No, and there’s no reason to think cars will stop improving either, but that doesn’t mean they will start flying.
The first error is in thinking that AI is converging towards a human brain. To treat this as a null hypothesis is incongruent both wrt the functional differences between the two and crucially empirical observations of the current trajectory of LLMs. We have seen rapid increases in ability, yes, but those abilities are very asymmetrical by domain. Pattern matching and shitposting? Absolutely crushing humans already. Novel conceptual ideas and consistency checked reasoning? Not so much, eg all that hype around PhD-level novel math problems died down as quickly as it had been manufactured. If they were converging on human brain function, why this vastly uneven ability increases?
The second error is to assume a superlinear ability improvement when the data has more or less run out and has to be slowly replenished over time, while avoiding the AI pollution in public sources. It’s like assuming oil will accelerate if it had run out and we needed to wait for more bio-matter to decompose for every new drop of crude. Can we improve engine design and make ICEs more efficient? Yes, but it’s a diminishing returns game. The scaling hypothesis was not exponential but sigmoid, which is in line with most paradigm shifts and novel discoveries.
> Making these sort of blanket assessments of AI, as if it were a singular, static phenomena is bad thinking.
I agree, but do you agree with yourself here? Ie:
> no reason to think that these tools won't vastly outperform us in the very near future
.. so back to single axis again? How is this different from saying calculators outperform humans?
Where can I go to see language model shitposters that are better than human shitposters?
(Always remember to use eye protection.)
But I think that everyone is lossing trust not because there is no potential that LLMs could write good code or not, it's the trust to the user who uses LLMs to uncontrollable-ly generate those patches without any knowledge, fact checks, and verifications. (many of them may not even know how to test it.)
In another word, while LLMs is potentially capable of being a good SWE, but the human behind it right now, is spamming, and doing non-sense works, and let the unpaid open source maintainers to review and feedback them (most of the time, manually).
Making these sort of blanket assessments of AI, as if it were a singular, static phenomena is bad thinking. You can say things like "AI Code bad!" about a particular model, or a particular model used in a particular context, and make sense. You cannot make generalized statements about LLMs as if they are uniform in their flaws and failure modes.
They're as bad now as they're ever going to be again, and they're getting better faster, at a rate outpacing the expectations and predictions of all the experts.
The best experts in the world, working on these systems, have a nearly universal sentiment of "holy shit" when working on and building better AI - we should probably pay attention to what they're seeing and saying.
There's a huge swathe of performance gains to be made in fixing awful human code. There's a ton of low hanging fruit to be gotten by doing repetitive and tedious stuff humans won't or can't do. Those two things mean at least 20 or more years of impressive utility from AI code can be had.
Things are just going to get faster, and weirder, and weirder faster.