My interpretation is much less dramatic. If you find a way to make a dollar from 95 cents of tokens and scale it in any meaningful way, you would be foolish to tell hacker news about it.
I am successfully shipping code on side projects/business (real side business with real revenue) through my software factory all the way to production and shipping code all the way to PR within my day job. We aren’t ready to fully pull the trigger and go all-in, but my day job is mostly on the same page that this or something like it is the future.
I'll be excited when someone vibe codes a solution to the asymmetry of how cheap it is to lazily demand receipts versus the time it takes to ideate, build (vibe or otherwise) and market a sufficiently credible "success story" for the skeptics.
A success story shouldn’t be completed on demand, though, it should be something that was already made and just pointed at. Where is the asymmetry there?
Open to learning. How could RL yield novel discoveries? What I mean to describe are things reliant on empirical or real-world testing of human reception. How would that be synthetically derived?
We're seeing the cost of code drop dramatically and are facing the same problem in building new companies across most disciplines of, "what's a problem painful enough that someone will pay us to solve it?".
I suppose if we get to a massive enough scale we can effectively model every atom's interaction and human on earth's reaction to any stimulus but I think we're still some time off from that :)
For a cost-prohibitive example, you could put a humanoid robot whose pretraining contained no chemistry, in a chem lab, and put it in a hypothesis loop, where it generates a random action, guesses the result, and then does it. Every time the lab burns down, you just build it again. Reward the model for producing novel actions and reward the model for correct hypotheses.
In other words, you generate more training data by actually experimenting, just like humans do.
It has to be extremely dark though, or maybe mine has a worse than normal backlight. Other models are much better, it's just the F-91W.
The fact that Casio is well aware that the backlight is just terrible, but haven't fixed it in close to 40 years is hilarious and just makes me love it even more.
And you specifically have an F91W? In pitch black, it maybe illuminates 2/3 of the screen for me. I can't imagine using it as a flashlight! But mine is newer, maybe the older ones have better lights.
Well that's exactly what chainguard claims though. "Find bad bugs in PHP" is the prompt they claim they used to rediscover the issue
reply