Incredible, how an entire religion has sprung up around AGI.
Fortunately, OpenAI already wrote theirs down. Well, Microsoft[0] says they did, anyway. Some people claimed it was a secret only a few years ago, and since then LLMs have made it so much harder to tell the difference between leaks and hallucinated news saying this, but I can say there's at least a claim of a leak[1].
[0] https://blogs.microsoft.com/blog/2026/02/27/microsoft-and-op...
[1] It talks about it, but links to a paywalled site, so I still don't know what it is: https://techcrunch.com/2024/12/26/microsoft-and-openai-have-...
- the prevalence "How many |r|'s are in the word 'strawberry'?" esque questions that cause(d) LLMs to stumble
- context window issues
It would be naive to claim that there does not exist, or even that it would be difficult to construct/train, an interrogator that could reliably distinguish between an LLM and human chat instance.
[0]: https://archive.computerhistory.org/projects/chess/related_m...
The Turing test could also be considered equivalent to "can humans come up with questions that break the AI?" and the answer to that is still yes I'd say.
As for writing in general slop score is still higher than a human baseline for all models[1], so all a human tester has to do is grade it and make the human write a bunch, the interrogator is allowed to submit an arbitrarily long list of questions.
The actual quote is this though:
> hitting an AGI milestone or pursuing an IPO
So it seems softer than actually achieving AGI or finalising an IPO.
A year ago I would have said that was crazy. In the last month, I've been using Claude Code to write 20kloc of Rust code every day (and I review all of it).
A week is now a day. If that figure doubles, I have no idea what will happen to us. And I think it's coming.
Only one of this can be true. It's not a shame to say you don't bother reviewing it, in the future that may well be the norm.
I'm prompting every change set, reviewing the outputs, then reviewing the total changes.
I'm sitting at my PC all day doing this - I used to be productive in short bursts, now I'm productive all the time. It's addictive.
I used Claude a lot on a recent project where it probably wrote 15-20k lines in a month, and it was overall excellent.
I can't get Augment / Opus 4.5 to edit a few C++ files from within VSCode without going off on a wild goose chase or getting stuck in an infinite loop after I tell that it should be doing this: "oh, you're right, I need to do X", "To do X, I must understand how to do Y", "I see now that to do Y, I should look at at Z". "Let me look at Z", followed by: "oh, you're right, I need to do X"..
Have you read Alexandrescu’s “modern c++”? It’s like a piece of modern art but completely not self aware. There’s just something about C++ that lures intellectuals in; like ice age mammoths to a tar pit.
Small wonder LLMs also fall victims to C++’s deranged ways.
I think you've crossed the line from being an AI maxi to just rage baiting. This comment is a pointless anecdote at best, please take your terrible FOMO takes elsewhere.
Reviewing 1k lines of code an hour is a breakneck pace, are you spending 20 hours a day reviewing code?
Building things at a mature company with a market is a lot different than hacking together your own tools. There are a lot more people you can let down at scale.
(Which for anyone familiar with your long comment history as a regular HN poster, is comically absurd to imply. You've been reliably adamant that AI will demolish this or that entire industry overnight for years at this point).
We'll see who's right. I never said "overnight". Let's check in at the decade's end.
Y'all dunked on me in 2019 when I said AI was coming for Hollywood. Have you seen Seedance 2.0?
It's coming for us too. I've written five nines, active-active systems that handle billions of dollars of money movement daily. These systems can work in those contexts. I didn't think we'd be here this soon, and I actually thought LLMs were a dead end. I was wrong.
I'm not trying to sell Claude Code. I hate the concept of hyperscaler companies. I want there to be viable open source coding models - there just aren't. I'm merely reporting on my findings.
I sit at my machine for hours now in a prompt, review, test cycle. It's addictive. I'm getting more done at a faster rate than any time in my professional career. I'm excited, and I'm also worried. I don't know what happens after this.
If you've seen how much I praise AI, then you've also seen how much I rail against monopolies. I am worried these giant companies are going to take the means of production from us. I don't think enough people are freaking out about this. It's a very real possibility.
I'm just going to keep building. But you should pay close attention to what's happening.
Y'all dunked on me in 2019 when I said AI was coming for Hollywood. Have you seen Seedance 2.0?
Being right at the wrong time is often worse than just being outright wrong, I've found.
If you're generating 20kLoC per day, you definitely aren't reviewing it!
Are they going to get stock for it or is it a PIPE?
Personally, I don’t think I want to get in on this at retail prices.
It can both be true at the same time that AI going to disrupt our world and that being an AI lab is a terrible business.
AGI is an IPO.
Yes and it's actually hilarious: a system that can perform most economically valuable work better than humans, or specifically when the AI generates $100 billion in profits.
Those conditions are an IPO or reaching AGI [1].
Nvidia and SofBank will pay in installments.
Also very interesting that Microsoft decided to not invest in this round. A PR statement was made though [2].
[1] https://americanbazaaronline.com/2026/02/26/amazon-to-invest...
[2] https://openai.com/index/continuing-microsoft-partnership/