Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- I suspect you could plug GPT4 from 2023 into the integrations and workflows and it would be about the same.
The "models are getting better with every breath you take" is just an unproven claim that the AI companies want everyone to believe and want the evangelists to use in every argument.
The AI companies have to put out new models regularly, not because of any improvement but because the failure to do so will signal stagnation. That's how it works in the tech industry. It's not like the wine or cheese industry where you just follow your hundreds-of-years-old recipe exactly and people buy the product because of that.
by kazinator - nowhere near plateau, but right at the inflection point of diminishing returns imoby kimjune01
- This is the best take. Return on capital is diminishing from intense competition so improvements get hidden away like the gems they are.
1) Frontier labs have no incentive to give the general public their best anymore; it's instantly distilled off of them. Why not charge governments and big corps real money to use the real good stuff instead? 2) So we get distilled-off-frontier public APIs like 5.6 and Fable/Opus. And the open source labs are distilling off of those. 3) There's a lot of benchmark hacking right now among all the publicly available models, actual usability of Opus for coding is far below its benchmarks suggest. 4) But context window, cybersecurity, logical coherence, and tool usage are absolutely better on Fable and Sol. It looks to me their internal tools definitely even better and not plateauing. But we won't get to use it.
by hiramwen - Obviously not, have you seen the 10 mathematical advances OpenAI found with Astra?
- I don't think the models are the bottleneck. Instead, how and where we deploy them remains significantly underrated and underutilized.
Like imagine how crazy that you can get human-like intelligence in a small device, we should be able to do more than a chat interface.
by _Yassine_ - I think the giant leap in AI technology was the Transformer. ChatGPT is based on it, with an incredible chat user interface. In terms of technology, there has been no similarly significant innovation since then. But user interfaces and training data keep improving, which continues to increase our productivity.
- > with an incredible chat user interface
what makes it incredible?
by andsoitis - No, LLMs have not plateaued, and each of the latest releases has been a proof of that.
Fable 5 is a model you instantly FEEL how smart and superior it is. GPT 5.6 Sol is a HUGE incremental improvement in multiple directions and dimensions.
DeepSeek v4 Flash 0731 is a huge improvement over the preview version, using exactly the same architecture.
Kimi K3 gets open weight models very, very close to the frontier.
No my friend, we are not done yet.
by iagooar - > you instantly FEEL how smart and superior it is
I felt taken by the change in the system model personality and writing style compared to opus, but I also found it to be much less impressive than I was expecting - let alone that the cost was incredibly high when not given for free.
Are you sure your reaction is not primarily to the improved ergonomics of Fable?
by taurath - In terms of the amount of information in the current LLMs, I think the largest is 5.6 trillion bytes of memory, it is miniscule to what is actually out there. There are over 13 zettabytes of information on the internet. Part of the capacity is about absorbing information. It has a lot further to go. If there are emergent properties with additional knowledge in LLMs, it is scratching the surface. I wonder what will happen with zettabyte computing.by NishanStepak
- AI is beginning to do PHD-level math.
If we're nearing the point where you can spin up 1,000 agents to look for ways improve existing models, then automatically run experiments to validate those ideas, we're more or less at RSI.
As always with AI progress, compute will bottleneck this early on, but a few efficiency improvements could dramatically increase this pace of progress.
I suspect we are at most 24 months from FOOM, but I suspect within about 6-12 months most frontier AI labs will be claiming the majority of their AI research will be AI-driven.
by kypro - I think the big change was agents. The LLM improvements after that point have been relatively minor in impact.by tripleee
- Agents are the proof that LLMs Plateaued : you need the loop to get further.by ldng
- I'm not answering your actual question, but... even if they have totally plateaued technically, there's still some more improvement left to be had from people learning how best to use them (and when not to).by AnimalMuppet
- how can you learn to use a tool that keeps changing unpredictably?by vrighter
- How much more productivity will we be able to squeeze from them?
Probably, depends on how you measure productivity.
If you measure productivity in terms of number of automated bureaucratic events (e.g. creating files, organizing files, generating lines of code, finding bugs in code, generating emails, responding to email) then yes productivity will continue to increase because LLM's are the killer app for increasing bureaucratic events.
If you measure productivity in terms of changes to the material world (e.g. traditional things like trade goods, buildings, food stuffs, irrigation systems, transportation networks, etc.) then no because LLM's have little meaningful impact on those activities...no AGI is going to harvest lettuce for our wedge salads).
by brudgers - The advances in robotics makes it look like a humanoid robot that could harvest lettuce is only a decade or two away. Making the robot is the easy part. The code to drive it has been the hard part. Until now, that is.by fragmede
- Obviously not. And if you look at the things LLMs do poorly there's clearly plenty of room for improvement.by wmf
- LLM shill argument in a nutshell: "while there is room for improvement in catalytic converters, they are improving practically by the month! Within a decade, internal combustion engines will put out air that is so breathable, it could be used for ventilating a maternity ward."by kazinator