6,608 karma · joined February 14, 2015
Anyhow, my research summary: Individual humans are so fucking expensive to train and upkeep (and this includes everything from before womb, where another human already limits their ability to work) You retain ~zero knowledge after death and start all over again for another measly 15 years of effective, productive work. Model training/r&d in relation, when deployed and used at scale, rounds to zero, even with the current retraining regime.
*Of course, the ratio can go to negative infinite if one assumes that models are doing 0 useful work currently and never will
On the margins, suppose the prompt is literally: "Build a feature complete, high polish Facebook clone". Facebook is complex but likely not super complicated tech, and still I would assume that (after having burned through a substantial amount of tokens) you would find substantial enough differences in the outcomes between different models on that prompt on various fronts.
The above ask is obviously not useful, but what's preventing you from taking on bigger chunks until you approach the limit? At some point you would hit a boundary, where the diff will be obvious.
But also, bigger projects need some amount of loc written and it's a bit silly to pretend that this is not the case or a bad thing.
So the answer to the question is roughly: Establishing that an agent can work in a large-ish code base is valuable, because 1) them not being able to do so has been a critique and 2) it's something that is required for a lot of software projects.
I don't understand. Guilty until proven innocent, because they... are too successful? What could possibly be the generalizable idea here?
Should we have a speed limit for too successful companies, even if they might be doing super valuable work? Who would we trust to be the judge of the potential havoc that bad capital allocation in such a moment might cause?
EDIT: To be more clear, I don't have any particular qualms with the S&P committee maintaining it's position. That part I find mostly interesting and goes towards the second paragraph.
The first one is reserved for the quote, which I do have qualms with. "Nobody knows" feels a bit weak when the implication, that someone could be doing something illegal, turns into a guiding principle.
More people can however work on a great plan to attack it, which only has to work once, to be worth it.
I don't think that follows. They expect things to improve, they do something about it and might (unreasonably) be frustrated by what they think is a policy that stands in the way of quicker progress. The first part is certain, the second part less so, and the third is just speculation.
It's clear that open source project are struggling to understand what is going on and coming up with plans – like everyone else – but clearly there are better and worse ways to proceed in this new world, if popularity, adoption and progress are things you want to focus on.
"Why would you not tell us how to get to the secret dimension if it's so great?" It's just a dumb question.
"If bank vaults are so great, why would you not advertise their locations". It's a mystery, is what it is.
Them not having your best interests at heart does not make them an adversary. Just like you not having the best interests of your grocer or work place or most other humans on the planet at heart does not make you an adversary.
> Or perhaps: Satisfaction in our work isn’t created by the boss. It’s what’s left if they don’t ruin it.
~no boss sets out to ruin employee satisfaction. It's a byproduct of having to integrate more realities into the smaller scope that employees usually care about, and that is just not easy.
Of course, most bosses are also not great at this – on average bosses are, like everything else, just average – but to assume that bosses "ruin" satisfaction and employees would be longterm fulfilled and create working companies if only left alone is polemic.
Are our institutions this incapable?