6 comments

[ 0.22 ms ] story [ 14.1 ms ] thread
dont need to freeze em, its already peaked

with nothing left to take , its inevitable, and could even regress after consuming its own slop

stop feeding the hype, let this bubble crash. i await an entire market correction around ai(.com)

those regulators are just grasping for power

People said that kind or thing continuously since GPT-4.

GPT-4 passed a lot of exams, but couldn't find and exploit novel zero-day issues to get out of its sandbox.

thats just a hype story, to keep the bubble going

ye they improve in coding a bit, but theres also alot of shitty security holes to find.. so its not really news

> thats just a hype story, to keep the bubble going

Not "just" a hype story, the events were discovered by outside parties, and subsequent open weights models replicate the capability. The early models simply could not do that, new models can.

Or just look at CVE publications. They've more doubled year-on-year. And even "it doubled" is understating it, because while it went from 128/day averaged over Nov 2024 to Oct 2025, to 267/day averaged over Nov 2025 to Oct 2026, it was about 548/day in specifically Oct 2026.

its totally a hype story right from the start.

the bot was instructed todo the things it did, its not rogue, and didnt break out of a super secure prison

its like saying , u told ur dog to fetch , and it tore thru and out of, a paper house , and fetched the ball

the cve thing, without more context, could be attributed to ai slop coding, or just population growth, like every other volume statistic on earth. more ppl, more bad coders, more cve. Yes ai contributes to locating low hanging fruit on code noone ever probably bothered to check in the first place. not really a win for ai, just a lose for human slop.

> the bot was instructed todo the things it did, its not rogue,

Incorrect, but irrelevant to my point.

Incorrect, because HuggingFace was not the target of the challenge.

Irrelevant, because models before this point were not capable of such things.

> and didnt break out of a super secure prison

And yet, it did find multiple novel zero days, at least one on the OpenAI and two on the HuggingFace side. And in

> the cve thing, without more context, could be attributed to ai slop coding,

Newer models (and again: you can verify this yourself with the newer open weight models which do not come from the OpenAI or Anthropic) find novel exploits in major open source repositories, and also in projects that refuse to accept patches of these findings because they hate AI so much.

> or just population growth

Population growth? Seriously? Even if literally every fertile woman on the planet got pregnant at the same time, the world population wouldn't rise that fast.

> Yes ai contributes to locating low hanging fruit on code noone ever probably bothered to check in the first place.

Firefox, the Linux Kernel, FreeBSD, Apache, and SQLite (amongst others) are your idea of "locating low hanging fruit on code noone ever probably bothered to check in the first place"?

You need to re-evaluate your understanding of the world.