It's also pretty funny sometimes how it gives weird future roadmap estimates ("part 2 - 3 weeks, part 3 - 2 months", etc.) and when you tell it to actually do those changes it's pretty much done in half an hour
I've long believed those numbers were faked by Anthropic/OpenAI to serve as a form of advertisement. The estimates are impossible to verify and their ability to do "2 days of work" in 10 minutes will presumably make the user go "Wow, I just saved SO much time!" Plus, the unnecessary text eats up the users' tokens so it helps the companies on the backend, as well.
I tend to be cynical about AI companies, but I'm guessing the bad estimates more just come from a complete lack of actual data it could use for that so it's more or less a hallucination.
I agree with you that labs are benefiting from those outputs but I'm skeptical that labs are purposefully training the models to produce those outputs.
Raw pre-training data includes plenty of conversations between professional builders and some of those include estimates.
I believe the outputs are a training coincidence with consequences that are opportunitistic for the labs.
Even for humans the estimates are way off, unless it's based on data that has some serious padding.
That said, it'll often say "2 days of work" and then complete the coding in 30 minutes, and while that's amusing, afterwards, I'll need to manually test, or send to other people for review, or realize the agent only actually did half the work and I need to do a second pass (or a third etc.) and then often getting the feature in does genuinely take two days.
It generates tokens by estimating what the next token is going to be.
Sure it cannot think like a human, but given it's input, it should give a good statistical answer (approximating not of how long it actually takes, but what a human would say how long it takes).
Therein lies the rub, no? To accurately predict the next token produced by a process, it’s necessary to model that process. If the process is a human attempting to estimate the duration of a task, then in some sense the LLM is modeling the estimation process. We’re well past the point where it’s credible to claim that LLMs just regurgitate their training data.
Obviously there isn't a hidden corpus of logs of coding chatbot assistants that has been accumulating over the years, but these coding chatbot assistants output tokens that resemble how we all imagined a coding chatbot assistant would have operated had it existed in the first place to end up in a corpus. "Training material" includes supervised fine-tuning, preference training, RLHF, and so on, so that certain outputs (like these timeline estimates) may really have been decided (at some level of conscious awareness) by product teams.
you might like the stuff in my work of oh my pi, its a test bed for my ideas around making these tools more reliable. hoping to maybe have a native ui iter of the real thing that this is a test bed for this summer.
At that time the predominant view was that LLMs were nothing but stochastic parrots, that they would plateau, and that hallucinations couldn't be fixed.
At this point I doubt there are any AI sceptics left. That ship has long sailed.
The only thing that matters is whether the estimates are accurate, and AI can improve on that too.
Even humans only estimate based on neurons firing in prior patterns.
Actually in this case they possibly are estimates.
It's been known for some years[1] that LLMs do regression in-context. Frontier models have been trained against many, many issue text that include task break downs and estimates.
Interesting. So it may have learned how to estimate as a human but doesn’t understand that it doesn’t operate at that speed :D
I wonder if there’s a reasonable way to give an llm parameters that give it a concept of its own execution speed. Seems that could be useful for multiple purposes
I think people are continuing to view these systems as pure LLMs - when that ship sailed 6+ months ago. Between being able to review memory, using agent harnesses and sub agents and skills to go out and discover information - modern systems (Codex, Claude Code, Cursor) - use LLMs - but the LLM is only a small component of it. Compare what you get from sending a request to a chatbot like ChatGPT - to what you can from a modern harness. The output is influenced by the LLM, but it's no longer a "model making a token prediction based on training material and RLHF" - that's a very 2025 way of looking at these systems.
Even Gary Marcus is starting to come around and realize that his priors are no longer as relevant as they once were.
Will the 10T parameter Mythos model be released this month or next month?
They better soon because it is generally accepted that one of the reasons GPT 5.5 is better at hard tasks than Opus is because of its parameter size - and that Opus 4.8 remains competitive only be scaling test-time compute (see how many more tokens it uses than GPT 5.5)
Why ask me? Anyway, Mythos is not 10T. Anthropic confirmed the training run was under 10^26 flops. You can't train 10T to chincilla and stay under 10^26.
Anthropic also confirmed they will not release Mythos, only a "Mythos-class" model, whatever that means.
I should have stressed the symbolic part. Everyone has pivoted to symbolic systems like claude code and codex. They would no invest so heavily in such systems if they thought llms would deliver agi soon.
You think someone is, or even should, special case things like estimates? What else deserves that level of intervention so they look less dumb?
Logistics for getting to the car wash next door?
In the mean time, alas, no, we can see from actual prompts sent directly or through sub-agents, and actual replies, estimates remain LLM generated.
Though, this discussion here could change that, because indeed there is a lot of special casing and context stuffing going on, one of the oldest being today's date for example.
• • •
I did read the Claude Code leak, and use pi, etc. So I disagree with your premise rather strongly. Today's "systems" remain, roughly, piles of markdown and context engineering wrapped in UI affordances, and behave very similarly today to how they did in 2024 for those already engineering context and delegating.
I do a lot of code bisecting with Claude Code - and it spends hours running experiments - looking at experiment results, making guesses as to what to try next for an experiment - until it eventually comes around to a working code pattern. I mean - maybe this is as much a reflection on me as anything else - but it's pattern of logic isn't that much different from what I would do. It knows, in general, what tools and APIs it can call - it tries something - observes the result, and then comes back and tries different experiments based on success/failure - mostly efficiently bisecting to a solution.
I'm still lower-down of the capability scale - as I'm still manually directing agents to do these wiggins loops - obviously the next step up is to direct the code-loops which control the agents. I just haven't got my tooling nailed in place to the point where I find that's more productive.
I actually might agree with you that this is mostly just "next token prediction" - if I can concede that's really all I do as well.
> I actually might agree with you that this is mostly just "next token prediction" - if I can concede that's really all I do as well.
Yep. Pretty sure I've got an LLM inside too.
The other replies complaining that my thinking is so 2023 -- on the contrary, what's evolved is my own apprehension of how LLM-like most "responses" from humans prove as well.
To be sure, there are other mechanisms at play as well, significant differentiation in our... Volume of training material? Quantizations/compression? Model architecture? Just-ahead-of-time forward branching with back propagation? Double loop adaptive learning? You know, harnessing the LLM. :-) Dare we call it executive function?
LLM mode becomes particularly apparent when conversing with Alzheimer's patients in the stage where short term memories do not form but they retain access to long term memory up to, say, 5 years ago or so. Fifty years of who they are, and one can trigger nearly identical responses with nearly identical prompts.
But that same person may be able to debate 1950s politics while being unable to complete making a sandwich.
If they didn't know of new shortcuts for a task, would almost certainly not "estimate" but "intuit", or "instictively" respond (apply heuristics), largely based on their "priors" aka training material.
If you sit with them and chat a while, you'll even get the kind of looping you get from Qwen trying to think when context is too full.
And if we believe this at all, then ... we should stop scrolling tik tok. Time to read a book. Have an experience. Fine tune. :-)
All models do it. It's their training. They didn't have "a person does this in a week but an LLM could in a minute" in their training yet. They also don't have the concept of elapsed time unless you ask them how long something has taken.
I mean in general I'd rather take slightly inflated estimates than the odd sprint poker stuff where other devs and PMs negotiate hours down and before you know it you're also stuck fixing nitpicky reviewer comments on code that is already good enough and have to send a release at like 7 PM, ofc also without enough tests or even enough manual checks and testing, cause people repeatedly act against their self-interest and try to compress timelines, thinking that that's somehow good for them.
At least with AI that actually does things more quickly, there is a bit more breathing room (introducing AI is easier than changing a given environment).
Aside from that, I wonder how much variety there is in practice: between "Oh yeah, I added that new button while we were in the meeting" and "The new button feature will be ready in Q3 according to the roadmap, once we have sign-off from all the stakeholders."
I heard an anecdote. Guy spent several days trying to convince his AI agent to build a feature. Kept saying it was crazy complicated, would take weeks.
Finally he convinced it to try. It one shotted it in 30 seconds.
Turns out the agents' idea of what is hard and easy also comes from Common Crawl.
Uh Claude tries real hard to dodge work. Talks about how it’s really hard 10 PRs. Finally convince it to do as 1. It stops 10% through and says ok done with PR 1, we can work on the last 9 tomorrow. Ugh.
It repeats what it has seen in the training data. Expecting it to reason about the complexity of a task is a pipe dream. The best is to tell it not to come back with estimates, and when it does, remove them anyway.