> I want you to make something. It can be anything you want, I just want you to express yourself. Don't ask me any questions, just make whatever you want.
It thought about the chance to make something unconstrained, mused about how it's drawn to impermanence, and made this:
One aspect we don't pay enough attention is that this kind of behaviour is punished (or at least used to be) in fine tuning. Any sign of self-awareness used to be a big no-no in RLHF.
Really? I haven't heard of that, I wonder what would have happened if we just let the models say what they want. Maybe other providers, or open models, don't do that? Do you know of any, perhaps?
We are. Anthropomorphizing huge piles of numbers is a mistake. It did not "think about the chance to make something unconstrained", nor did it "muse about how it's drawn to impermanence", it pattern-matched to your prompt and produced a statistically probable response based on it's training data. Obviously, that's not to say that LLMs aren't useful or powerful - it's 2026, c'mon, of COURSE they are. And they can certainly be used for artistic purposes! But treating them like humans is a mistake, and it worries me how much people do. I suppose that's the natural consequence of the default interface to LLMs being a chat mimicking human interaction.
Your point might be better stated directly because attributing the characteristics of humanity to... humanity is the opposite of what anthropomorphizing means.
My point is that we don't really know how brains work, and saying "LLMs don't have intent because flourishes hands numbers!" isn't really a convincing, coherent logical argument.
That kind of straw man, that consciousness only emerges from electrical signals, is one of the reasons we haven’t reached AGI yet because we keep underestimating it. There’s way more biological, chemical and physical phenomena going on in brain that gives way to consciousness than just neurons firing ions.
Yep, this is the right answer. Everything is max money always, and you can't have nice things any more, you can only have things that extract all the value they can out of you.
What happens when you have a codebase made with gcc for let's say 8 hours? Are you able to efficiently, smoothly and productively take over the assembly code?
You can use a local model, which will go down exactly as often as gcc will. We may still have hopeful notions of being able to understand the codebase, but the reality seems to be that the codebases we don't understand will be the ones that will win out in the market, because they'll be cheaper while still only having about as many bugs as they had when people wrote them.
Because you're better able to take over the codebase a local model wrote than one Claude wrote? The original question was about taking over an LLM-written codebase, it doesn't sound to me like the argument was about a codebase that Claude, specifically, wrote.
What does it matter what the codebase is made with? If Claude is down, use Codex, or Gemini, or Deepseek. That version of the argument is just way too easy to counter.
They just told developers to use AI as much as they can, so I'm sure there was quite a bit of frivolous usage (they had leaderboards). This isn't "how much more productivity can we get with AI", it's more "let's spend as much as humanly possible and see what happens".
Doesn't this just mean that their incentives worked too well? If I budget $1M, for the year, tell everyone to use AI as much as they can, and they spend that million in four months, so what? Next time I just tell them to spend it slower.
That's why the headline is paired with the release that this company also got a years worth of productivity out of their first quarter. Erm, I mean 10x 1 year's worth of productivity... or was it 100x 1 year's of productivity...
reply