anon373839

> We've been seeing that headline for a few weeks now and I really don't understand the problem.

It’s powerful symbolism. It reminds me of that tone-deaf iPad ad that sparked outrage in 2024. The one where all the cultural artifacts were crushed in an industrial press to make a soulless slab of glass. And that was Apple, who is generally well-liked by the public.

But it’s the most useless canary ever. We already know that certain topics are taboo in China.

As for all the other uses the models have, it seems pretty clear they’re not doing anything weird. If they were, people would be posting examples of that and not of Tiananmen Square.

This, 100%. I don’t think the industry knows how to scale LLMs’ general intelligence much further. The training paradigm is about maximizing very specific behaviors / very specific tasks, but doing lots and lots of them. Which can create the illusion of general intelligence if your tasks are similar to the ones the models were fitted for.

… while concentrating immensely more wealth and power in the hands of a few weird and depraved people.

> how do you differentiate a non-addictive platform vs an addictive platform?

Legal systems have dealt with these sorts of slippery definitional issues for centuries! In this specific case, I would venture that the applicable legal standard would be some form of a balancing-the-factors test, where there is a list of criteria that a court weighs to determine if a given situation meets a legal threshold. It’s purposefully designed to have flexibility and ambiguity, with the expectation that, over time, courts will develop precedential law defining how the factors apply in various contexts.

For something like “addictive social media”, imagine a standard with factors such as:

- Whether an algorithm determines what material is presented to the user, based on a statistical model of user behavior;

- Whether the product was developed with the input of behavioral psychologists;

- Whether a significant portion of the users exhibit compulsive use of the product;

- Any mechanisms implemented l by the developer to deter compulsive use;

and so on.

Police can and should be reined in. There are so many better levers that can be pulled to do this, if only the political will were present. A technofeudalist surveillance state is a poor solution.

> How does this have value? Is any and every sentence in an e-mail considered 'fact' and thus to be fed into the AI?

I have a hunch what this is for. AI companies want to make bigger inroads into nontechnical work settings. But LLM progress outside of fields where verifiable rewards for RL post-training can be synthetically generated (coding, math) has been pretty flat. Buying years of operational data from a company like an airline could be used to reconstruct long-horizon task trajectories in areas like customer service or marketing.

The gains from increased parameter scaling are sublinear: there's no more hockey-stick improvement to be seen going in that direction. That doesn't mean some improvement isn't possible - it's just going to be increasingly not worth doing.

Also, I think the fact that small open-weights models are catching up to the frontier rather than the frontier rapidly pulling away is evidence of this. In fact, by far the most dramatic capability increase story over the past two years has been the gains made in the small-parameter regime.

One might think, "hey, this agentic coding thing was a pretty big deal!", but I think it's a bit of a distraction because models only recently became optimized for this specific use case. It's not like they suddenly gained so much general intelligence that they magically had the ability to use a coding harness. No, the labs started spinning up a bunch of RL environments and generating rewards over long-horizon trajectories of combining these tools. It's an excellent application of LLMs but care needs to be taken interpreting how much "progress" has been mae in terms of raw generalized capability.

> The top model from 2025 looks silly compared to the top model of the first half of 2026. Do you feel like progress has stalled?

I do. Pre-training is where the industry saw the “emergent properties” of LLMs arise and, for a time, people thought you could just keep scaling up bigger and bigger models but then the incremental gains from doing this did plateau. Labs will still do bigger models (Bytedance has a 10T planned), but these are sparse architectures and they aren’t going to have mind-blowingly greater intelligence. Fable didn’t either.

Scaling pre-training tokens also flattened out. What is still delivering gains is scaling RL on verifiable tasks. But that’s not general intelligence - it’s fitting models to specific tasks, which ML has always been good at. More importantly, most tasks to which humans apply their intelligence don't have computationally verifiable answers.

The models are post-trained on these prompt additions so they’re more structural than thinking of them as “system prompts” suggests. (All LLMs ever see is tokens going in, so even the concept of a system prompt is just formatting they’ve seen in post-training.)

You can also apply fixed token budgets for the reasoning blocks, though it will decrease quality in some cases.

> Qwen 3.6 35B A3B on medium thinking mode

Qwen 3.6 doesn’t have configurable reasoning effort, does it?

Agree. We’re going to hear a lot of buzz about recursive self-improvement in the near future, which I’d cynically say is meant to address the naked emperor you just called out.

Can labs get a boost augmenting more of their processes with automation? Sure. Will it result in a self-sustaining takeoff to infinity? … No. It will saturate too. Just my opinion.

This comment seems needlessly insulting and also kind of obtuse?

I don’t think the GP meant that it’s impossible to dispose of a large amount of money. But it is effectively impossible to spend down a certain level of accumulated wealth by purchasing goods and services that any human being or their family could require to maintain even lavish standards of living.

The examples you gave are conversions of wealth to power, using money to reshape the world. Which, I think, is kind of the crux of the issue. Perhaps you were alluding to this at the end of your comment.

> I’ve been in the “overweight” BMI range with low body fat before. It’s not too hard to get there

Did you do that as a natural lifter? It’s hard for me to imagine most guys being able to get into the “overweight” range at < 15% body fat without some assistance.

> A small model won't know the tax code of every city in the world, as it's probably impossible to fit, and it's the kind of thing that unless you already know about, it's very hard to search for

What makes it hard to search for? If you tell me you want to open a lemonade stand in East Bumbling Falls, I won't know that city’s municipal ordinances. But I will know that cities tend to have ordinances. So I’d search broadly for what codes that city has. Then I’d go through each one looking for anything that might affect a lemonade stand.

> You just can't compress the entire human knowledge into a 30GB file

Fortunately, that isn’t necessary! What LLMs need is a level of fluency with key concepts so that they can (1) make effective use of retrieval tools and (2) understand the material in the context window. 30GB-sized models can absolutely store enough knowledge to do this.

Here is an example from the field of law. Most lawyers who have litigated contract disputes in California know about Civil Code § 1717, which makes any contract providing for attorney fees to a prevailing party mutual, so that even if the contract was written to be one-sided, it won’t be enforced that way. It’s a simple enough concept, but there are many more particulars to it, such as what happens when the fee provision is only written to apply to part of the contract. (Answer: it depends on other facts.)

When a lawyer recognizes that they’re in a situation where § 1717 is relevant, the first thing they will do is pull the statute and read it, because nobody has it memorized. And they don’t need to.

In my experience, the facts that are compressed away in small models are ones you don’t need them to memorize. They need familiarity with the essential concepts in a field, so that they will have better comprehension of material put into the context (and make better retrieval decisions). They don’t need to know very particular details which are a hallucination risk and should be derived from the context instead.

Holy Streisand effect, that’s some background!

I don’t think this works, for a few reasons. First, intelligence gains from scaling the models bigger is sublinear now. So they could eke out a little extra performance, but the increased cost will eventually eclipse the economic value gained from this.

Second, humongous models are impractical even for them to deploy widely. They’re best used as teachers for smaller, more efficient models that can crank out the volume they need to sell.

Finally, there is a data wall. Sure, they can keep scaling RL on math problems and code. But with everything else, where will the supervision come from when they need several orders of magnitude more?

Yes, RSI seems to be the new AI industry McGuffin of 2026, just as agentic capability has become table stakes and scaremongering has become a punchline.

Yeah, Dax from OpenCode said that it appears to just be traffic shaping, nothing to do with the inference economics. He also said that OC have already replicated the inference cost in internal experiments.

I feel like the industry has quietly moved past the Bitter Lesson. In 2023 the story was naive parameter/data scaling and “emergent” intelligence properties. But there wasn’t enough data or compute to keep pushing in that direction, and the gains from it have been sublinear anyway.

Now, the labs spend enormous effort curating data pipelines to fit the models to a large assortment of very specific tools, tasks, harnesses, domains, etc. They also kind of fit to benchmarks by creating loads of synthetic training data that resembles benchmark tasks. None of this feels like the “scale up primitive methods and turn off your brain” message Sutton originally delivered.

But I thought Anthropic was an altruistic organization devoted to the betterment of humanity…

It seems to me that the model struggles to have enough general intelligence, knowledge, or reasoning capacity for arbitrary prompted tool calling. At this size, not surprising.

I am VERY interested in seeing how it could perform with some fine-tuning for a specific family of tools/tasks. That would be a great addition to the demo.

Actually I have been quite curious about this. I have a Qwen 3.5 800M model kept in memory just for this type of processing, and I set my small_model setting to use it. But it isn’t receiving any requests and somehow the titles are still generated…

> albeit at a lower tok/s rate

Not always! I get 80-100 tok/s from Qwen 3.6 35B-A3B on a MacBook Pro thanks to MTP. With long contexts that dips to around 50-60. However, prefill is much slower than API models. So it becomes really, really, really critical to not have cache misses.

I almost feel like he is just trolling for attention to justify his job. He posted a fake-outrage meltdown tweet about commodification of LLMs being the enemy of America or some nonsense. Paragraphs without a single uppercase character. In fact lately he seems to be alternating between normal English prose and “i am ai bro” affected writing, which I assume is some kind of engagement A/B testing.

Recurring subscription fees are not “tools” the employee owns, they are employer business expenses.

There are statutes requiring employers to fully reimburse business expenses incurred by the employee. Example: Cal. Lab. Code § 2802.

LLMs aren't finished products, though. And since they are basically a distillation of the collective works of all of humanity, it would make a lot of sense for them to wind up as public goods rather than some special advantage held by a select few.

I don’t quite understand what the argument is. Government does as a matter of fact operate on the frontier of new technology. (It’s how we got the web.) Why shouldn’t it, exactly?

Also, the US has been involved in AI research since the 1940s. So it’s not exactly a new thing.