I, for one, welcome our new Chinese overlords.
[flagged]
Congratulations, it turned out great!
Open Weights is where the action is at in the past couple months, I’d have to think the US frontier labs are getting nervous. Like Anthropic hasn’t released anything pushing the frontier since “the event” earlier this summer.
[flagged]
I was briefly impressed that https://hylang.org/ had released a 4.0 version!
I tried benchmarking it, but it keeps timing out/rate limiting, so the current provider(s) are unusable.
Unfortunately codebuddy login is not working for me in the United States of America
> open-sources
link to source code?
I am going to bring up graph issue for everyone of these announcements.
They all suck.
They shoulda put their stick where they belong, not at far left.
It just makes comparison to Deepseek 90% of them time as Hy4 has nothing to show off.
I'm liking where LLMs are headed:
They can do the difficult small level optimization, the boring but tedious code but cannot be tasteful.
That means I'm more valuable and more productive. Good stuff
This guy gets it.
Maybe's its a problem with the hosting at novita.ai but I didn't got much useful out of this model as a coding agent.
Novita does not offer Hy4-preview on either OpenRouter or their own model list. Maybe you confused it with Hy3.
You are correct.
I experimented with Hy3 for a project and was surprised with how good it was. I don't know if it's good for coding, but as a general purpose agentic model, it was only beaten by deepseek4-flash in our tests. It was so close to deepseek behaviour I kept thinking it must have been forked from it.
For the last few days I've been experimenting with the _free_ version of Hy3 offered by Opencode Go and I was also surprised to see how (relatively) good it is on coding tasks too.
The free quota from Opencode Go is also surprisingly generous, I perhaps hit limits one or two times and I've been using it _a lot_ for implementation tasks (using e.g. GLM-5.3-flash for working on specs and planning next steps).
is it just me or are the bar charts in the blog post strange? Higher numbers don't seem to correspond correctly to their actual height?
Noticed that as well
Probably AI generated.
But, what bars are clearly off? I couldn't spot any.
Looks okay to me.
The first column has both the Hy4 and Hy3 scores overlaid on one another (Hy4 is darker blue and the taller one), with both scores written below the top of the respective bar - maybe you're seeing that?
I wish model providers would stop committing chart crimes in their releases.
- if you're gonna order the rest of the bar chart by rank, order your model accordingly.
- if you're gonna highlight a winner in a table of benchmarks, don't highlight your entire model row in the table.
Etc etc
Read websites through llm.
I used Hy3 quite a bit for the type of tasks it was suited for. Excited about this. My one concern over Hy3 was speed. In theory, it could be served much faster as a smaller model but it was relatively slow everywhere I could get it (including from Tencent directly) but also several other inference providers.
In my evals, I saw an unprecedented jump between preview and final release on Hy3, from unusable to competitive. Did you see similar in preview vs release version?
Oh yes! I forgot about that. Yes, you can see this in benchmarks about hy3 preview and hy3 release still today because they measured them separately - it was significant.
Genuine Q about word optimization/token density:
If we create a stripped-down vocabulary with greater token density to use less resources and to resolve ambiguities earlier in the semantic process, aren't we creating NEWSPEAK and dragging along the worst aspects of it? The ambiguity and multi-valence of words is what creates more connections between words, increases the directionality of associations, and expands the potential subtlety and depth of meaning. By paring down (or requiring verifiability) we make it harder to say certain things, or at least make it harder to unintentionally say something that makes MORE or DEEPER sense than what we intended. If the token density becomes extreme, you're left with something like a calculator.
Maybe this is the ultimate path toward better coding? But the worse path toward better genuine thinking?
Nah reducing token length means that we're just reducing English down towards a programming language like a nice demi-glace
This is one of the dangers. AI boosters would say that humans already do this compression and it was accelerated by mass media and then the internet, and that model memory + context can be broad enough that compared to human capabilities the opportunities for depth and variability are even greater. But I think we know which way this optimization usually goes. Even the notion of a "fine-tune for subtlety" is a contradiction.
I don't think so. It's pretty clear that LLMs use the higher level layers for reasoning, so a bit of logorrhea very possibly enriches the result quality.
[flagged]
Vague and without substance. It easily passes as sarcasm, which means criticism but without any commentary, else it is sincere... but doesn't have any commentary.
Seriously, without China we'd just have two parasitic companies hoarding this tech and deciding whom and how is allowed to use it.
Claude and OpenAi are not allowed in Venezuela, so I thank China too and I swear to god I'll never use them and will be rooting for chinese models forever
Im not particularly fan of the chinese, but no chinese model asked for my citizen card yet to complete a task. And apparently no chinese provider uses persona to manage this kyc information. OpenAI does, in EU space. Just saying.
> Notably, Hy4 preview also contributed to its own development process, participating for the first time in the automated optimization of training methods, data strategies, evaluation frameworks, and low-level operators. The model proposed approaches, ran experiments, and iterated based on the results, with the resulting code, logs, and feedback feeding into subsequent rounds of exploration. This established an early-stage recursive self-improvement loop.
This reminds me of one of the predictions from https://ai-2027.com/ . Only that there it's "OpenBrain" doing this, not the Chinese. And the authors of that paper were also slightly wrong about "Mid 2026: China Wakes Up": China woke up already a while ago. And:
> But China is falling behind on AI algorithms due to their weaker models. The Chinese intelligence agencies—among the best in the world—double down on their plans to steal OpenBrain’s weights.
No need to steal anything, they have already caught up.
And then there's this prediction for February 2027:
> Officials are most interested in its cyberwarfare capabilities: Agent-2 is “only” a little worse than the best human hackers
I think we're past that point now, too…
Don’t need a “better” hacker if you have ten thousand AIs all trying literally every single possible thing to exploit a system with. The main issue is that this will eventually bring down the exploitation cost enough to target very minor targets who weren’t worth it before.
The AI 2027 paper/website is exactly the same as random guesses from tech bros after a couple of beers telling you what they think the future will be. It has nothing to do with political theory, economic theory, game theory, or any other quasi-scientific or rigorous evaluation of real world events and predictable outcomes. It's just vibes. If they're wrong nobody will notice, if they're right people will call them geniuses.
Just like how Windows 95 contributed to its own development process.
If the distillation "attacks" created useful inputs to open weight models, ai-2027 was directionally correct that the Chinese would find ways to extract IP from western firms. (Scaled account creation and grinding outputs etc is not a dramatic story element as spies, though!)
Whether the distillation has constituted "attacks" or has or will meet the bar of "stealing" IP is not super interesting to me, though.
Is anyone here working on a problem for which current generation LLMs are inadequate, but that could possibly be solved by the next release of a first tier LLM?
Or is it like bicycles? Unless your problem is named Tadej, you don't need a $13,000 bike.
I was given a picture cube, which is like a Rubik's cube but every side is a unique picture. It came scrambled and I don't have an original reference image. I like to take videos of it and give it to llms to solve. I call it my agi test because it hasn't been solved yet
I asked a current generation LLM to make me $1k a week and it hasn't so far.
For me personally, the answer is no. Fable is adequate to do basically anything I want to do. My perspective, broadly speaking, is that we've saturated most of the benchmarks because we've largely saturated our capacity to verify models' work at scale. What's left is context-bound verification, i.e. the problem of ensuring that output matches intent and ambiguities in prompting were resolved correctly. Further advances in autonomy do not make that latter verification problem easier. If anything they make it harder as the output per task becomes more complex and therefore more taxing for a human to verify.
The solution to that (to my mind) would be not a better model but a basic shift in architecture beyond the current paradigm and into a setup where agents have durable, plastic memories and undergo contextual individuation over time. But at that point agents start to become quasi-persons and not tools.
Scientific physics simulations - even the frontier models just engage in rationalization of obviously unphysical results instead of understanding the system. They have the rote knowledge but fail to apply it unless their hand is held through the process.
3d modeling to an STL a part compatible to a visible cable raceway still fails even if I let Claude Fable use me as a robot that measures with calipers.
I saw a laptop earlier in the train that I asked ChatGPT, Claude and Gemini what it was, providing a brand, screen size and ports description. Gemini could never figure it out, Claude and ChatGPT eventually did, after multiple rounds of indirection, giving completely wrong answers (there was a perfect match for the problem statement, they all explored alternatives first). LLMs are (probably) amazing at things I don't care about, and still suck at the mundane stuff you would have the marketing tell you they excel at.
Yes. Most of us are, still. The frontier is currently both at expanding ‘common sense’ / non-cheating outcomes for imprecisely specified software (that’s all software), and at expanding autonomy - ability to work longer unsupervised with success, oh, and also at expanding outside contextual reasoning about what’s being built, as in “hmm, that doesn’t look right or make sense, let me explore that.”
My experience is that even Opus 5 still tends to write buggy or low-quality code and makes serious mistakes when analyzing code. It's a lot better than before but still not something I trust. I've had less experience with Fable since I can't use it at work; I hear it's a step up but still has its limits.
For large tasks like a web browser or a compiler, even expensive swarms of frontier LLMs have not been shown capable of producing codebases that actually work. (Anthropic built a C compiler with Opus 4.6 but it lacked optimizations and apparently hit a complexity wall.)
I also want to use LLMs for reverse engineering, but apparently it's pretty hit-or-miss, especially if you're forced to use open-source models to avoid restrictions.
This reply is particularly interesting to me because most of my experience with actually using LLMs to get work done is with coding agents. But I only have a fairly narrow set of experiences: two pretty large solo Flutter projects. I am currently really pleased with Gemini as a coding agent. It could improve, but I think improvements are going to come from marginal gains in the harness and training material so it can catch things like misconfigured permissions in platform specific areas.
It's also interesting because, while coding agents are important and are a notable success, they are never going to be a multi trillion dollar business. And are there any other domains where LLMs have such a large impact?
I think the tech analogy for frontier models is going to be super computers.
Super computers keep getting better but most people don't need them for most things.
2000's supercomputer is today's (highest end) smartphone performance tho
Hardware debugging and firmware details lead to thinking/testing loops on all but the frontier here.
[dead]
Yes, lots – I think that folks will hopefully discover more of these as they scale up their ambition, now that LLMs make a lot of previously difficult things far easier.
Infra as code and devops shit. Fable is there in general because things it doesn’t know I can point at documentation and have it do a reasonable job. Opus 5 sucks. If I don’t have fable quota, I drop to Opus 4.8 and hold its hand.
I get buy with very cheap models and actually using my brain, you don't need these SOTA models. China will definitely win this AI 'war'
If the models stay open, it seems like everybody but anthropic/openai wins. i literally can’t see a downside. We can post-train the models to know about tienanmen square.
I want to be able to generate my own Simlilirian movie by dumping the content of a book into an LLM.
Both animated and live action results would be acceptable.
Unfortunately most existing LLMs lack the capability to maintain context across tens of thousands of frames.
Since live action results are acceptable, this is already possible with current day LLMs. Just instruct one to hire a writer, director, cast, and crew to make the movie.
Plus, the token costs involved should be pretty low! (Other costs may not be.)
How would a computer generated video be live action?
I think this is the best and most realistic reply so far: the ability to do this is close enough, and things like AI music are hints that there is a business model for this. Maybe I'm just jaded about CGI effects in movies currently, but I think the fact that people except that kind of thing as entertainment means you might get away with a fully AI movie that people will pay for.
There are two more points in favor of this kind of AI movie project: there's zero chance that anyone would greenlight a Hollywood budget for the Silmarillion, and it is beyond human capability to write that screenplay.
The results are boring. Not because the content is boring, but because you can so easily remix the results. Human curation is what creates value with these, not dumping and consuming. A personal perspective of a human being ups the respect, where the exact same sentences generated by an LLM carry no such value.
That sounds like an interesting challenge. Have you seriously considered solving it? Because in about 10 seconds I came up with a process that should work, provided enough compute power. Simply model the traditional film making process by starting with a script, character stories. Design your world, then design the storyboard, and all the scenes. Create a list of all the visual elements that need to be replicated between all the scenes. Then you have to built prompts and reference art of the objects, faces, people. Make sure to do multiple takes of each scene, and have the vLLM critique and analyze the performances and technicalities. Should work?
I think, also, like in the traditional film makers career, this process should be built iteratively, start with a fast food commercial, then do a music video, then you can probably do a short film. Continue to improve the process, and one day I’m sure the LLM film studio can make you any movie you want, provided you have enough tokens.
On that topic, check higgsfield cinema studio 4; they already provide amazing tech for the cinematic experience, somewhat similar to what you are describing.
Nobody wants to watch such films, they want to muck with the filmmaker, the generation apparatus. That's the product, if there is one here, and absolutely not the 3 hour epic that's spit out with 4 variations to choose between. That's work. We'll have other LLMs pointlessly tell us which should be watched, we'll view a summary, and vote the Oscar on that.
I'm getting a Poe's Law feeling. I'm genuinely unsure about whether this post is a stone cold parody or not. I think I need to turn off the internet and go to bed.
EDIT: Your username doesn't help, either.
A spanish rock solved that problem for free.
¿Como?
This is the exact same type of comment I heard about computer hardware upgrades for three decades in a row.
“Very few people actually require a Pentium workstation, a 486 is perfectly adequate for the majority”
The logical fallacy is taking an extant distribution of “product capability” that is priced to fit what the market will bear and assuming the “next upgrade” simply tacks on a little bit more to the right hand rail of that curve.
No!
It shifts the entire curve!
Everything for everyone gets better and the top 1% of the most demanding users will continue to pay the same-ish premium.
“Nothing” will change.
Look at it this way: you can buy a $200 laptop for your kid or a $20,000 Mac with an M5 Ultra processor.
BOTH are vastly more powerful than either a $200 PC or a $20,000 “workstation” from 20+ years ago.
Look at: https://arena.ai/leaderboard/text?q=openai&utm_source=chatgp...
The “budget” 5.5 Instant model beats o1 and o3 which were “pro” models at the time of their release!
Intel didn't just surf some natural wave of demand for higher power personal computers. Intel found new needs for powerful PCs, especially in gaming, and they put a lot of marketing and industry relations dollars behind PC gaming.
In other words. PC users didn't figure out that they could buy super powerful PCs and play games on them, that was a carefully managed market transition.
What is going to do the same for LLMs?
> Intel found new needs for powerful PCs
It wasn't "Intel" that found new uses for PCs, it was everybody who found new uses for them. Billions of people and millions of companies found uses for "more computer power".
It was only the journalists with limited imaginations (and no industry experience) who struggled to come up with potential uses.
> carefully managed market transition.
You make it sound like a conspiracy! It wasn't. It was simple capitalist competition. If Intel hadn't improved their products, their competitors would have left them behind.
That very nearly happened ten years ago because Intel become stuck on the 14nm process and their products stagnated while Apple, ARM, and AMD lapped them repeatedly.
> What is going to do the same for LLMs?
Everybody.
Are you saying that unless you're "carefully managed" by some third-party, you could not find any use for "unlimited intelligence on tap"?
> Tencent has released and open-sourced Tencent Hy4 preview, a next-generation large language model with 770B total parameters and 49B active parameters, and a context window exceeding 1M tokens.
There are no open source models, at least not useful ones (yet) [0]. Open weight is not the same as open source. The current "open weight" models are just opaque binary blobs you can run on your own computer instead of through a web API.
Imagine thinking that running a Photoshop binary on your own computer instead of through a SaaS web app means that it's "open source". Of course you think that's ridiculous.
You can open source dataset without all the details how it was assembled.
Models are lossy compressed datasets you can pick up and amend (fine tune / continue training / alter) according to license they were released under.
Hy4 is released under OSI approved Apache License 2.0.
Parent poster is technically right - open “source” implies the source used to make something is open. The model source is training data and code, not just weights.
But the reality is, the weights are a useful artifact that you can use to create derivative works. So, dismissing it as a photoshop binary is as technically wrong as calling it open source.
Countries that aren’t competitive need access to training datasets so that they may train their own similarly capable models and be sure of the inputs. Governments cannot blindly trust open weight models from China and the US.
So windows is open source because the binaries are a lossy compression of the original source?
Weights are not binary. A model is created at init time, with random values. After that, it is being modified using data. The key point is that the labs modify the models "as weights". That means that weights are the intended / preferred way of modifying a model. Which, coincidentally, matches the definition of source in Apache 2.0. There is no "higher level" place where editing takes place. It all happens in weight space. Through the license you get the same rights as the lab that created it: view, inspect, run, modify, re-release. That's it. That's the only thing a license can grant you.
The rest is semantics, misunderstandings, and FUD. A model released under an open source license is open source. Training data is lab knowhow / IP. Which, historically, has never been required for any open source release.
Hy4 apparently has ludicrous traction on OpenRouter already (https://openrouter.ai/tencent/hy4-preview), with trillions of tokens processed in a couple days: more than GLM 5.3 in a week. That said, it's relatively cheap with a 5% cache cost when everyone is still doing 10%/20% cache costs, so Hy4 may be more compelling.
It's very likely tencent games those stats, buying their own tokens.
Which explains why almost none of my request go though
If you’re Tencent you can just plug it into some field somewhere that lots of people see right? Like how Meta could put their model on Instagram search
[flagged]
i'd be curious if openrouter is just being gamed by these publishers by paying for the exposure.
wouldn't trust they dont do Capitalism like the rest of the AI field.
Of course they are. Of course they do. Nobody should be surprised by this.
>dont do Capitalism like the rest of the AI field
Like lobbying the US president to harm their competitors?
I would suggest "lobbying" is not the correct word to describe all the corruption going on in the current USA administration cesspool.
lobbying/legalised bribery hard to say where one ends and another begins at times.
Well, it sure as hell isn't capitalism.
Lmao that’s exactly capitalism
Where in the Wealth of Nations does a Trump appear?
i mean, theres capitalism as the ideal, and there is capitalism in practice, so maybe you are both right...
I wrote about this a couple of weeks ago. It's actually often the biggest cost and it tends to be hidden away on most platforms!
https://martinalderson.com/posts/watch-out-for-cache-read-co...
Btw I still haven't came across any decent model that is <$0.01/MTok cache costs apart from deepseek thru their official API (even with the price increases).
Seems like a bit of an opportunity for someone to take - drop cache read costs significantly.
For me, an average long session results in about 200-300M cached input, 4-800K input, 2-400K output. Mostly the lower bound. Output depends on how much the model thinks.
There are two problems here:
- cache hit pricing (both Muse Spark 1.2 Contributor and MiMo 2.5 are around the $0.002-3/M mark)
- cache persistence time
Muse Spark drops the cache in less than 5m. MiMo keeps it around for at least an hour based on my experience with whoever is serving it for OpenCode. This difference itself will inflate bills massively.
A 500K token input repeatedly read by MS 1.2 for full input price 12 times an hour = $0.60. You would be expecting $0.012. So a 50x difference. Same thing on MiMo 2.5 is $0.018 because of longer cache times.
That's because Deepseek invented the paradigm of prompt caching, they are the SOTA when it comes these techniques. Despite them open sourcing all their research, nobody beats them.
edit: I do wish openrouter would let you sort providers by Cache Hit % and Cache cost. These are the only things that matter to me at this point when choosing a provider.
>Deepseek invented the paradigm of prompt caching
Caching was always here, you don't need to do anything special to get it on a single user local backend running a base model or a chatbot in the first place. Among commercial providers, OpenAI adopted it in 4o first.
Cache hit % on openrouter is not a good metric, it's mainly driven by openrouter's own provider juggling than the providers themselves
> Cache hit %
I thought you had to actively manage caches, do you not?
This is not true, there isn't even a way to see a cache hit % model for a specific model, that wouldn't make any sense. You are confusing what I'm saying with cache cost, that has nothing to do with effective cache hit %. I'm talking about when you click on a specific provider for a specific model, you can scroll down on the view and see their cache hit % for that model [0].
These cache Hit % are accurate, I've done a ton of testing of this myself. The cache hit % is one of the most important metrics as far as estimating cost. There are many providers with cheap cache reads, but have an effective cache hit % of 30%, making their cheaper cache pricing meaningless compared to another provider who charges more but has a 85% cache hit percentage.
[0]: https://openrouter.ai/deepseek/deepseek-v4-flash-0731?endpoi...
scroll down on the provider/model card and you'll see a field called cache hit %, its different for every provider/model.
I don't use routing on openrouter, I strictly use models with a single provider and no fallback, at least for use with harnesses its pretty dumb to route requests to multiple providers you are busting your cache every other request and increasing costs by 20-50%.
I think you're arguing the same general point that the person you're responding to is. But you're saying he's not understanding - he understands that they report a cache hit % but you can't look at that public metric with any level of accuracy _because_ most people aren't pinning their providers and they _are_ getting juggled around which is bringing that metric down. That's not to say that specific providers might have issues or worse cache implementations - but it stands that if openrouter is juggling the requests back and forth by default then _that alone_ is breaking caches on those requests in huge numbers.
I know what they're saying. Why would openrouter calculate it thay way lol. They obviously dont. Think for a sec, they arent idiots.
OpenRouter randomizes which provider gets your request by default right? I think you have to pass a specific provider in the request to prevent that. (Or set up a preset or something.)
This behavior makes it so you don't benefit much from the caching, unless you pin it to a single provider.
> This behavior makes it so you don't benefit much from the caching
I don't believe this is correct? AFAIK once it routes you to a provider for a given conversation that choice is sticky unless you hit technical difficulties. (It's more complicated than that, they recently added named routing strategies that you can append to the model name.)
IMO the relevant metric is cache TTL which isn't typically published AFAIK.
> OpenRouter randomizes which provider gets your request by default right?
I'm not sure it's wholey accurate to say they "randomize" the provider, rather my assumption based on usage is that it's something like cheapest-ish/responded to the request within some reasonable-ish time/etc algorithm that chooses the provider on each request - which seems, remarkably questionable in terms of optimizing for user experience or hidden user costs.
> This behavior makes it so you don't benefit much from the caching, unless you pin it to a single provider.
I so very much recommend this approach. My avenues that automate llm calls to openrouter are setup to make api reqs to openrouter to determine best price/response/etc and then pin the request to that (and, preferably, a fallback if there's reasonable difference between #1 and #2) provider for that session. Otherwise you're going to have a bad time.
I'd imagine this could make things interesting in cases where one provider is offering different quants than the others and openrouter is just swapping you back and forth on a long agentic session.
You can click the table headers to sort Ascending/Descending.
You can sort by the cost cache cose, but you cannot by cache hit %. You have to click on the provider and see what their cache hit % is. A provider could have a super low cache cost, but a 50% cache hit percentage, making the cheap price of cache read's meaningless.
Wait, what does that number mean? I thought it always uses the cache price when the prefix matches.
When the prefix matches a request sent to the same Providor. The thing is the TTL is different for each provider, some cache for 5 minutes some cache for 1hr. Its ideal to only use one provider per agent session / and per model with the best cache hit % if you care about costs.
> [...] Let's maybe add a helmet? It could improve riding theme, but may obscure head. Maybe a small cycling cap or helmet? The user didn't ask; can add red helmet? Might be cute. But pelican with big beak; a helmet might obscure. Better maybe no.
> Maybe add sunglasses? no.
> Maybe add water? no.
https://tools.simonwillison.net/markdown-svg-renderer#url=ht...
> Let's maybe add comments? The final code can have comments. Fine.
Did he seriously automate away one of the best quirks of his blog posts, i.e. evaluating new models with a touch of fun? I read AI slop all day, thanks.
No one’s talking about how good the final product is.
Edit: someone else commented that as I was typing this, lol.
This is actually pretty good!
If someone can look at that reasoning trace and see a stochastic parrot next word prediction machine, we don't understand those words in the same way.
LLM is "stochastic parrot next word prediction machine"; it's just that this "stochastic parrot next word prediction machine" have proven to be smarter than most people. I mean, this already happened with AlphaGo too.
I still call them stochastic parrots, but believe what they are revealing is that we are all stochastic parrots to some extent. I simply don't see how biological computation (i.e. thinking) can be anything else. Similar to the reveal in west world, we are likely much simpler than we give ourselves credit for.
A "train of thought" can be seen as a trace of a depth first search where the preceding trace is used to guide termination and next expansion decisions. A similar concept, "taboo search", exists in classical constraint optimization where previous solutions are fit to a model that guides future expansion (but as the name "taboo" implies, away from uninteresting solutions).
We also have harnesses that perform breath first search.
If I tried to describe what it means to "think deeply", I would probably say a combination of both.
Ultimately I believe that we will surpass human capabilities but fail with alignment. Handing the world's resources over to stochastic systems that can evolve faster than we can reason about them simply leaves too many "interesting" outcomes that do not end well. I also expect the failure modes will be totally non-obvious.
The stochastic parrot epithet is so 4 months ago
Yeah, it has been clear for a long time that there is reasoning and mental modeling going on here.
The other option is that you do understand those words the same way, and the people making these (now nonsensical) anti-AI claims simply aren’t talking about the same programs/models we are. Their idea of SOTA is when chatgpt.com launched.
If you took a point sample pre-Opus, and didn’t write a good prompt, of course you would think all AI programming was worthless slop.
[flagged]
Simon’s karma exceeds yours by about 1,300x — so most of us disagree with you. Like me, I disagree with you. I like the pelican benchmarking, and I like it when Simon stays on top of models for both public good reasons and because it saves me time and effort.
If I had to choose between more or less pelicans, I’d prefer more pelicans.
Comments that the HN community find interesting are surfaced higher.
Just tap on the [-], and upvote what you find more interesting :)
Nah, keep it up. I just come to the reviews for the pelican!
I think the pelican thing is a bit tired at this point, however I find Simon's comments to be high quality.
Same.
Is the broken English an optimization or a byproduct of the model being developed in China?
qwen3.8-flash-next also 'thinks' like this in its thinking stage before output, watching it 'think' in opencode, but it produces syntax correct and grammatically correct code comments, changelogs and readme type files.
Likely something that was first made especially obvious by Chinese models and then became something worth optimizing for in English too.
Chinese can be extremely information-dense in token terms, though it depends on the tokenizer. Roughly speaking, you can pack more "meaning" into a short sequence than English often allows for. That's why "caveman" reasoning is a pretty good fit.
There's a difference between bolting caveman speak onto an existing model and training a model to reason that way, though. If you just force an existing model to be concise in outputs, you're artificially reducing its available reasoning steps and can possibly prevent useful exploration or verification. If it's trained specifically to use compressed reasoning, it can learn to represent the same intermediate ideas in fewer generated tokens, cutting the number of sequential inference steps without necessarily sacrificing the useful reasoning itself.
It's not so much inherently a Chinese-model trait, but Chinese models could definitely have helped demonstrate how effective very compressed reasoning traces can be.
There are few tests of this, but one example I thought was interesting was here: https://github.com/PastaPastaPasta/llm-chinese-english
I wouldn't say it was Chinese specifically that was emulated, but it got people thinking about tokenizers and representation efficiency, and how natural English is rather inefficient.
Saving tokens
More intelligent and shorter:
Maybe add a small cycling cap or helmet if it doesn’t obscure the head.
Optimization. Why use many word when few word do trick?
Optimization on a idiosyncrasy. The same thing that makes Claude repeat "That was the most important thing you said in this whole conversation" is what makes grug speak optimize on token usage.
Real humans get non-primary information from word variation. It's reasonable to hypothesize that it has a role in thinking things, because it endures. Our languages need to breathe over time, and flourishing might be one of the aspects that allows that breathing space.
What I find funny about "why use many word when few word do trick?" is that it's only slightly shorter than the regular "why use many words when few words do the trick?"
I always figured that was part of the joke, because a writer came up with it, and a writer would know (I assume?).
> Why use many word when few word do trick?
Be concise.
OR
OR
etc."Omit needless words."
-- William Strunk Jr. and E.B. White., The Elements of Style
When GPT-5.6-sol's reasoning traces were leaked, they also used "caveman speak". Definitely a token efficiency optimization
I can't help but imagine agents using caveman speak sometimes start behaving in a stereotypically caveman manner, even if it's subtle. Is there a chance the agent does less reasoning because of it?
Just so we are clear, no "caveman" spoke English. "Caveman speak" is just shortening the vocabulary of english, not a "caveman language". Given this, your concerns for "stereotypical caveman manner" makes very little sense since what caveman are you talking about?
Training a variant to reason in early modern english in the style of the tudor elites might be an amusing way to test for that.
"Neuralese"
some people made a 'caveman' speak qwen as a joke
It's not exactly a joke, it does reduce the amount of tokens. However, it does not improve performance (fine tunes are finnecky things, hard to get one right).
Personally the only 'enthusiast' modified qwen 3.6 27b or 3.6 35b-a3b I've found useful are the ones that have been run through heretic and adversarial data sets for innocent/dangerous prompts, to produce uncensored LLMs. They have some niche non-coding uses for things that a commercial LLM will never talk about.
I think those are mostly vapor that runs on the small culture of "models should not be censored" thing. But from my experience, they unlock nothing meaningful.
Fine-tuning is great for really small models on specific applications, but it's not something that can essentially improve a more generic model.
That said, there seems to be a fine line in quantization+finetuning that could recover performance. It's just hard to get a hold of it (I feel it in some models, but it's hard to say yet; lots of small labs working on this RN).
The most interesting use I've found for them so far is strictly as a novelty. Give a chat session with one to a completely non technical person, who at least knows that openai and anthropic have some guard rails on stuff, and tell them to wild with something like "give me the precursors and chemical formulas for the precusors for crystal meth" and watch it answer.
Yep, but that's not changing the quality of the model. It's not an optimization in any sense (and it's a hit on productive workflows, possibly).
This is also likely to stop working as censoring moves to the training data source.