This is a solid intro.
I had no idea Lean was so Haskell-y and readable/practical in this way! For some reason I assumed it was just a clunky abstruse prover IDE type of thing, I guess I conflated it with older coq/Rocq and all the gigantic illegible ai math papers in recent news
Oh wow, throwback!
Pretty remarkable how looking thru these janky old images actually did give me a sort of fuzzy nostalgic feeling. The tells/flaws are such a recognizable style.
I remember absolutely abusing the DALLE endpoints when we first got access after the waitlist ended. This was in my old job, where work looked a lot different from today. Also just a vastly different time in my life, funny how much time flies in four years.
(I do wish this had a before/after sort of comparison to demonstrate newer sota models handling the same prompts)
Ah this is a great post, thanks for sharing this. I completely agree with you!
Fun anecdote related to the first point about AI posters etc: my non-technical mother and her group of elderly Mahjong friends, who don’t care about or are completely ambivalent about AI, ate at a small restaurant in their town recently, and they were confused and annoyed by the new menu that was one of those maximalist and grotesque one-shot cheap model designs. Now it’s spoiled for her and she notices the awful styles everywhere in person and online. So I do think the tide there must be turning soon, at least against that minimal effort slop ad/menu copy
I love the Eno quote, I think I’ll check out that diary
They do have Anthropic and OpenAi as vendors, who are at the center of this hack as well, and who have been beating the "distillation attack" drum on a regular basis.
OpenCode, and the lot, will need to release an open, public, dataset containing the original queries (and other relevant details) that the hacked keys were used for. I'm unwilling to trust those industry players at their word. We need to do our own community analyses.
We don't yet know what the hacked api keys were used for. We would need grand transparency, like an open dataset with the original queries, so that the public can validate through a collaborative effort. I'm not keen to accept their early conclusions, eventual RCA summary stats, or METR reports.
If OpenCode is not releasing all of the details about the exposure at the moment, it likely would be a good idea to rotation keys for the connected services at this point
reply