I wonder about the world where, instead of investing in AI, everyone invested in API.
Like, surfacing APIs, fostering interoperability... I don't want an AI agent, but I might be interested in an agent operating with fixed rules, and with a limited set of capabilities.
Instead we're trying to train systems to move a mouse in a browser and praying it doesn't accidentally send 60 pairs of shoes to a random address in Topeka.
I was hoping to see the questions (which I can probably find online), but also the answers from models and the judge's scores! Am I missing a link? Without that I can't tell whether I should be impressed or not.
Looking at the replies to, say, https://x.com/aoc, it seems pretty steered to me.
If you want to hear replies from her supporters, good luck getting through a few hundred blue check marks. Compare this to something like https://x.com/RonDeSantis, which is an full of adulation for the guy.
In my estimation, X is tilted pretty far right at this point, simply because paid blue check marks are a sign of pride or shame, based on your political affiliation.
That's what I find most offensive about the use of LLMs in education: it can readily produce something in the shape of a logical argument, without actually being correct.
I'm worried that a generation might learn that that's good enough.
"Speak confidently as if you are right, but listen carefully as if you are wrong."
I hate-hate-hate this advice. When I assert something confidently, I'm willing to put money on it.
The first time I ran into someone I respected who was confidently wrong, it threw me for a loop. It makes you seem like a blowhard bullshitter! Guessing is fine, but say it's a guess.
Edited to add: I suppose this is a variant of Cunningham's Law, but I still hate it.