You would be surprised - we're the 4th largest independent distributor of LLMs in the world - and nearly every Fortune 500 company has utilized either our RL fine-tuning package or used our quants and models - we for example collab directly with large labs to release models with bug fixes.
Well, who would have possibly thought that going from 'AI for the benefit of humanity' to becoming a software vendor for the Department of War is the ultimate rug-pull?
Meanwhile, the actual enterprise market, i.e., the adults in the room, already left for Anthropic. Why? Because Anthropic doesn't treat their core model like a weekend side-quest while they're busy chasing hardware fantasies and search engine clones.
OpenAI’s moat is evaporating in real-time, and it’s well-deserved. You can’t build a 'padded room' for the military and expect the tech world to keep buying the safety-first copium. They fumbled the trust, and now they’re fumbling the market.
The paper's critique of the 'data wall' and language-centrism is spot on. We’ve been treating AI training like an assembly line where the machine is passive, and then we wonder why it fails in non-stationary environments. It’s the ultimate 'padded room' architecture: the model is isolated from reality and relies on human-curated data to even function.
The proposed System M (Meta-control) is a nice theoretical fix, but the implementation is where the wheels usually come off. Integrating observation (A) and action (B) sounds great until the agent starts hallucinating its own feedback loops. Unless we can move away from this 'outsourced learning' where humans have to fix every domain mismatch, we're just building increasingly expensive parrots. I’m skeptical if 'bilevel optimization' is enough to bridge that gap or if we’re just adding another layer of complexity to a fundamentally limited transformer architecture.
Pdf24 has been supporting offline usage since forever and works like a charm. What you state in your post in simply wrong and misleading. I guess the "vibe" got too heated...
Crazy how OAI is way behind now and the only one to blame is Sam, his ego and lust for influence. Their downwards trajectory of paying accounts since "the move" (DoW deal) is an open secret. If you had placed a new CEO at OAI six months ago and told him to destroy the company, it would have been hard for that CEO to do a better job at that than Sam did. Should have left when he was let go but decided to go full Greg and MAGA instead. Here we are. Go Dario
I heard him say that too. And he's probably right. But it's more like every knitter now has access to an automated loom.
Oddly I feel AI is getting me off the endless learn new tech churn. I was looking at a few odd ball programming books on my shelf, graphics programming from scratch and retro game dev (c64 edition and nes editions) and thinking I might now have time to work through these instead of learning technology x.
you make a good point. I lost interest around "MCP" in all this; now we're up to people not understanding map reduce and manually garbage collecting for the AI.
knitting machines don’t generate the design from a prompt, and neither does industrial knitwear production facilities. In fact, knitting machines have quite a lot of manual input that goes into the final product, including careful programming.
Not equally true at all. Far from it. If you have ever seen people use knitting machine you would know the amount of skill required to operate one is far beyond creating a prompt. Same is true of looms, etc.
In fact this whole analogy makes no sense, a knitting machine is far closer to a compiler in this analogy then it is to a language model. Many would argue that automatic looms were the first compilers of the industrial age, and I would agree with that argument.
I was never talking about a knitting machine in the first place. Rather, I was referring to the old lady sitting on her sofa, knitting a sock she could also buy for a dollar, but decides to do it herself for the love of the game and nostalgia: a hobby.
The "art" of programming is going exactly that route, maybe with a little fewer ladies and more men.
I didn’t hear the exact analogy so I made some assumption. But I fail to see any insightful analogy which could make such predictions, unless the analogy is operating on top of some flawed assumptions about industrial knitware production.
An old lady could equally sit in front of her desktop PC write some HTML, and upload a blog page with her amazing knitting projects, or she could get pintrest. This was true before LLMs, and it is still true today.
Another potential flaw is the assumption that professional knitwear design does not exist. It does. Plenty of people work in industrial scale knitwear products. We have people designing new products, making patterns and recipes, we have manual labor in the production, operating machines or even knitting by hand. Case in point, travel anywhere and go to a local market popular with tourists, and you will see plenty of mass produced knitted products, most of them took great skill to design and produce. Nothing compatible to prompting an LLM to do this for you.
Not for long, presumably. Apparently the majority of marketable skills will come from a handful of capex heavy, trillion dollar corporations and you will like it.
The OpenClaw inventor? Ok, sure. I think this is sort of cute. The idea that it is just great that all knowledge work would just be a "hobby" when that logically a world in which there would be no leisure would be quite amusing if it is wasn't so depressing.
Agreed. The idea is nice and honorable. At the same time, if AI has been proving one thing, it's that quality usually reigns over control and trust (except for some sensitive sectors and applications). Of course it's less capital-intense, so makes sense for a comparably little EU startup to focus on that niche. Likely won't spin the top line needle much, though, for the reasons stated.
Most Copilot customers use Copilot because Microsoft has been able to pinky promise some level of control for their sensitive data. That's why many don't get to use Claude or Codex or Mistral directly at work and instead are forced through their lobotomised Copilot flavours.
Remember, as of yet, companies haven't been able to actually measure the value of LLMs ... so it's all in the hands of Legal to choose which models you can use based on marketing and big words.
EU could help them very much if they would start enforcing the Laws, so that no US Company can process European data, due to the Americans not willing to budge on Cloud Act.
That would also help to reduce our dependency on American Hyperscalers, which is much needed given how untrustworthy the US is right now. (And also hostile towards Europe as their new security strategy lays out)
How about not making these unenforced laws in the first place so that European companies could actually have a chance at competing? We're going to suffer the externalities of AI either way, but at least there would be a chance that a European company could be relevant.
The AI Act absolutely befuddled me. How could you release relatively strict regulation for a technology that isn't really being used yet and is in the early stages of development? How did they not foresee this kneecapping AI investment and development in Europe? If I were a tinfoil hat wearer I'd probably say that this was intentional sabotage, because this was such an obvious consequence.
Mistral is great, but they haven't kept up with Qwen (at least with Mistral Small 4). Leanstral seems interesting, so we'll have to see how it does.
Because the AI act was mostly written to address issues with ML products and services. It was mostly done before ChatGPT happened, so all the foundation model stuff got shoehorned in.
Speaking as someone who's been doing stats and ML for a while now, the AI act is pretty good. The compliance burden falls mostly on the companies big enough to handle it.
>Because the AI act was mostly written to address issues with ML products and services. It was mostly done before ChatGPT happened, so all the foundation model stuff got shoehorned in.
It's not an excuse. Anybody with half a working brain should've been able to tell that this was going to happen. You can't regulate a field in its infancy and expect it to ever function.
>The compliance burden falls mostly on the companies big enough to handle it.
You mean it falls on anyone that tries to compete with a model. There's a random 10^25 FLOPS compute rule in there. The B300 does 2500-3750 TFLOPS at fp16. 200 of these can hit that compute number in 6 months, which means that in a few years time pretty much every model is going to hit that.
And if somebody figures out fp8 training then it would only take 10 of these GPUs to hit it in 6 months.
The copyright rule and having to disclose what was trained on also means that it will be impossible to have enough training data for an EU model. And this even applies to people that make the model free and open weights.
I don't see how it is possible for any European AI model to compete. Even if these restrictions were lifted it would still push away investors because of the increased risk of stupid regulation.
> It's not an excuse. Anybody with half a working brain should've been able to tell that this was going to happen. You can't regulate a field in its infancy and expect it to ever function.
As I said, the core of the AI act was written about supervised ML, not generative ML, as generative ML wasn't as big a deal pre Chat GPT.
> You mean it falls on anyone that tries to compete with a model. There's a random 10^25 FLOPS compute rule in there. The B300 does 2500-3750 TFLOPS at fp16. 200 of these can hit that compute number in 6 months, which means that in a few years time pretty much every model is going to hit that.
As I also said, the foundation model stuff (including this flops thing) is incredibly stupid. I agree with you on this, but my point is that the core of the AI act was supposed to cover the ML systems built since approx 2010.
> The copyright rule and having to disclose what was trained on also means that it will be impossible to have enough training data for an EU model. And this even applies to people that make the model free and open weights.
Again, you're talking about generative stuff (makes sense given the absurdly misleading name now) whereas I'm talking about the original AI act, which I read well before ChatGPT happened.
The training data thing is a tradeoff, like copyright is far too invasive (IMO) and it's good to be able to use this information for other purposes. However, I personally would be super worried about an ML team that couldn't tell me what data went into their model. Like, the data is core to all ML/AI approaches so that lack of understanding would make me very sceptical of any performance claims.
Lets be real, the AI companies don't want to say what's in their models because of the rampant copyright infringement, not because of any technical incapability.
Ha, keep putting your prompts and workflows into cloud models. They are not okay with being a platform, they intend to cannibalize all businesses. Quality doesn't always reign over control and trust. Your data and original ideas are your edge and moat.
The same old speech that has been used throughout history. When cars were invented people complained to everyone that Ford intended to cannbolize all horse drawn carriages. When manufacturing was invented it cannibalized the work of all the sewing and knitting companies that had women making one item at a time. When Google was invented it cannabolized libraries, and encyclopedias, etc. etc.
Yet nobody wants a horse drawn carriage, nor to knit their own sweaters, nor go to the library to look things up in a physical encyclopedia.
Literate programming sounds great in a blog post, but it falls apart the moment an agent starts hallucinating between the prose and the actual implementation. We’re already struggling with docstrings getting out of sync; adding a layer of philosophical "intent" just gives the agent more room to confidently output garbage. If you need a wall of text to make an agent understand your repo, your abstractions are probably just bad. It feels like we're trying to fix a lack of structural clarity with more tokens.
This feels like a desperate attempt to stay relevant in a post-LLM world. They’re basically wrapping an LLM in a "professional" skin and calling it an expert review. The problem is that once you start letting an AI "expert" dictate tone and logic, you effectively lobotomize the writer’s original intent. We’re reaching a point where AI is just reviewing other AI-generated text, creating a feedback loop of pure mediocrity. Copium for middle management, if you ask me.
Grammarly even from the start was very distracting to me even as a someone using english as a second language to communicate. I have developed my own taste and way of articulating thoughts, but grammarly (and LLMs today) forced me to remove that layer of personality from my texts which I didn't wanted to let go. Sure I sounded less professional, but that was the image I wanted to project anyways.
Unrelated but surprising to me that I've found built-in grammar checking within JetBrains IDEs far more useful at catching grammar mistakes while not forcing me to rewrite entire sentences.
JetBrains’s default grammar checking plugin[1] is actually built on languagetool[2], a pretty decent grammar checker that also happens to be partly open source and self-hostable[3]. Sadly, they have lately shoved in a few (thankfully optional) crappy LLM-based features (that don’t even work well in the first place) and coated their landing page in endless AI keywords, but their core engine is still more traditional and open-source, and hasn’t really seemed to change in years. You can just run it on your own device and point their browser and editor extensions to it.
> The problem is that once you start letting an AI "expert" dictate tone and logic, you effectively lobotomize the writer’s original intent
Isn't that what grammarly has always been, since long before the invention of the transformer? They give you a long list of suggestions, and unless you write a corporate press release half of them are best ignored. The skill is in choosing which half to ignore
I disagree. You write when you have something to say. A service like Grammarly tries to help you convey what you want to say, but better. What you want to say is still up to you.
Words paint the picture, but the meaning of the picture is what matters.
Children and young students, certainly. Adult students: almost 100%. If writing is your job, then by definition, and your problem is more often finding something to say, not writing it.
You’re not counting all the office workers who have to write reports or emails, or all the scammers who write those websites to manipulate SEO or show you ads.
Everyone should think twice about putting their name on AI garbage, or garbage of any kind. But wishing doesn’t stop it from happening, especially when companies are explicitly selling you on doing just that. Remember the Apple Intelligence office ads?
It's great. Now that fancy writing is cheap and infinite, fields whose entire scholarship value was in obscurantist jargon bending have to actually start to turn on their brains and care about making more sense than an LLM can.
That's a beautiful Kafkatrap you've constructed. Not much of an argument though. Maybe there's another explanation for this though. Perhaps you think you know much more about different fields than you actually do?
Maybe not. But academia is going to change. Status will still have to be allocated by some mechanism but the classic journals and reviews based system will crumble under the weight of LLMs. Of course this will upset a great many of people who enjoy the current state of things.
Sam really fumbled the top position in a matter of months, and spectacularly so. Wow. It appears that people are much more excited by Anthropic and Google releases, and there are good reasons for that which were absolutely avoidable.
Also, never saw any Unsloth related software in production to this day. Feels strongly like a non-essential tool for hobby LLM wizards.