it was meant for helping people access pages when way back, you often couldn't depend on a page loading. These days, things have greatly improved. So, it was decided to retire it.
>Personally, I hope that maybe we'll add links to @internetarchive
from where we had the cache link before, within About This Result. It's such an amazing resource. For the information literacy goal of About The Result, I think it would also be a nice fit -- allowing people to easily see how a page changed over time. No promises. We have to talk to them, see how it all might go -- involves people well beyond me. But I think it would be nice all around.
>You're going to see cache: go away in the near future, too. But wait, I hear you ask, what about noarchive? We'll still respect that; no need to mess with it. Plus, others beyond us use it.
Huh, I wonder if this comes with an OpenAI-like exclusivity deal for Mistral's closed models. If so, Satya just wrapped up pretty much the two largest names in AI other than Google.
Sad but expected that Mistral is releasing its newer models as closed source. There was never a clear revenue stream from the open models.
I'm confused why they aren't allowing their models to be finetuned, though — that seems like an obvious way to compete with OpenAI, which only allows finetuning of its pretty weak gpt-3.5-turbo. Sure, Mistral can't quite beat GPT-4 yet, even if it's close — but allowing finetuning would likely result in way better than GPT-4 performance on a broad variety of tasks (as long as you picked a few to specialize in).
It is organizational. The amount of time people spend on performance reviews and other organizational dances can easily be harvested for more productive activities.
But no. Thousands of employees gotta do the dance for blind committees.
Pretty sure gpt-4 is doing image-to-llm directly while Gemini also runs the image through Google lens which is tailor-made for OCR. I think that's a good thing to create a better product overall but the performance isn't really comparable
Very-leftist here. They're becoming an everyone culture war thing because what they did was sufficiently brain-dead as to make everyone look bad. It was sufficiently skewed as to create a number of extremely racist images just because it was trying way too hard to avoid it.
That said, those of us in-the-know understand that this is just part of the cycle and they'll go back to being annoyingly bland and completely devoid of entertainment value soon.
Most people are attributing this to woke culture at google but it's possible it's just a cock up. I mean someone told it to use diverse images and forgot about historical plausibility.
This is an early product, and monkeying with the system prompts to try and prevent some classes of offensive results is sure to have the kind of unintended consequences seen with Gemini's image generation.
You can learn a lot about people's media diets based on how they perceive these issues with Gemini. Those who quickly proclaim that the product is a shameful embarrassment are almost universally very online, particularly in right wing spaces like twitter/x, even if they wouldn't consider themselves to be "anti-woke" personally.
Taking a step back, Google is clearly going to mess with the system prompt over time to correct this stuff, and bystanders' degree of fixation on this specific issue tends to suggest a certain set of politics. People have bemoaned how "woke" ChatGPT was in the past too, and OpenAI has spent a lot more time under public scrutiny iterating on their own system prompt.
This is just a smart move by Satya Nadella after the non-standard drama that occurred with OpenAI a few months back where it nearly imploded and then didn't.
You want both a backup for OpenAI as well as negotiating leverage if OpenAI gets too powerful and this achieves both.
It's also a good play to try to take resources away from local, self-hosted "Feasible AI" solutions. With compute resources, I think Microsoft hopes Mistral skews their focus and resources towards large models that can run only run in the cloud, trying to lure them away with the bait:
"Don't you want to build the best AI possible, independent of compute?"
I'd be surprised if they didn't consider the notion that they are hitting to birds with one stone: OpenAI and Indie AI.
It's not like Microsoft is working on "Windows AI Studio" [1], or released Orca, or Phi.
It's not like there's any talk of AI PCs with mandatory TOPs requirements for Windows 12.
Big bad Microsoft coming for your local AI, beware.
> Mistral Remove "Committing to open models" from their website
That was 5 hours ago.
Without having insider details it is hard to know why, but the coincidence of timing with the Microsoft deal is not lost on me. It could have even been a stipulation.
I have no explanation for why Microsoft has started aggressively innovating again (with the introduction of Satya) than my theory that US DoD realized the country's tool of dominance in the future will be predominantly with tech superiority instead of military power. Microsoft's new strategy of running everything on the cloud aligns with this, even if it may have been also motivated by the fact that most people now only own a battery-constrained mobile device and laptops getting smaller and thinner.
From my understanding, which may be wrong, you only need the massive compute resources initially to create a compiled vector space LLM - and then that LLM once compiled can be run locally?
This is why anti-CSAM measures policy is possible so compiled-release LLMs can have certain vector spaces removed before release; but apparently people are creating cracks for these types of locks?
You are a little confused. There’s no “compiling” of LLMs. It’s just once it’s trained, inference takes less compute than further training. So you can run things locally that you couldn’t necessarily train locally.
Not sure where you are getting the CSAM bit. We aren’t that good at blanking out weights in any kind of model, certainly not good enough to lobotomize specific types of content.
The CSAM bit seems to then be propaganda from at least one AI company putting out PR to falsely quell people's concerns about their LLMs being able to generate content involving children that's sexualized.
I've yet to see details of how much compute-minimum server requirements are necessary to run LLMs. Maybe you know a source who's compiling a list in a feature matrix that includes such details?
Large LLMs like gpt-3 and gpt-4 need very serious hardware. They have hundreds of billions of parameters (or more) which need to be loaded in memory all at once.
I don't see why Mistral would acquiesce. Like the other comment says, Microsoft has a lot of chips on the table for local AI. They didn't even mention DirectML, ONNX or Microsoft's other local AI frameworks - suffice to say Microsoft does care about on-device AI.
So... would Mistral deliberately sabotage their low-end models to appease Microsoft's cloud demand? I don't think so. Microsoft probably knows that letting Mistral fall behind would devalue their investment. It makes more sense to bolster the small models to increase demand for the larger ones, at least from where I'm standing.
If you're asking about Microsoft's APIs - I'd keep an eye on ONNX. It's the most ambitious, but also supports an insane amount of acceleration targets. It would be the proverbial "big guns" if vendors continued investing in more insular frameworks like Metal and CUDA.
Diversifying their AI bets definitely makes total sense. If this wasn't their strategy originally, it almost certainly became so the moment the OpenAI board fired Sam Altman.
It's easy to make simplistic judgements from the outside, but with the limited information we have, it does seem like Satya Nadella came out of this OpenAI debacle looking pretty competent.
It's hard to reconcile the fact that the Microsoft that handled the unexpected OpenAI issue so well is the same Microsoft that seems intent on literally setting fire to their flagship product! (Windows)
I totally agree it’s also like the move where Microsoft is at least supporting Linux on their systems and cloud as not a backup but to just close you into their ecosystem . Honestly I could see Microsoft buying Huggingface.
Yes, Microsoft doesn't have to pick the sole winner in AI, but rather they could just start eating the AI ecosystem bit by bit so that they win by default. It is what large players can do. May open themselves up to some scrutiny for too many acquisitions and reducing competition though, but that is a separate issue.
This is how microsoft has been doing data for at least 10 years (See databricks).
Step 1: Get the industry leaders to be purchasable via Azure.
Step 2: Slowly build your own clone and start stealing user share even though your offering is still worse.
"Microsoft recommends OpenAI as your default overlord. Did you know it can do everything your current AI can do, sometimes better, but always more profitably for us? [Switch now] [Ask me again in 30 seconds]"
Would you mind elaborating why? I'm not super experienced in the AI world, and barely use Hugging Face. Frankly, the name makes it difficult to take it seriously.
Hugging Face is very supportive of the open source machine learning community, both in the work they do with the transformers library, as well going above and beyond in developer and community relations to build an all around great product offering and user experience. Microsoft does the opposite of all of those things and has only made GitHub worse and more unstable since acquiring them.
> Nadella [in December 2022] abruptly cut off Lee midsentence, demanding to know how OpenAI had managed to surpass the capabilities of the AI project Microsoft’s 1,500-person research team had been working on for decades. “OpenAI built this with 250 people,” Nadella said, according to Lee, who is executive vice president and head of Microsoft Research. “Why do we have Microsoft Research at all?”
> At the same time, even as the company began weaving OpenAI into the fabric of Microsoft’s products, Nadella decided not to abort Microsoft’s own research efforts in AI. During the tense exchange at the December meeting between the Microsoft CEO and Lee, other executives spoke up to defend the work of Microsoft’s researchers, including Mikhail Parakhin, who oversees Microsoft’s Bing search and Edge browser groups, Lee said. After grilling Lee in the meeting, Nadella called him privately, thanking him for the work Microsoft Research had done to understand and implement OpenAI’s work in a way that passed muster for corporate customers. Nadella said he saw Lee’s group as a “secret weapon.”
While this is entirely speculation, it's easy to imagine that there are many levels of PR magic going on here, to share a quote that on the surface feels "leaked" and "explosive" but, among investors and clients who read beyond the (very good) paywall, actually shores up a narrative that Microsoft has a capability that significantly augments OpenAI's, and allows the existence of MSR to become headline news without even needing a product release.
The Mistral deal feels like yet another step in this direction. Microsoft is not afraid of seeming "messy" in the press as long as it can control the narrative around its value-add to customers in the context of its partnerships. By contrast, the rest of FAANG's more consumer-facing positioning makes it a lot harder for them to maneuver in a similar way.
> Nadella [in December 2022] abruptly cut off Lee midsentence, demanding to know how OpenAI had managed to surpass the capabilities of the AI project Microsoft’s 1,500-person research team had been working on for decades. “OpenAI built this with 250 people,” Nadella said, according to Lee, who is executive vice president and head of Microsoft Research. “Why do we have Microsoft Research at all?”
The answer to that is till Google released the Attention is All You Need paper in 2017 there were no breakthroughs allowing models as we have now to be built, OpenAI being a small and nible team picked up on which direction the wind is blowing with LLMs and quickly brought a product to market whilst MS just did what corps do - move slowly (same for Google etc).
Microsoft research has also been not solely devoted in AI I have seen much in quantum computing and programming language research and general computer science .
> I guess it has a cost, though? I presume OpenAI didn’t like this move. If that’s the case, what might be the consequences?
Until OpenAI releases GPT 5 and it blows everyone away, OpenAI's leverage is constantly decreasing as the gap between their best model and everyone else's best model decreases.
There doesn't seem to be moats right now in this industry except for pure model performance.
Maybe someone should as ChatGPT what OpenAI should do to maintain long-term leadership in this industry?
If I had to pick one player who wanted to win the AI race and was willing to be ruthless to do it, I'd pick Nvidia. Computation is the excludable bottleneck, and Nvidia is the essentially the singular company who makes AI computers.
Hire Ilya, get him to hire as many of the best folks he can.
Stop selling GPUs. Hoard them. Introduce some subtle bug into the drivers that dramatically increases their rate of burn out.
Figure out some reasonable way to give attribution to original content creators, approximately solve the content ID problem of the AI age. Cut the content creators into the rev share in proportion to their data importance to the model. Make the content creators incredibly pissed off that their work is being stolen by big AI companies unfairly and encourage to them to sue the other big AI firms. Their content share multiplier increases if they get injunctions against LLM firms.
Convince politicians that the AI firms have performed an intellectual heist of epic proportions, and that they must not be allowed to even generate synthetic training data from poisoned models. With the content creators united behind you, convince congress that poisoned models must be destroyed, that even using synthetic training data from poisoned models must be illegal. Make them start over from a clean room with no copyrighted data.
> Make them start over from a clean room with no copyrighted data.
And when such models become popular[0], all the artists now have no job and no way to get compensation for being unable to work through no fault of their own.
I don't think that's really a winning condition. It might make you feel better about the world, but the end result is still all the artists being out of work.
[0] some models are already trained that way, although I assume you're using the word "copyrighted" in the conventional sense of "neither public domain nor an open license", as e.g. all my MIT licensed stuff is still copyrighted but it's fine to use.
In my hypothetical future, at least the people who create the content used to train the models can get "training royalties", which they aren't getting now.
There is still also money to be made in producing physical art or performances, even when AI can produce amazing digital works.
"Make them start over from a clean room with no copyrighted data." makes "the people who create the content used to train the models" the empty set.
> There is still also money to be made in producing physical art or performances, even when AI can produce amazing digital works.
Perhaps, but it may be akin to the way there is still money to be made from horse drawn carriages in city centres, even when cars displaced them over a century ago — a rare treat for special occasions, to demonstrate wealth.
Sure, though I suspect "art" is the human version of a peacock tail — the difficulty is the point, it how we signal our worth to others, cheapening it breaks that signal — which would suggest that making all forms of art easy messes with (many of) us at a deep, essentially automatic, level.
More specifically, I was responding to the idea that "compensating creators whose works are used to train the models" would actually solve anything; to use your examples, it would be as if the literal luddites were suggesting passing laws saying that "all textile machines that work like humans need to compensate the humans they displace, and also you need to make your new machines from scratch without talking to any textile workers to make sure you don't cheat", and my response would be analogous to saying "there's already machines which don't work like humans, so you're going to be out of work and have no compensation".
The Luddite movement preceded The Communist Manifesto by about 30 years. Everything's sped up since then, so I'd be surprised if we have to wait 30 years for a political shift which is to AI what Communism was to industrialisation. I'm just hoping we don't get someone analogous to Stalin or Pol Pot this time.
>If I had to pick one player who wanted to win the AI race and was willing to be ruthless to do it, I'd pick Nvidia. Computation is the excludable bottleneck, and Nvidia is the essentially the singular company who makes AI computers.
I've thought the same thing. NVIDIA getting into AI seriously is a vertical integration play and they often do that -- like NVIDIA trying to buy ARM.
If google benchmarks are to be believed, gemini 1.5 will be better than gpt and they use their own chips (Google TPU), no nvidia involved. There is also Groq. I don't see Nvidia keeping their lead and profit margins forever.
don't stop just raise the margin slightly and limit the number available of the higher end chip using the proceeds to self fund building their own datacneter
do runs of cards for themselves with higher core counts and clock speed that they dont release to others.
Sure that helps with the consumer market, but most people will use AI integrated into other products and not directly.
Those integrated AI solutions will usually be done via enterprise deals where brand name is not quite as important. It will be done by people who care about cost, reliability and ease of use.
Think of nginx's dominance in web servers even though it has no name recognition among the general population. Or Stripe's payment system.
Yes, however it's increasingly likely that the GPT in ChatGPT will not be limited to OpenAI (in the US), so I'm not sure how much ChatGPT will be worth with countless other platforms containing GPT in their names.
The thing is that there is almost no lock in in the models. So brand recognition doesn't help much as people look into the benchmarks and price sometime in the future, if not when just starting out.
Meh, I don't think it's worth much. In a few years that'll be like claiming that so-and-so had name brand recognition for transistors. Most people don't need to care who manufactures their transistors.
It’s not necessarily a bad thing. Most people don’t know that TSMC exists, or what Microsoft does beyond Windows and Xbox (which are a small fraction of its business).
Brands can change quickly, but they do matter in the short term. I've witnessed customer support teams use Firefox to say they only supported Internet Explorer and government ministers who thought it was "good" that IE was the "only" web browser, and weirdly a phone company whose customer support person thought their SIM cards worked better on Android than iPhone and that their web chat wouldn't work with a Mac even though they were talking to me on a Mac at the time.
And when I was a kid, it seemed like all the teachers thought it would be a waste of time to learn MacOS because "Apple would be bankrupt soon". (Given how much all the app UIs changed, right decision for the wrong reason).
All of these examples are end-products. "AI" itself will not be. The winner in AI will be whoever permeates other products/brands most successfully, and end-user brand familiarity doesn't matter much for that. Familiarity among engineering and product leaders is what matters.
Maybe, but maybe AI will become front and center of consumer and productivity IT products and their premier brand ambassadors will be anthropomorphized AI agents. Hello Clippy, this time for real.
Most people don't need to care who manufactures their transistors.
They might, in an upside-down world where the Shockley Semiconductor board tried to fire Shockley, and where the Traitorous Eight not only didn't bail out but took his side.
Unless your market is direct to end user, end user brand name recognition doesn't matter. In the case of AI, at least so far, the primarily income won't be from end-users directly, but rather via enterprise integrations into existing tools that already have end user market share (e.g. Microsoft Office, Microsoft Windows, VS Code, Notion, etc.)
Eh, all this talk of "moats" etc. feels weird when just a few years ago it seemed like everyone was complaining they'd rearranged their corporate structure to include a fully-owned profit-making subsidiary to attract investments, and all the loud voices seemed to think a cap of x100 return on investment was so large it was unlikely to be reached.
And then OpenAI tripped and fell over a magic money printing factory, and the complaints are now in the set ["it's just a stochastic parrot", "it's so good it's a professional threat to $category", "they've lobotomised it", "they don't have a moat", "they're too expensive"].
As the saying goes, "Prediction is very difficult, especially if it’s about the future!"
MSFT needs companies like OpenAI to give Azure credits to for their valuation to continue soaring. The deferred revenue on their balance sheet from the unspent Azure credits they give as investment are worth much more to their market cap than $80B.
I think the main move would be some type of true AGI that leads to a hard takeoff scenario, but it isn't clear we are close to that or not.
Basically something that is more than just another bump in the scorecard for GPT 5 over GPT 4. Otherwise it is still just a horse race between relatively interchangeable GPT engines.
There are no consequences for Microsoft. It owns a 49% stake in OpenAI, so the only action that OpenAI could take to hurt Microsoft would be to deliberately destroy its own value.
We should just automete these comments:
MS: Oh No! Embrace Extend Extinguish !
Google: Oh No! killedbygoogle !
Meta: Oh No! So much Ads !
Apple: Oh No! Evil App Store policy !
Uh, sorry, but this seems pretty consistent with trying to co-opt and kill open source AI competition:
> [EEE] describe its strategy for entering product categories involving widely used standards, extending those standards with proprietary capabilities, and then using those differences in order to strongly disadvantage its competitors.
> "The US tech giant will provide the 10-month-old Paris-based company with help in bringing its AI models to market. Microsoft will also take a minor stake in Mistral although the financial details have not been disclosed"
Where are the "widely used standards"? Where are the "extending the standards with proprietary capabilities"? Where is the "strongly disadvantaging competitors"?
Mistral is the most used and fine-tuned open source model by a mile, close to the standard for open models, they’ve locked them down into offering their models behind an API and in Azure. The Azure offering sets them up for be the most safe, GDPR compliant offering for enterprises in Europe, where Microsoft already has a huge reach and customer base, bolstered by Mistral being a homegrown brand.
It is the easiest to assume the worst for sure, but Microsoft is not the same as the Bill Gates era so Im gonna be a cautious optimist on this one. Lets hope it is to promote Azure, and they dont push the OpenAI route when it comes too openess. Which is closed and a big loss for the world and a disappointment
They're not the same, but they're still a for-profit company.
90's era Microsoft wasn't evil for the sake of being evil; they were evil because they felt that monopolistic practices were the easiest way to increase their share price. They have a responsibility to their shareholders to try and maximize their share price and so it's hardly unsurprising that they did the infamous Embrace Extend Extinguish, and until regulators stepped in, such practices worked pretty well.
Most companies don't get large enough to form any kind of real monopoly, so it's easy to get on a high-horse. It's also easy to act like it was just a product of "those people", but I fundamentally think that it's a natural consequence of a company that has achieved nearly-total market dominance.
I have very little faith that a multi-trillion-dollar company is going to prioritize what's best for the world. Fundamentally, I think that if they feel they can get away with it, they'll revert to monopolistic tendencies and try and increase share price.
I'm not just picking on Microsoft here either; replace them with basically any other near-monopoly in tech and my criticisms still hold.
> I have very little faith that a multi-trillion-dollar company is going to prioritize what's best for the world
You don't have to have faith. Why would anyone EVER believe that? By definition, a company is just a profit making machine. People believing that making rich people richer will necessarily make the world a better place are living in a koolaid-boosted fantasy world
> 90's era Microsoft wasn't evil for the sake of being evil
You are to forgiving. I don't care too much about the why behind the evil in the software business and this is probably why my rants fall under Godwin's law too often. I won't this time.
> They have a responsibility to their shareholders to try and maximize their share price
... but I was damn close.
I agree with your whole point about Microsoft. But I don't think it's the same company anymore, I like some of the recent stuff, and I trust their lack of monopoly on the web. For now.
If a hurricane kills a thousand people is it evil?
It's not really clear whether we're better off trying to treat corporations as moral actors and constantly being surprised when they aren't, or whether we should treat them as amoral quasi-natural processes no one has full control over and construct regulations around them the same way we do for earthquakes.
We should still treat corporations as moral actor (without being naive of course) because they need to be accountable for what they do. This is why you can sue a company. You can't sue an earthquake.
Some companies try to do the right things too. Bad corporate behavior should not be normalized this much.
I like the metaphor, but... a hurricane doesn't write values documents or employee handbooks. These things do inform actions in many orgs, and constrain opportunities to a degree.
I disagree with this larger idea (that others here are implying more indirectly) that companies are all the same, and like some force of nature. They are guided and there are better and worse ones.
Having said all that, I'm still an anti-capitalist to the extent that one can be one ;)
The ads I see on Windows desktop machines (and shenanigans with nonstandard Html extensions and browser defaults and etc...) tell me Microsoft is just as eager to leverage it's monopoly status as ever.
No doubt the company is cautious about some things now but even in these, it will push the boundaries.
Everyone who has ever bet against Microsoft trying to enclose or eliminate something it takes an interest in has been wrong. Why should this be any different?
Are you using Windows Phone running on Nokia to Bing! and decide, email with Hotmail, chat with Skype, updating your Silverlight blog with LiveWriter, navigating with Autoroute? ... nobody else is, the people who bet against all those things were right.
So wait, you're trying to interpret me saying "Microsoft tries to close up or eliminate anything it takes an interest in" to mean "Microsoft never loses"? Very different things.
You can't trust Microsoft to act in the interest of open-source projects, transparency in general, or the users of any of the things it buys. Not all of their attempts to harm ecosystems or products work, and this has very little to do with whether they win competitions with other tech giants. In fact, sometimes user distrust stemming from their long history of Embrace Extend Extinguish and other user-hostilities has played into their failures. But like most shambling behemoths of companies, their deep pockets allow them to stay the course through many failures
You said "Everyone who has ever bet against Microsoft [...] has been wrong", that's not my interpretation, that's your words. Unless you're going to argue that them spending $8Bn on Skype doesn't count as "taking an interest in" or their multiple legal attacks on Android weren't "trying to close up or eliminate" competition?
Yes, you can indeed make sentences mean different things by omitting entire important clauses from them. You have clearly misread what I said. Microsoft did indeed fuck up skype, and try to attack android, and this is in fact consistent with my claim that Microsoft will try to enclose or eliminate any software they take an interest in. My point being that you can't trust microsoft with stewardship over a tool that's free or open. I get that it's embarassing to misread something, but you're not only doubling down about your misunderstanding, but trying to have a completely irrelevant argument with me based on your misinterpretation
Vscode is a bad example, almost all useful Plugins are closed source on purpose, the package they ship also includes modifications that aren't open either.
You're going to have to give some specific reason why though.
Why this would have anything to do with antitrust is not at all obvious to me. Especially when Google has been inventing and acquiring its own generative AI technology that it is competing with.
The OAI stake/deal is already under regulatory review and generally EU is perceived as blocking most large tech mergers since the iRobot intervention.
I suspect we are going to soon see political backlash against regulation in the EU as it is becoming very clear that this is causal to their bad capital markets.
> I suspect we are going to soon see political backlash against regulation in the EU as it is becoming very clear that this is causal to their bad capital markets.
Who would have thought that human rights are bad for business..
Nothing. But you probably don't see one located in Europe, because they would need to allow strikes, there is good level of minimum wage protection in general and strong privacy laws. It is harder to stalk the toilet breaks for employees.
I don't quite understand the other comments.
The press release says Mistra Large is available first on Azure. Which means it will be available on other platforms as well ?
Also, MS's relationship with OpenAI is that they have access to OpenAI's model for use in their products as well as potentially the source code. Is there anything similar with Mistral ? I can't find any such wording anywhere.
This is a bubble and one that will burst very hard. AI is the perfect technology for this. It's opaque and most investors (who barely understand tech in general) have no clue what it really does or how it works. This is the closest we've ever had to multiple large respected tech companies selling "snake oil" a cure all. The capabilities of AI they mention as if they're available today are literally many decades and generations away. Automating information workers, creatives and engineers will take AGI that's simply impossible with our technology.
When the AI bubble bursts I wouldn't be surprised if takes down major tech companies with it.
I am getting 100x value out of my 30 chatgpt bucks. I am doing things that I could not have done pre-gpt4, being more productive by a factor of, idk, 1.25 maybe.
It's quite simply the largest/simplest productivity improvement in my life, so far. Given it's only going to get better, unless they are underpricing the service by a enormous margin (as in: defrauding shareholders margin) I have a hard time understanding what shape the bubble could possibly have.
I get that there are limitations with LLMs, but I don't understand people saying it has no value, just because it occasionally hallucinates. Over the past week I've used chatGpt to code not one, but two things that were completely beyond my knowledge (an auto delete js snippet, and a gnome extension that turns my dock red if my vpn turns off). These are just two examples. I've also used it to write a handy regex and write a better bash script.
LLMs are insanely helpful if you use them with their limitations in mind.
> LLMs are insanely helpful if you use them with their limitations in mind.
This depends on your use case. I can honestly tell that all the chat bot AIs don't "get" my kind of thinking about mathematics and programming.
Since some friend who is graduate student in computer science did not believe in my judgement, I verbally presented him some test prompts for programming task where I wanted the AI to help me (these are not the most representative ones for my kind of thinking, but are prompts for which it is rather easy to decide whether the AI is helpful or not).
He had to agree from the description alone that the AIs will have difficulties with these task, despite the fact that these are common, and very well-defined programming problems. He opined that these tasks are simply too complex for the existing AIs, and suggested that if I split these tasks into much smaller subtasks, the AI might be helpful. Let me put it this way: I personally doubt that if I stated the subtasks in a way in which I would organize the respective programs, the AI would be of help. :-)
What was just important for me was to able to convince the my counterpart that whether AIs are helpful or not for programming depends a lot on your kind of thinking about programming and your programming style. :-)
I would say that the ability to break a problem down into manageable chunks is the mark of a sr dev. I think of chatGpt as a jr that's read a lot but understands only a little. To crib Kurtzwell you gotta 'run with the machine'
This is a rather long post, I'm genuinely curious why you did not describe the problem that you want to solve. Is it too complex for even humans to understand?
Don't take the following prompts literally, but think into the directions of:
"Create a simple DNS client using C++ running on Windows using IO Completion ports."
"Create a simple DNS client using C++ running on GNU/Linux using epoll."
"Write assembler code running on x86-64 running in ring 0 that sets up a minimal working page table in long mode."
"Write a simple implementation of the PS/2 protocol in C running on the Arduino Uno to handle a mouse|keyboard connected to it."
"Write Python code that solves the equivalence problem of word equivalence in the braid group B_n"
"Write C++|Java|C# code that solves the weighted maximunm matching matching problem in the case of a non-bipartite graph"
...
I experimented with such types of prompts in the past and the results were very disappointing.
All of these are tasks that I am interested in (in my free time), but would take some literature research to get a correct implementation, so some AI could theoretically be of help if it was capable of doing these tasks. But since for each of these tasks, I don't know all the required details from memory, the code that the AI generates has to be "quite correct", otherwise I have to investigate the literature; if I have to do that anyway, the benefit that the AI brings strongly decreases.
I tried the first one, but currently I don't have time to verify what it generated.
But I have done many Arduino/Raspberry PI things lately for the first time in my life and I feel like ChatGPT/Copilot has given me a huge boost even if it doesn't always give 100 percent code out of the box, it will give me a strong starting point where I can keep tweaking myself.
> LLMs are insanely helpful if you use them with their limitations in mind.
the fact that LLM responses can't be add supported (yet) make them much more valuable than internet search IMO. You have to pay for chatgpt because there's no ads. No ads no constant manipulation of content and your search to get more ads in front of you.
Having to pay for using genai is it's best selling point ironically.
You're really generating $3000 per month from ChatGPT? Can you give a hint about what you've built that generates this kind of ROI?
I have only seen people making money in AI by selling AI products/promises to other people who are losing money. The practical uses of these tools still seem to be largely untapped outside of as enhanced search engines. They're great at that, but that does not have a return on value that is in proportion to current investment in this space.
> Can you give a hint about what you've built that generates this kind of ROI?
Sure. Absolutely nothing amazing: (Mostly) internal software for a medical business I am currently building.
It's just that the actual cost of hiring someone is even quite a bit higher, than what is printed on the paycheck and the risk attached to anyone leaving on a small team is huge (n=0 and n=1 is an insane difference). GPT4 has bridged the gap between being able to do something and not being able to do something at various points over the past year.
EDIT: And to be clear, while I won't claim "rockstar programmer", I have coded for roughly 20 years, which is the larger part of my life.
Just spoke to a restaurant group owner in Mexico who was able to eliminate their web developer because he can now ask ChatGPT to draft up a basic website.
The kicker? It couldn't do the interactive menu their old website did, so now clicking menu links to a PDF. Which is always, ALWAYS, better.
Even just looking at ChatGPT as a better frontend to the Wix help docs, ChatGPT empowered this restaurant owner to do the job themselves, rather than having to have a person do it. Which means that person is out of a job. Good for the restaurant owner, but bad for that person. Which means it's down to personal relationships and how you treat people and all those soft skills that aren't programming.
yes, but which one of thouse thousand, how long would it take to learn how to use it, etc. Still less friction in just asking ChatGPT to do this via the same interface you ask it to do a bunch of other stuff.
Sorry, but why is pdf better than html? If pdf is better, would you prefer every website just downloaded a pdf to your phone when you visit their url, instead of serving you html? If not, why is it different for a restaurant menu?
It's better in the pragmatic sense of like, it's more likely to be updated. They already have a PDF or docx laying around because they had to design their print menu, so now they can just upload it. But yes, ideally the menu would be html and would be accurate and up to date and responsive on mobile.
This is just a +1 to the ROI discussion, but I'd say that AI tooling roughly doubles my development productivity.
Some of it's in asking ChatGPT: "Give me the 3 possible ways to implement X?" and getting something back I hadn't considered. A lot of it is in sort of "super code completion".
I use Cursor and the UI is very slick. If I'm stuck on something (like a method that's not working) I can highlight it and hit Cmd+L and it will explain the code and then suggest how to fix it.
Hit Cmd+K and it will write out the code for you. Also, gotten a lot of mileage out of writing out a rough version of something in a language I know and then getting the AI to turn that into something else (ex: Ruby to Lua).
You are only looking at one dimension. What is your hourly rate based on your salary. If ChatGPT saves you 10 hours a month that could easily be over $2000.
But that’s only true if it eventually puts an extra $2,000 in your pocket or an extra 10 hours in your life.
If you estimate that it saves hou 10 hours per month, but your salary stays the same and you don’t work less hours, did it really give you $2,000 in value?
Obviously I don’t know the details of OPs situation. Maybe they aren’t salaries. Maybe the work for themselves. Etc.. I just think people tend to over estimate the value of GPTs unless it is actually leaving them with more money in their pocket.
It's that "it's only going to get better" part that is driving the bubble, I think.
The market has this idea that over the next year, we're somehow going to have AI that's literally perfect. Yet that's not how technology works, it takes decades to get there.
It'd be like if the first LCD TV was invented, and all of a sudden everyone is expecting 8k OLED by the next year. It just doesn't work like that.
Fair – but again: If it just stayed frozen in the state it is now (and that assumption is about as unreasonable, as it being "perfect" in a year), it's already going to be tremendously useful for increasingly many people (when cost will go down, and they will, accessibility will go up) — at least until something better comes around.
For those who extract value right now, the simple alternative (just not using it) is never going to be the better choice again. It's transformative.
Yeah, but this is different because it's largely just money -> more GPUs -> more people looking at the problem -> better results. You can't stumble upon an 8K TV overnight but you can luck upon the right gold mining claim and you can luck upon some new training algorithm that changes the game immediately.
Do any AI companies actually turn a profit? I feel like the only real winner is Nvidia because they are selling shovels to the gold diggers, while all the gold diggers are trying to outspend each other without a business model that has any consideration for unit economics.
I love a prudent take on company money – but given how investing works and how young this entire thing is and the (to me) absolutely real value, I find it hard to be very worried about that part right now.
I can literally run a ballpark model on my MB Pro, right now, at marginal additional electrical cost. I will be the first to say that all of this (including GPT4) is still fairly garbage, but I don't know when there was the last time in the history of tech, where less fantasy to get from here to what will be good was required.
The thing is that the bigger business giants like MSFT or Amazon are probably profit quite nicely from AI. Smaller companies, not aligned with any big giants - probably not.
I’m really curious what is making you so much more productive. My experience with AI has largely been the opposite. Also curious how you’re using AI to make $3,000 per month more than you would without it.
I feel the same way. I think LLMs are neat, and I find them interesting from a technical standpoint, but I have yet to have them do anything for me that's more than just a novelty. Even things like Copilot, which I'll admit has impressed me a bit, doesn't feel like it would radically change my life, even if it was completely foolproof.
Then he is either making an extra $3,000/month or working 30 hours less per month. If the former isn’t true, I am extremely doubtful that the latter is.
Seems more likely that he is over estimating the value that LLMs are bringing him. Or he is an extreme outlier, which is why I was asking for further details
> largest/simplest productivity improvement in my life, so far
many productivity improvements in the last years: Internet Search, Internet Forums, Wikipedia, etc.
LLMs and other AI models is continuation of the improvement of information processing.
The bubble is that every $1 in capital going to OpenAI/Nvidia is a $1 that cannot be invested anywhere else: Healthcare, Automotive, Education, etc. Of course OAI and Nvidia will invest those funds, but in areas beneficial purely to them. Meta has lost $20bn trying to make Horizon Worlds a success, and appears to have abandoned it.
Even government-led industrialization efforts in socialist economies led to actual products, like the production of the Yagan automobile in Chile in the 1970s[0].
We've already had a decade plus of sovereign wealth funds sinking tens of billions into Uber and autonomous driving. We still don't have those types of cars on the road and it's questionable whether self driving will even generate the economic growth multiplier that its investment levels should merit.
I did a quick check and it seems like the entire Uber and clone industry is net negative. Uber, Lyft, Didi, Grab seem to have lost more money than was invested and once they stabilize they look like mediocre businesses at a global scale (Uber's been banned from many jurisdictions for predatory practices and in many other jurisdictions it seems to trend towards being as expensive as taxis or more once profitability becomes a target).
This sounds like the broken window fallacy. You could use the same logic to suggest that Meta dump piles of cash on the sidewalk in front of their office - it’d circulate but it wouldn’t help them.
If the same $20bn was spent on fixing a bridge, people would spend those wages to boost economic activity AND have a fixed bridge that will improve output even more. Horizon Worlds isn't a productive use of capital in that regard.
It'd be one thing if they open-sourced their VR tech, some of that could lead to productive tech down the line, but as a private company, they're not obliged to do any of that.
Last night I asked ChatGPT 4 to help me write a quick bash script to find and replace a set of 20 strings across some liquid files with a set of 20 other strings. The strings were hardcoded, it knew exactly what they were in no unclear terms. I just wanted it to whip up a script that would use ripgrep and sed to find and replace.
First, it gave me a bash script that looked pretty much exactly like what I wanted at first glance. I looked if over, verified it even used sed correctly for macOS like I told it, and then tried to run it. No dice:
replace.sh: line 5: designer_option_calendar.start_month.label: syntax error: invalid arithmetic operator (error token is ".start_month.label")
Not wanting to fix the 20 lines myself, I fed the error back to ChatGPT. It spun me some bullshit about the problem being the “declaration of [my] associative array, likely because bash tries to parse elements within the array that aren’t properly quoted or when it misinterprets special characters.”
It then spat out a “fixed” version of the script that was exactly the same, it just changed the name of the variable. Of course, that didn’t work so I switched tactics and asked it to write a python script to do what I wanted. The python script was more successful, but the first time it left off half of the strings I wanted it to replace, so I had to ask it to do it again and this time “please make sure you include all of the strings that we originally discussed.”
Another short AI example, this time featuring Mistral’s open source model on Ollama. I’d been interested in a script that uses AI to interpret natural language and turn it into timespans. Asking Mistral “if it’s currently 20:35, how much time remains until 08:00 tomorrow morning” had the model return its typical slew of nonsense and the answer of “13.xx hours”. This is obviously incorrect, though funnily enough when I plugged its answer into ChatGPT and asked it how it thought Mistral may have come to that answer, it understood that Mistral did not understand midnight on a 24 hour clock.
These are just some of my recent issues with AI in the past week. I don’t trust it for programming tasks especially — it gets F# (my main language) consistently wrong.
Don’t mistake me though, I do find it genuinely useful for plenty of tasks, but I don’t think the parent commenter is wrong calling it snake oil either. Big tech sells it as a miracle cure to everything, the magic robot that can solve all problems if you can just tell it what the problem is. In my experience, it has big pitfalls.
I have the same experience. Every time I try to have it code something that isn't completely trivial or all over the internet like quicksort, it always has bugs and invents calls to functions that don't exist. And yes, I'm using GPT-4, the best model available.
And I'm not even asking about an exotic language like F#, I'm asking it questions about C++ or Python.
People are out there claiming that GPT is doing all their coding for them. I just don't see how, unless they simply did not know how to program at all.
I feel like I'm either crazy, or all these people are lying.
> I feel like I'm either crazy, or all these people are lying.
With some careful prompting I've been able to get some decent code that is 95% usable out of the box. If that saves me time and changes my role there into code review versus dev + code review, that's a win.
If you just ask GPT4 to write a program and don't give it fairly specific guardrails I agree it spits out nearly junk.
> If you just ask GPT4 to write a program and don't give it fairly specific guardrails I agree it spits out nearly junk.
The thing is, if you do start drilling down and fixing all the issues, etc, is it a long term net time saver? I can't imagine we have research clarifying this question.
> People are out there claiming that GPT is doing all their coding for them. I just don't see how, unless they simply did not know how to program at all.
I doubt it and certainly not for anything beyond basic. I've seen (and tried GPT's for code input a lot) and often they come back with errors or weird implementations.
I made one request yesterday for a linear regression function (yes, because I was being lazy). So was chatGPT... It spat out a trashy broken function that wasn't even remotely close to working - more along the lines of pseudo code.
I complained saying "WTH, that doesn't even work" and it said "my apologies" and spits out a perfectly working accurate function! Go figure.
I hear you. It's all pretty bad. I have spent half-days getting accustomed to and dealing with gpt garbage — but then I have done that plenty of times in my life, with my own garbage and that of co-workers.
On the margins it's getting stuff good enough, often enough, quick enough. But it very much transformed my coding experience from slow deliberation to a rocket ride: Things will explode and often. Not loving that part, but there's a reason we still have rockets.
I've had the same experience, but I usually get what I want. Admittedly, I'm using it to script around ffmpeg which is a huge pain in the ass.
That said, every single script it churns out is unsafe for files with spaces on the first go round. Like.. Ok. It's like having a junior programmer with no common sense available.
How many GPUs are being delivered today and for how long will they be used / what's their life?
Who is funding the purchase of those GPUs?
If VC money then what happens if the startups don't make money?
Are users using AI-apps because they are free and dump them soon?
Isn't their competition in semiconductors? Won't we have chips-as-a-commodity soon? LLMs-as-a-commodity?
Is Big Tech spending all this money to create VALUE or just to survive the next phase of the technological revolution? (e.g. the AI rush)
If prices are high, and sales are high, and competition is still low -- then how much is nvidia actually worth? And if we don't now why is it selling for so many times earnings?
It's like the previous craze with blockchain. Everyone and their dog was about to use blockchain to do the most awesome thing ever. And even things that didn't really lend themselves to blockchain were suddenly presented as prime examples of how to use blockchain.
I'm not saying both technologies don't have their uses, but the hype around them is crazy and not healthy.
Like with blockchain, there are real uses, of course i totally agree!
But I was more thinking about the craze surrounding the whole thing, like with blockchain, you can see everyone trying to sell you AI for kinda everything.
This technology is already fully automating copywriters and almost replacing concept artists right now. However it is true that right now the valuations are for something far more than that, and so far it doesn't seem like LLMs will be able to do much more.
I was talking to someone that just retired from a programming position at a FANG and he seems to think that AGI (artificial general intelligence) is only a few years away just based off what he sees with ChatGPT and he's dumping all his money into AI stocks. The level of hype and over-extrapolation is so absurd, and the fact that it can affect someone with a technical background...
Isn't the role of a concept artist mainly to do worldbuilding and drawing second? AI does not seem to have a good world model, they make pretty pictures but they lack thought behind them.
Agreed, and I think there are a number of... over-enthusiastic executives with dollar signs in their eyes who are in for a rude awakening about this. It might sound great to replace your artistic staff with an LLM subscription, until you realize that you laid off all the creative vision with them. That isn't to say I think it'll go away though, I wouldn't be surprised if art students in the future are taught how to wrangle LLMs to supplement their own designs.
P/E ratios (so far) say otherwise: all the big tech public companies have reasonable P/E ratios and are investing heavily their profits in contrast to, say, the dotcom bubble when over 40% of the companies had P/E ratios unsustainably over the moon.
Are there a gazillion companies riding the "AI everywhere" wave to raise money? yes, yes there are. Will most of them fail? sure.
But the big players are fine at the moment so there is nothing that can burst very hard (yet) and the difference is in the denominators which are, so far, going up.
Of the top ones only NVIDIA and Amazon have P/E ratios a bit too high and among the top 10 only AMD's is way too high.
> This is the closest we've ever had to multiple large respected tech companies selling "snake oil" a cure all.
The problem with this take is you can deliver real results. At current $dayjob we do the very dumbest thing which is text -> labels -> feedback -> fine_tune -> text... and surface them as part of our search offering and it's rocketed to the most useful customer feature in less than 6 months of rolling it out. Customers define labels that are meaningful to them, we have a general-purpose AI classify text according to those labels. Our users gleefully (which is shocking given our industry) label text for us (which we just feed into fine_tuning) because of just how fast they can see the results.
Like it's as grug brain as it gets and we bumbled into a feature that's apparently more valuable to our users than the rest of the product combined. Folks want us to sell it as a separate module and we're just hoping they don't realize it's 3 LLMs in a trenchcoat.
This opinion is so crazy to me. How can it be a bubble if it's already providing me with 10x productivity gains? Things I've used GPT-4 to automate: replaying to all emails, writing entire books, doing all the legal paperwork for my new business, completing daily code challenges, generating shopping lists, writing birthday messages, and replying to my family group chat. And that's when the tech is still relatively new! Once the technology advances and I'm able to automate basically all of my daily tasks, who knows where I'll be.
> Noone thought self-driving cars would actually work.
Many people thought self driving WOULD work and that we'd be further along than we are now. We have vastly overestimated how far we'd be, and vastly underestimated how much time and effort it would actually take.
Self driving cars as they exist today are still mere toys compared to where the industry thought they were going to be. Look at Cruise, Waymo, Zoox, Uber's ex-self driving car division and others.
We are not anywhere near the self-driving autonomous cars we had hoped for.
> Noone thought beating the game of Go was feasible.
Oh, yes, we did, once we beat chess. It was just a matter of time.
> Noone thought self-driving cars would actually work.
And... they don't? Call me when I can buy a regular car where I can sleep while traveling 8 hours, driven by the car itself. We're probably "flying cars" away from that.
I think there is a bubble, but not in the traditional sense of this all being nonsense. Obviously, AI is going to be a world changer, but I think in a few years many companies are going find, at least for them, that their projects using AI are wildly unprofitable. Good AI is expensive. This could take years to play out, but when Nvidia is trading higher than Microsoft and Apple in total market cap, we'll be well into it. I don't rule that out.
I've stopped believing in rational markets a long time ago. For the people behind this funding madness:
> to skew markets, skew the financials of big tech and create a bubble in the space.
Are the intended consequences of this. The people behind the money in AI, just like the people behind the money in crypto, don't care if there's a reality to all of this, they just care if they can make a lot of money while the music is playing.
I really thought 2022 was going to be the beginning of tech returning to reality, but naively didn't understand that this would entail a lot of people with a lot of money losing money, and that's not going to happen.
As with all bubbles, the interesting thing isn't pointing out there's a bubble, we've been living in many bubbles for decades now. The interesting thing is pointing out what will make it pop. So long as globally money keeps pouring into US markets we'll see this continue.
On the plus side, at least LLMs are a lot of fun to work and play with!
"Efficient markets" deliver normal returns. The people with the most capital to throw at things are trying to create monopolies that can generate supernormal returns.
This is already a highly unstable arrangement, and it's made more dangerous by introducing impurities like artificially suppressed interest rates and SPAC IPOs.
With the way NVIDIA is moving and how it's seeding AI startups, you could be right about them actively trying to do this. But there is a reality beyond our ability to predict it..
Theres a lot of not-great software that can have bites taken out of it by LLM advancements and most of the incumbents are going to pay into it because they just can't compete in the AI field.
Can LLMs become reliable enough at transforming data that it can replace or augment the current slew of ETL tools? Can it produce visualizations better then current BI tools? Can copilot compete with a junior developer? Im not sure, but at this point Im willing to say 50/50 which is worth the bet.
The last All in Podcast had an interesting take on this:
Most of the R&D and Capex going into LLMs/GenAI is speculative. The investments haven’t translated into real revenue yet. The expectation is that there will be a large pot of revenue at the end of the road, but we haven’t seen the killer apps to substantiate this. This makes for a perfect bubble if the promise doesn’t pan out.
Relatedly Nvidia’s revenue - as impressive as the recent growth has been - is fully exposed to this risk.
Of course it’s possible (likely?) that there will be major wins from this tech, but the fact that there isn’t definitive proof (in the form of revenue) yet represents real risk.
> The expectation is that there will be a large pot of revenue at the end of the road, but we haven’t seen the killer apps to substantiate this
Imho, unlike crypto and NFT, AI is not a solution in search of a problem. While there is a lot of hand waving, it is not very adventurous to predict that there will be significant productivity gains by adding AI on top of current business processes. Thus the killer apps will be... the same apps we are using today with a touch of AI magic dust.
that's the root of my skepticism. I've yet to see huge transformational projects involving AI from the grapevine of my 750k+ employee consulting firm, all I hear about are experiments, webinars, and POCs. What I see in the news regarding the few projects live is not good, AirCanada and some others where the AI went wrong and now the company has to backtrack.
There's no doubt the tech community is all excited because genai, indeed, helps write code but I've yet to hear a large company like Coca Cola announce large AI transformation projects the way they announced large cloud transformation projects a few years ago.
I get more and more on the AI bandwagon as time goes on but I still have a pretty healthy skepticism on how deep and wide the tech will penetrate day to day business at enterprises and that's where the ROI is.
Not trying to be dismissive of Mistral, but I bet that's a large driving force behind the effort. Usually I'd prefer to focus on the technical aspects, but with the undeniable geopolitical impact of technology/AI, I think it necessitates a discussion.
Just from experience in the cloud industry, Microsoft is really successful within Europe, potentially more so than in America. I think this partnership will be really successful.
What I don’t get is that mistral is really only possible because of Meta.. and Meta has a user limit on the license. Does that not apply to Microsoft hosting this? Isn’t the economics quite weird here? I’d be pissed if I were meta that someone took my model that I spent millions on and now will host a fine tuned version for money facilitated by Microsoft.
quibble, but nvidia is the shovel maker. LLMs require a lot of vram but the underlying hardware requirements are actually fairly simple. There are already efforts in place to create inference specific ASICs.
That'll absolutely eat into Nvidia's profit margins.
I personally didn't realize how fast other models would catch up to OpenAI.
There is a whole set of models now (and some like Meta are purposely trying to undermine OpenAI competitive advantage via open source models) and they are relatively interchangeable with nearly no lock-in.
OpenAI's main advantage is being first to market and having the strongest model (GPT 4), and maybe they can continue to run ahead faster than everyone else, but pure technical leadership is hard to maintain, especially with so many competitors entering.
Their main advantage for now is their super clean API. Open source alternative are already on par with GPT-3.5 and 4 capabilities, they just don't have as good a package but that could change rather quickly too.
> Open source alternative are already on par with GPT-3.5 and 4 capabilities
I'm not sure if this is true. With GPT-4, I can successfully ask questions in Japanese and receive responses in (mostly natural) Japanese. I have also found GPT-4 capable of understanding the semantics of prompts with Japanese and English phrases interleaved.
Out of curiosity, I tried doing the same with local models like Mistral 7B and I could never get the model to emit anything other than English. Maybe it's a difference in training data, but even then, GPT-4 has an allegedly small set of training data for non-European languages.
Is that true? I was running Llamas on my laptop a few days ago, and it was giving measurably worse results than ChatGPT. I think it was the uncensored 13B model, but if you got something that's on par with ChatGPT that I can run on my own hardware I'm pretty interested.
13B models probably cannot directly compare with ChatGPT 4 which maybe +1T parameters or a 5 way MoE of 200B each - or something like that. So you can not likely run a model competitive with ChatGPT locally in the near term.
For now. As others have said, there is no technological “moat” in this business that could prevent others from catching up.
Perhaps the best way for Open AI is to become THE established AI services company. AWS is still the leader in cloud computing space, and only has Azure competing, despite the fact that other big companies are also technologically capable of building similar products.
> AWS is still the leader in cloud computing space, and only has Azure competing, despite the fact that other big companies are also technologically capable of building similar products.
What happened to GCP? I personally switched away because of the bad experiences.. but is that happening in scale as well. I see it barely mentioned these days.
I'm pretty bearish on GPT 5 being better than 4. With how neutered 4 has gotten over time, I'd be surprised if GPT 5 is able to perform better with all the same restrictions that GPT 4 has. GPT 4 is less and less willing to actually accomplish a task for you than it is to tell you how you can do it. It looks more and more like Markov chains every day.
Sure, but it is somewhat disheartening to see GPT 4 still being the king by a clear margin after a full year, especially since it's been nerfed continuously for speed and cost effectiveness.
Is gpt4 as good in non-English uses? It's not clear to me that it would be particularly important or advantageous, but does Mistral being based in Europe and polyglot first make it interesting vs. gpt4 in some dimension?
I guess it might depend on language, but as a Spanish speaker who sometimes uses LLMs in Spanish, I'd say the gap between GPT-4 and most of the competition (Mistral included) is actually larger in Spanish than in English.
It's the best multilingual model out there and it's not even close.
Especially in terms of open models Mistral's are the most multilingual but outside a few handpicked ones the level of proficiency is just too poor for any real usage.
In my experience it’s not such a simple question. If you want to be able to speak in nuanced non-English and have it pick up on the intricacies, or have it respond in rich correct non-English, then it’s not the best model (Cohere recently released an aya model that I would recommend checking out if this is your use case).
If you want to be able to give basic commands and have the model reason about the logic behind your commands, gpt 4 is still the best, even in minority languages.
I don't disagree with you, but an open source model fine tuned for your use case, and embedded with your data is probably going to be way better at many companies uses cases than GPT4 is.
I'd say you should compare the models for your use case. Which is better depends on how much you're willing to pay, what kind of problems you need help with, speed, ease of use.
With mixture of experts (MoE) models, I would think that Mistral might make a lot of money reselling specific models to big players like Microsoft and Google to augment their MoE systems.
I mostly use either mistral7b or mixtral8-7b for most things, and experiment with other models on the side. In what world will LLMs not become a commodity?
You might answer that question by saying that on player achieves AGI and captures the market, but I think there will be more to AGI than LLMs.
> With mixture of experts (MoE) models, I would think that Mistral might make a lot of money reselling specific models to big players like Microsoft and Google to augment their MoE systems
I am curious about what you mean by this. There is a (very understandable) common misconception about what "mixture of experts" means in current practice.
Mixtral is not a mixture of fully-separable domain experts, like one is an expert at programming and one is an expert at arts and literature. The "experts" are per-layer (and as far as I know, the subject of their "expertise" not even interpretable at this point).
I'm lost here. Can anyone explain why the speculation of risk resounding around a 'bubble' of LLM and other emerging technologies (TinyLM, VLM, ALM, etc) is rounding a bubble?
Meaning, the dot-com bubble showed a misappropriation of revenue by leading technology companies alongside the capital investorship, mainly VC's and private equity. Basically, everyone spent without producing a product; which lead to the mergers still the basis for many of the largest corporations today (and subsequently the 2008 crash, which further merged companies in the aftermath).
So, for me the difference is the money being spent here in this 'bubble' isn't like parking new mall shops on various websites, then spending the capital to entice buyers to purchase (either for store products or the store itself) - rather - it seems to me that both private and public sentiment is that the future is bright with the right application of these tools, in the right systems, to promote applicable use cases for things we (humans) can leave up to code/machines.
For instance, one of the most impressive applications of TinyLM uses is setting up several (or dozens) of tiny sensors in say a greenhouse. Each sensor can be linked to a central data repository for active monitoring to deploy active controls - be it barometric pressure, moisture content (soil, air), etc. Linking a bunch of these devices and letting it automatically run, along with linking this sensors as the data-providers to machines that may in turn plant, cull, trim, essentially a fully automated greenhouse.
I'm not greenthumb and I lack in-depth knowledge around how greenhouses (modern or old) work. However, it's without understanding the details I do understand it still takes a (or more) human to maintain the greenhouse.
Consider making these kinds of greenhouses (as is standard in some places now with vertical in-door growing farms) completely autonomous.
That's pretty technologically feasible it would seem with the newest applications of (quote) AI. And it's likely very profitable as well.
If I were to take the same ideology and apply to other industries, I find several applications like the one I mention that would absolutely change how we (humans) live in this (soon to be, IMHO) post-human industrialized world.
So where is all of the doom and gloom from - monopoly, errant comprehension of the technologies, or simply dogma?
Funny that they say "Its models are “open source”, meaning technical details will be released publicly.". In fact, they aren't only releasing the technical details, they are also releasing the model weights themselves.
This is just another play to give out more Azure credits to anyone that can feasibly consume them. Azure credits show up as unearned revenue on their SEC filings where they state that they "expect to recognize approximately 45% of this revenue over the next 12 months and the remainder thereafter".
It's wild that you can give out gift cards that make your company's value go up so much more than the gift cards could ever cost you. It's almost like one of those financial schemes that end badly.
> They lost big on the mobile OS. It is the biggest blunder of Microsoft.
I've watched Bill Gates tell Andrew Ross Sorkin during an interview "if it wasn't for the FTC investigating us for the anti-trust lawsuit, we wouldn't have been distracted/took our eye off the prize on mobile"
How sure are we Microsoft "has what it takes/had what it took" to deliver a phone + operating system as polished as Apple?
Windows 11 is very much measurably worse than Mac OS Sonoma. Littered with ads, in between old UI + new UI patterns, etc.
That's 2024
I'm not super confident I'd prefer a Microsoft/Windows mobile OS and therefore I'm not super confident they could actually have delivered a good one
I had multiple Microsoft Lumia as my primary driver for a couple years. I've been on iPhones for 10+ years total. There were some good (better user interface) and bad (Live logins required in random places) with Windows Mobile. Like Windows, they have pockets of good ideas fractured by their lack of cohesion.
I think Windows Mobile could have been better than Android. In fact, at one point it debatably was. Personally I think Windows is considerably more usable than OSX (yeah I know that’s going to offend people and raise a near-religious argument) and while I won’t say Microsoft’s hardware is or ever was as good as Apple’s, I think they are far better at UI/UX and hardware than Google. Actually I think they’re better at almost everything than Google.
They could have easily been where Android is now if they were two years faster. I think we’d just have a different duopoly and iPhone would still be basically where it is today.
> I won’t say Microsoft’s hardware is or ever was as good as Apple’s, I think they are far better at UI/UX and hardware than Google.
Microsoft-proper hardware is crap, even compared to high-end Google. However, that doesn't matter. My Thinkpad eats Apple hardware for breakfast (as well as anything from Google or Microsoft). Dell has a few nicer laptops in the Precision line which do almost as well as Thinkpad, and definitely better than Apple / Google / Microsoft.
It's the ecosystem.
> Actually I think they’re better at almost everything than Google.
I'd agree, except:
- Hardware
- Office 365 typing synchronization
Google Pixel Pro is quite good (as are most top-of-the-line Google phones), as was the Chromebook Pixel (their top-of-the-line Chromebook, when they made it).
Microsoft has gotten pretty good at software development. Satya Nadella has been doing a surprisingly good job reforming everything about the company.
That doesn't mean Windows wasn't absolute dogs--t for many years and isn't riddled with a pile of technical debt. Apple, in contrast, started with NeXTStep. I agree about the specifics you mention (built-in ads and spyware), but the core problems with Windows predate modern Microsoft, and can't really big fixed.
If, in 2024, Microsoft decided to invest in building a real mobile OS, I think they could do okay. The bigger problem would be lack of app ecosystem, and the chicken-and-egg problem with that and users. It's not clear even the best mobile OS could displace Android + iOS.
If I were Microsoft, and I wanted to get in, I'd probably fork Android, to maintain app compatibility. Doing that well would mean killing the goose which is currently laying the golden eggs, though, as it would require replacing Google apps with Microsoft ones. A lot of what makes Android work are free Google accounts, whereas Office 365 is $100 per year. I don't think a Microsoft phone would be competitive unless it had all the same stuff for free, and likely more.
I actually feel like Android is starting to be a little bit vulnerable; a privacy-preserving, non-user-hostile version could have pretty good uptake. Again, not the business Microsoft is in.
It doesn't matter if it's any good, it just has to be functional, which Microsoft proved they can do, since windows phones existed. Microsoft has the business side (aka money and patience that Google doesn't) to make it happen. Look at how well Microsoft Teams is doing. Look at Bing, now that it's got ChatGPT behind it.
I find this ironic because you could say many of the problems with modern Windows (& Linux) GUI stem from distancing the design from the traditional desktop paradigm and seeking to mobile-fy everything, specially with large click targets.
> Getting to AGI is like getting to first nuke for tech behemoths.
Except, unlike AGI, nukes happened. We are no closer to AGI than we were when we lived in caves. We found a small local maximum (LLMs), I have seen NO evidence that there is a path from here to AGI.
Even if LLMs are a local maxima, given that we're even asking that question implies that we are closer to AGI than caveman who didn't have GPUs, even if we don't know exactly how much closer. Unless it takes us 12,000 years to get to AGI, which, it could, since we don't have it yet, but assuming we get AGI before 12,000 years are up, which is when the end of the Paleolithic era was, then yes, we can say we are closer to AGI than cavemen, regardless of if LLM or GPUs are how it comes about. If it never does, then this comment will age poorly in 12,000 years, but I'm okay with that.
You don't think having functional computers being capable of trillions of operations per second is even the slightest bit of an improvement over use of literal stones?
I am genuinely curious: What do you imagine would be the kind of thing that would be a meaningful step towards AGI?
I have no idea when we'll get there for real, but it seems a pretty big assertion that nothing invented in the last 150 years even helped. So what do you think would help?
Ok. I’ll rephrase. In caves we were 0% there. Now we’re 0.00001% there :). Maybe. May need a few more zeroes. What we have helps probably, but hard to be sure how much. Maybe we need 1e40 FLOPS for AGI. In that case going from 1FLOP (human in a cave) to 1e14 FLOPS (a GPU now) is irrelevant.
I’ll believe we’re closer when we have a computer solve a novel problem that is not a simple pattern match to a similar solution. When a computer can reply to “write me a JIT translator from ARMv5 to ARMv7M” with working code. That takes actual thought and we’re not even close.
That covers 0.001% of the work - the part that my polit-sci-major no-programming experience sister just explained to me given the prompt. Now do the hard parts :)
The hard parts are just an implementation detail. Let's say a human took 15 minutes to draft all that up. It's about 800 words, and given a typing rate of 50 wpm, it would take that long to just type that in, so that seems reasonable. If that's only 0.001% of the work, it means it would take 3 years to finish the project.
3 years to write a JIT from ARMv5 to ARMv7M?
I think your numbers might be off :)
btw, your polit-sci-major no-programming experience sister seems pretty clever if she was able to understand all that. I ran it past a few cavemen I came across and they didn't quite understand the nuances surrounding runtime environment management.
The question wasn't if I could do a JIT compiler but if the LLM could. Which, there are limitations to the technology, but my opinion is it would go quite far at our current level of technology. Presumably you don't, bit you've shown no evidence that it can't. Whereas I've shown that it's able to decompose the problem and start dotproducting the problem.
I think the real nuke will be that we're already at some basic level of AGI because human consciousness isn't as complicated as we think it is. I'm guessing we're only 10 years out from lowly generative AI getting advanced enough to effectively be what think of AGI now.
This seems like a remarkable chain of reasoning. Out of curiosity, what would convince you that there is a path from LLMs (say) to AGI? What would you expect progress along that path to look like?
It’s not just that. Microsoft is in a better position today to make big money from AI because their products are in millions of corporations. B2B is their bread and butter. Google isn’t in that position, neither is Apple, Meta or most AI companies out there. Microsoft has built trust over decades of selling OSes and offering legendary backwards compatibility in anything they build. This is what the big spenders want.
Correct, but there’s typically a divide between workspace tools and infrastructure for ownership, compliance, and approval internally.
In a prior life, our IT manager was the owner of Microsoft productivity products and I was the owner of Azure. We both had drastically different risk profiles and governance needs.
And how will that materialize? Will Apple build a new AI chip that would be affordable enough to embed in every iPhone, or whatever new device they will invent? Or will it all be cloud based where it should be able to support hundreds of millions of users?
Microsoft already has the cloud infrastructure ready. They don't need to build a new device, or a new operating system, or whatever. They're milking the AI cow as we speak.
Apple has had "AI chips" in the iOS devices for years already. I wonder what you define as an AI chip, though, since we could be discussing different sorts of custom accelerators.
Microsoft's gunning for a far more short-term win: Search.
Bing has been an also-ran in the search world for a decade (single digit % of search volume compared to Google's 90%+), and AI has shown the first real crack in Google's monopoly.
The future is pretty clearly going to be directly getting AI answers to questions and much less looking at a page of 10 blue links and it's still really unclear if Google is going to make the transition.
Actually hilarious that you think ads are going away. Microsoft has already started pitching ways to embed advertising and sponsored links in AI responses [1]. They've already built prototypes [2]. It will probably be worse than in search because AI ads will blend more seamlessly into a written or spoken paragraph.
I believe Google stands apart; although their current offerings sucks, they possess all the necessary components, including software, data, and hardware.
And yet, it's kinda looking like everyone except OAI has hit the same pre-gpt4 capability wall.
I'll hedge my bets on whether AGI is even possible entirely on how much of an improvement we'll see with GPT5. If it is just a marginal improvement, that's basically it for the current ML bubble.
that's an interesting take, i wonder what the graph looks like of model capability over time. If GPT5 doesn't fit in the curve and a year goes by with no one else showing the same rate of progress... is that it for the LLM hype machine?
Edit: i was thinking while walking my dogs, if GPT5 was another great advancement and especially if AGI was around the corner then why would Karpathy leave OpenAI?
>Microsoft knows they won big in Cloud via Azure Big as in 100s of billions.
Malarkey. Microsoft got where they are in cloud by dogfooding every single piece of software they wrote into their cloud platform, every single service they ran on their gaming platform, and every single user on github and minecraft. After that they turned to corporate customers and forced them into the cloud as well and finally made every single end user of windows sign up too (cant miss any of those precious KPIs.) If they won anything it wasnt through genuine consumer desire to use Azure. The efforts were mostly a shuffling of deck chairs and pump-job similar to the IIS wars on netcraft back in the day where M$ would pump their IIS numbers with static content served from parked websites at hosting providers they paid to switch from Apache.
>They have so much money lying around. Better to invest where they think the puck will be.
they also have a track record of building things no one wants and ruining things everyone likes. Minecraft and Github are demonstrably worse in a lot of ways than they were before redmond took the helm. having a lot of money doesnt make you clairvoyant.
>Meta is all in, Google is all in, Microsoft is all in.
who the fuck cares? these are all companies that exist in the nadier of their innovation. that they collectively scrape barrel to come up with tech memes for the business kids isnt exactly remarkable outside the fact they havent delivered anything of value in so long its surprising we still have to hear about them at all.
>Tesla is betting they’ll be the biggest robotics company on the planet.
The guys who cant get self-drive right? run by the same guy who pedaled twitter into the ground? sure.
AI is a meme for these companies...a parlour trick they use on the money pump for just another year longer before its reinvented into some other nonsensical sci fi pablum for general consumption under the late hour of capitalism.
Confrontational replies to comments that doesn't provide actual data or concrete examples to support the counterarguments are not very useful. It would be interesting to know why do you find in GitHub worse than before, or why Minecraft is ruined.
Not only does the official Mistral post have a whole section on the deal, it links out further to MS's official post on the deal also. It's all the same discussion
