Which parts of the writing are yours, and which were by LLMs? The density of LLM constructs here is very high, and the style doesn't feel consistent with your old blog posts. (And I wish it was written in the style of those older posts, which read clearly human).
Like, did you write this in another language and use LLMs to translate? Write an outline and LLMs wrote out the rest? Write this amount of text that was then edited by LLMs?
My blog posts are written more slowly - they are smaller and have low frequency. I write them on my own.
This text was drafted by myself and Claude Code "improved" the English, with lots of interference and rewriting from me - I can see that you do not like the "improvements" and maybe the overall result is not as good as it should. Not being an English native it is harder for me to judge (it does look OK to me) but this comment of yours is important.
I will consider rewriting this calmly. But it was important to get that data out - we have a social emergency in several countries.
I read the entire article, and it was very well written. I had no problems reading any of it.
I have come to loathe the posts from people who are so filled with AI angst (hate) that they're looking to jump on anything they can. I would ignore the previous suggestion.
As a data point for you, using LLMs to "improve" your English nowadays has the opposite effect. I'd rather have spelling/grammar errors than read LLM output as an article.
For what it’s worth, I would much rather read quirky ESL English than Claudish. Claudish gives me a strong impression of laziness/SEO/race-to-the-bottom/enshittification/slop. Quirky ESL English gives the exact opposite impression: there’s a real person on the other end who’s really trying!
(Also, your English is clearly very good, and doesn’t even need a cleanup.)
> - Why are they stacking Fable xHigh against Carbon Med?
Because on that specific benchmark (FrontierCode 2.0), Medium is Carbon's best effort level and xhigh is Fable 5.2's best effort level. You can see that on the graph your next point is about.
And it's just that one benchmark that's at medium. "All scores show each model’s highest-available effort unless otherwise noted".
> - How did they produce such a bizarre effort-intel curve for Carbon that goes against every available benchmark?
The graph is labeled as being about FrontierCode 2.0. Those appear to be Carbon's scores on the official FrontierCode 2.0 leaderboard: https://cognition.com/blog/frontier-code-2.0
> - Why did they leave Carbon-- positioned primarily as a cyber model-- off of the single cyber benchmark comparison?
> - Why do they switch to comparing against Argon (rather than carbon) for robodojo?
Because no numbers are available for Carbon for those benchmarks, and the model isn't publicly available so they couldn't run the benchmarks themselves?
And from your followup:
> where's Astra on that CVE chart
To run Astra on a vulnerability finding eval, you'd need to be part of OpenAI's verified cybersecurity program. Otherwise the results would be tainted by refusals. Do you expect that OpenAI would grant such access in this case?
> Because on that specific benchmark (FrontierCode 2.0), Medium is Carbon's best effort level
I am referring to the chart that cross-compares multiple benchmarks, which Artificial Analysis clearly shows Carbon-high as superior on. Further, AA's FrontierCode chart places Carbon-high as having a higher score on Frontier 2.0 than Carbon-med (https://artificialanalysis.ai/evaluations/frontiercode-2-0?m...)
Where it gets very interesting is when you look at the Cognition "FrontierCode" publish date (2026.10.27), and the revision as of 10.28 in web archive (https://web.archive.org/web/20261028175237/https://cognition...). This revision has a VERY different Carbon effort / intelligence curve where high beats medium.
It is only with the latest revision, as of today, that all of a sudden Carbon's benchmark curve goes cone-shaped. Curious, indeed....
Isn't it against TOS for OpenAI to use their models in the development of competing models? So, any employee who does so would be opening themselves up to civil liability at AI scales. IANAL, obviously, so genuinely curious.
Maybe they contract out the work to other companies in other countries who promise not to access OpenAI and then sign up to do things they couldn't do without accessing OpenAI? Like independent promoter payola.
Him banging a Chinese intelligence asset was already reported in the news last month, and that story broke on a Thursday. What's new today is his apology letter/admission, and the board investigation. The timing of those announcements was up to Nvidia, so it being outside of trading hours is not at all suspicious.
There's some kind of a cognitive bias here similar to the base rate fallacy but not quite the same.
Yes, 40% is pretty high in the abstract. But a 60% chance of getting removed is insanely high compared to the base rate for this concrete case! Before this news broke, what were the odds of Huang not being a CEO of Nvidia a year from now for any reason. I'd definitely have taken the under at 1%, and the over maybe at 0.1%?
Like, what other risks were on the horizon? Him getting bored or tired? Not that man. A terminal illness that was diagnosed too late to treat? That happens very rarely for the ultra rich with their access to infinite high quality health care. A freak accident, e.g. airplane crash? Possible, but the rates are <<1%. Targeted violence? Unlikely, Nvidia is too far removed to be hated that strongly by anyone. The board firing him for some other reason? It'd have to be something that can't be papered over with hush money (e.g. any standard sex scandal almost certainly could; it's only the national security aspects that make it impossible here). A hostile takeover or similar that leads to him being ousted? Logistically and financially impossible.
Going from so rare it's not worth planning for to even odds is a pretty darn big shift from a single event!
> Why would the volume that any of this be increasing exponentially?
Because the limiting factor for software has always been the supply, not the demand. When the average human engineer would ship 100 lines of production code / week, you could only afford to do the most critical things. But if code gets cheaper to produce, the projects that were marginal become worth doing, and the projects that were obviously insane become marginal. And that's exactly what is happening as models get cheaper and better, leading to the frontier of worthwhile projects expanding rapidly.
> And what are all the engineers doing when (besides the model quality) their products don't seem to be changing very much?
That's got some nice But other than that, Mrs. Lincoln, how was the play? energy to it!
I’m going to assume people are asking for the benefits. Assuming that when there are so many commits it breaks git and CI, there would be a tangible thing that we can look and touch to say these 10 million PR in the last week brought you bla, cause 10 million PR are what it took to have say cloud functions or redis.
>That's got some nice But other than that, Mrs. Lincoln, how was the play? energy to it!
So you think that the move from 4o to Astra required something like 1000x the code to pull off? I find it hard to believe as an MLE, but since I don't work with LLMs I might be wrong about how complicated the implementations have become.
I do know people who work at OAI and it seems like none of them work directly on the models, but that's n=1.
Local officials granting tax breaks to data centers are just total morons. There is no polite way to say it. They have these companies over the barrel, any negotiation should be about how much extra money the data center shovels directly to the locals to make them enthusiastic about the DC, not about tax concessions.
But amazingly, they might not be the biggest morons of the story. This data center will make Google billions in profit per year. What beancounter is choosing to jeopardize that for avoiding $3M/year in taxes? How out of touch do you have to be to be to make that call, and why does that person still have a job?
Yes governments are all rotten and corrupt to the core. We should be castrating them and limiting their size and power as much as possible, everywhere.
No. The solution is a complete overhaul of government to actually represent the will of the people, not to remove the only means the public has over their lives and community.
> We should be castrating them and limiting their size and power as much as possible, everywhere.
You may have different intent but that is what your wrote. These representatives are the voice of the people, if you limit their power then Google can build datacenters without even needing to ask.
No. If models become better, the natural expectation is for hardware to get more valuable, not less. (The hardware that was previously able to achieve X can now do 2X. Why would that make the hardware worth less?)
Likewise if models become commoditized, it's raising the value of the existing hardware (you'd expect the hardware owner to keep a bigger slice of the surplus when that's the scarce component).
For model improvements to be bad for the value of hardware, what you'd need is for demand for intelligence to be inelastic. But everything we've seen so far suggests that is not the case: the demand for intelligence (whether human or artificial) is highly elastic both in price and in quality.
The growth doesn't need to be from Search. Even with the current level of search growth, Google's total revenue is up 25% year on year, operating income is up 30% year on year.
I don't know that I got anything out of this other than that it took a long time. I think there is an interesting article to be written on this. Like, what were the concrete requirements? Why were those the requirements? Maybe show how one letter evolved over time as the project went on, and explain why.
But instead the article is effectively the Gravitational Pull of Pepsi[0].
Hmm. I would have appreciated more depth and details, but it did cover requirements to look good and maintain identity across print, screen, merchandise, etc.
Which makes sense but that alone sent me spinning. Not easy.
But how many brands are really, honestly, displayed at that variety of sizes, in that variety of media, across language barriers and purposes. Some, sure. Not many.
That is not actually a typical playbook. Normal businesses do not want to be regulated, and will fight it tooth and nail. So regulation happens when bad things happen, the public wants change, and the businesses can't fight it or set favorable terms. (At most they can minimize the badness of the terms). Regulatory capture happens much later, and slowly. It's not where you start at.
> That is not actually a typical playbook. Normal businesses do not want to be regulated, and will fight it tooth and nail.
Sure. If you exclude, for example, the entire automotive industry, the entire telecom industry, the entire airline industry, the entire energy industry, and the entire healthcare industry from the category of "normal businesses" then you can perhaps make an argument that most "normal businesses" do not want to be regulated.
> Regulatory capture happens much later, and slowly.
Regulatory capture often happens quite early, especially when new technologies emerge, and companies that are successful early on aim to consolidate their hold on the market. This happened with the telephone industry around the turn of the 20th century, the broadcasting industry starting in the 1920s, etc. Regulation always tends toward market consolidation, so the people who are aiming for market consolidation tend to advocate for it. And sometimes the most effective advocacy is pretending to oppose it.
Like, did you write this in another language and use LLMs to translate? Write an outline and LLMs wrote out the rest? Write this amount of text that was then edited by LLMs?
reply