pu_pe

The company sounds like a bunch of hot air to me. From their about page:

> At the heart of Multiverse's platform is CompactifAI, a compression technology that applies tensor networks, a mathematical framework from quantum physics, to the problem of AI model compression. This application was pioneered by co-founder and Chief Scientific Officer Dr. Román Orús and reduces the size of large language models by up to 80-95% with immaterial accuracy loss.

Ironic, considering they are releasing a 438B model that loses to a 27B one. From another part:

> Singularity Machine Learning is a cloud service that uses quantum machine learning for solving supervised learning problems.

I wouldn't be surprised if these guys just finetuned an open Chinese model and called it a day.

Almost everything this administration does is inflationary. You cannot have tariffs, oil crisis, tax cuts, high debt spending without long-term investors getting worried that they won't meaningfully get their money back in 10 or 20 years from now.

I wish this would be true, but simply sampling unusual ideas is something eminently automatable. We already have temperature and other settings to guide a LLM towards this kind of "weird". So I don't think it's just about thinking outside the box as most weird ideas are probably also stupid, you can probably almost exhaust the search space of interesting ideas quite easily.

Maybe we will need humans to experience entirely new things instead. No one knows the taste of a new, bioengineered animal. No one knows what it's like to stimulate specific cell types in their brain. And so on. There are things that are and will remain uniquely human, forever.

Benchmarks got a little bump from this: https://xcancel.com/deepseek_ai/status/2087864585504305397?s...

The paper underlying this blog post is fundamentally flawed because of benchmark ceilings. If we define only simple tasks like asking what is the capital of France, all models will converge to 100%, obviously. But as bigger models get more capable we want them to replace more and more complex tasks, in as short time as possible.

Then of course there is the economics of it. Do people prefer to spend $5000 upfront to get things done 5x slower, or would they rather pay $20 a month for that?

Which percentage of people have GPUs capable of running Qwen3.8 27B? I am one of those, and for my job I am still resorting to hyperscalers because tasks are completed faster and more accurately that way. Even if we assume that models will no longer improve and we reach a point where everyone can run Fable in their laptop, surely running 1000x Fable agents would give you an advantage.

I think access to compute will matter just as much, if not more, as access to models.

> Overall my view is that AI is structurally a technology that tends to concentrate power, for reasons that have nothing to do with regulation (more to do with the extreme implications of the scaling laws). Open-weights do help some with this but are nowhere near a sufficient solution because they simply shift the concentration somewhat to those with the most compute and chips (which are roughly the frontier labs plus maybe hardware providers).

This is actually a good point. Open models are getting better and better, some of them might even be useful in consumer hardware now. But if AI performance is still correlated with compute power, then no doubt power will remain with the people owning the chips.

Seems to be SOTA for its size. Hopefully independent benchmarks will come soon.

Nice trick. Couldn't you embed the query though, compare it to the embedding of the categories, then ship only categories that are close to it in the prompt to a smaller model?

This now places deepseek flash v4 from DeepSeek themselves at higher prices than openrouter (depending on caching). Will be interesting to see if third party prices remain the same.

> What happens to the stock price of a company that missed its revenue targets by a factor of 5?

It depends. Tesla is a good example of how those things are not as clear cut as you might think.

The world is starving for more AI compute. Nvidia's GPUs are selling for 2x or 3x their former price, RAM prices increased even more, all because demand is huge right now. If anything buying those chips at bulk discount and then selling them right back to consumers would already be profitable.

Be careful that you are not basing your argument on whether you personally like lemonade.

If a kid's lemonade stand increased its revenue from $1B/year to $100B/year in two years, we would be paying serious attention to their product. People seem to love that lemonade! At this level of increased demand, maybe it's ok to let this kid borrow $1T over 5 years, because even if his growth slows down to 2x/year instead of 10x/year, he will still be covering that investment. That's hopeful and risky, but not at all insane in my view.

That's only true if the value of keeping your data and code private is zero. And in that case, Anthropic and OpenAI subscription plans may be even cheaper per day.

Yeah but you can probably have everything ready and then accelerate as necessary. Meta itself did this when releasing Llama 4, it was a really botched release right when they were feeling the heat from DeepSeek and others.

Based on the benchmarks, it seems that Muse Glimmer barely edges out against Qwen3.6 27B, except for tool-calling skills (MCP, etc.). I wouldn't be surprised if they released it now because they are afraid they wouldn't beat Qwen3.8 27B.

This will have general inflationary consequences for consumer products (phones, consoles, laptops, etc.). On top of current uncertainties regarding oil and fertilizers, I think 2% inflation in the US and Europe would be a very optimistic target.

Jevons Paradox is called a paradox because it is non-intuitive. It is very intuitive to conclude that when costs go up, people will use less of that thing.

I meant that the current business model of the hyperscalers is highly profitable, and that model is selling compute that you could build yourself but you'd rather not to. If OpenAI or Anthropic default because people prefer open source models, they will still have compute for rent. The only scenario in which they are truly screwed is if demand for AI goes down.

Profit margin determines whether OpenAI and Anthropic survive. The hyperscalers would be fine with a model where they make huge investments in infrastructure and you pay to access them per use/per hour. That is already their model for cloud compute and it is highly profitable, even though the vast majority of what people do with it (storage, VMs, etc.) is a commodity.

> The US has slid below Hungary in democracy, Hungary was also straight sliding towards dictatorship. Hungary did a 180 though by electing Peter Maygar. Hopefully the US can do a 180 in the midterms.

Considering TFA is from 2014 and knowing the Democratic Party, I think instead of a 180 you could at best expect a 20

When the Wright Brothers first came to Europe in 1908, their prototype could fly for several kilometers of controlled flight. The best European prototypes at that time were absolutely inferior, which is why the Wright Brothers immediately managed to sign contracts with European companies. They introduced the three-axis control system which was a major innovation, basically the opposite of a patent troll.

Anthropic is perceived as the leading AI company in the world. Even if they started selling inference alone, people would prefer to pay them a premium for it rather than figuring out how to operate opencode or other replacements. Add to that government contracts, enterprise and other markets, and I think Anthropic would be totally fine even if models plateau.

Transistor count has increased exponentially for decades and so did demand for compute. I think we will see a similar phenomenon with AI.

> And the fundamentals here are OpenAI and Anthropic, which are massively valued companies. They have humongous commitments and are generating real revenue on the order of twenty billion a year.

I think the size of their commitments is predicated on demand. Anthropic's annualized revenue run rate is now close to $50 billion, a fivefold increase from a year before [1]. They are making big investments, like $200 billion on Google's TPUs over the next five years [2], but those numbers seem justified by their expected revenue this year alone. If Anthropic cannot capture that revenue, someone else will.

Stock market valuations are a different beast, I personally think we have been due for a correction for ages now. But criticism of AI investment and particularly betting that it will all come crashing soon appears misguided to me. I can see a future where AI expenditures shifts around, not a future where everyone simply stops spending in AI all of a sudden.

[1] https://www.marketscale.com/industries/software-and-technolo...

[2] https://www.resultsense.com/news/2026-05-06-anthropic-200bn-...

It's the end of many eras, all at once. I think the same dilemma will appear for games, music and film too. As our ability to detect AI content deteriorates most authors will resort to it, at varying degrees.

Because there are many legitimate uses for nucleotide synthesis, and we would have to slow scientists down by adding more regulation. The other problem is that it's not like it's very difficult to do it, so a motivated person could still do it even without access to these services.

I wouldn't bet on token consumption going down any time soon. In the past year or so, the increase has been exponential [1].

[1] https://en.macromicro.me/charts/148532/world-openrouter-toke...

> Exponential View, a consultancy, counts $175bn of generative-AI revenue, on an annualised basis, in June. In a recent paper Anton Korinek of Anthropic and Patrick McKelvey of the Bank of Canada estimate total “AI services” revenue. Adapting their methodology, we reckon this was $220bn (again annualised) in the first quarter of this year. Ramp’s data imply that 2-3% of business spending now goes on AI, pointing to $170bn a year.

So between $170bn-220bn in annualized AI revenue today. Maybe it doesn't cover trillion-dollar bets but this is a very substantial number.

From what I understand, the rare books in question are not some historically relevant medieval manuscripts, but rather some relatively recent books (still under copyright) for which there are few print copies available for purchase.

I am not sure that physically destructing one copy of this type of book to preserve its contents digitally is so bad. Pretty much anything that is still under copyright should be valuable only for its content, not for the physical medium it's printed on.

I am fascinated by the discourse around this Bun rewrite. I read a lot of drama and personal accusations, there are pieces like this one trying to extract clues, and it seems everyone has a deeper ideological concern behind whatever they are trying to say. For this article, it seems to be skepticism towards AI and how successful it can be at replacing programmers. Other takes, like the one from the Zig maintainer, were also along those lines but more about the open source ethics and the future of that in a LLM world.

I am mostly bullish on AI capabilities, so from my perspective I don't see why we should be skeptical that frontier LLMs guided by experienced devs can translate whole libraries like that. Good follow-up questions would be how expensive it currently is to do so, and whether we will see people branching into all-in on AI versus no-AI camps as happened in this case.

A lot of people here are focusing on ADHD, but I think the author is commenting on a different phenomenon, because they were able to concentrate before and now it's just much harder. I feel it and many people I talk to feel it as well.

It's difficult for me to separate whether this is a result of other changes in my life over time or simply a reflex of ageing in general, though I do think it's a wider societal shift. In the same way that older people find it difficult to watch newer movies (and vice-versa), there is a certain pace associated with culture as well. We have, on average, shifted from books to TikTok in the span of a generation.