Gotta be honest, LLMs are still subpar compared to what I do.

Can they produce more code more quickly? For sure.

Can they go deeper on a topic instantly? Definitely.

Would I say their output is better? Almost certainly not.

Maybe that's true, and maybe you're an outlier. But from what we've been seeing, the frontier LLMs have torn apart many codebases, finding bugs littered everywhere, some over a decade old.

At this point it seems impossible for human to write better/more efficient code than a swarm of agents -- to be clear, multiple coordinated LLMs, not one-shot builds with a single agent.

▲

0c3ca83 2 hours ago | parent [-]

How much money was put into this, and what would putting a similar amount of money into fuzzing, static analysis, and human researchers look like?

	▲	adventured an hour ago \| parent [-]
		Ten median US software developers cost about $1.5 - $1.7 million per year - including all benefits - as of a couple of years ago. A frontier AI is at least as capable as ten median human software developers. It's likely well beyond that now. Over the course of a year, would you have to spend $1.5 million to get the output of those ten median humans? No chance in hell. You'd have to spend less than 10% of that. Then on top of that: the AI is improving rapidly; the humans go on vacation, the humans decline, the humans get old, the humans get sick, the humans take parental leave, the human struggle to adapt, the humans are resistant to change, the humans riot, the humans protest, the humans need healthcare, the humans want paid more, the humans can only work for 4-6 hours at peak while writing software (slowly). Do you know any humans that are experts at a dozen programming languages? That human does not exist. The frontier AI are omnimaths.

▲

nonethewiser 2 hours ago | parent | prev [-]

>Gotta be honest, LLMs are still subpar compared to what I do.

Im sorry... what?

You are replying to this:

> Our first formalization of Fermat’s Last Theorem, performed by an unreleased research version of Claude, took 11 days. It relied on a purpose-built harness along with guidance from a human researcher. By contrast, a team of 1,000 Fable 5.2 agents fully formalized the theorem in just 30 hours, with minimal human intervention. Fable 5.2 started from minimal prompts, built its own ways to coordinate, and used no mathematical references, only a list of intermediate theorem statements to score progress. It needed one-third of the action tokens as our first formalization and produced a proof seven times shorter. As can be seen in the figure below, Opus 5.5 (with the same 1,000-agent setup) completed only 26% of the formalization in 18 hours, and barely made any more progress after 30.

	▲	adventured an hour ago \| parent [-]
		HN has become so miserable over the past ~12-24 months because it's a site filled full of highly paid techies living in the deepest pit of denial that has ever existed. It's all going away. Here's a prediction: HN will effectively die with the software developer.