Aidan Vyas

Empirical asset pricing and artificial intelligence. [email protected]

[Aidan Vyas @aidanvyas]

Had a blast playing the @FiveThirtyEight NFL forecasting game this season! @NateSilver538 @friscojosh @Neil_Paine @greerreNFL

[Aidan Vyas @aidanvyas]

@astropol0 This seems like such a chart crime?

[Aidan Vyas @aidanvyas]

@Thackej @calebwatney I guess I was specifically thinking about the case of looking at an order book. Beyond basic heuristics, no human could look at an order book and make predictions on futures prices, though quasi-LLMs can do this quite succesfully.

[Aidan Vyas @aidanvyas]

@JoePostingg You could also imagine a world where the “best” generals are just the luckiest.

I.e., they are perceived to be great because they were lucky.

[Aidan Vyas @aidanvyas]

@alexolegimas I’ve never tried the 99 cent pizza, but it’s on my bucket list (though I have low expectations). I haven’t been impressed with the $1.50 slices so far, though perhaps I’m a bit of a pizza snob.

[Aidan Vyas @aidanvyas]

@rand_longevity I believe this is just Google crawling their page. It’s been updated pretty regularly the past week.

[Aidan Vyas @aidanvyas]

@YudaB11 @joaogsantosc @AnthropicAI At first I thought there was a chance that I would get one too, but he got an email, which I haven’t received, so it looks unlikely.

[Aidan Vyas @aidanvyas]

@alliekmiller In one of the commit messages, it claimed to be:

Co-Authored-By: Claude Fable 5 <[email protected]>"

even though the model was clearly set to Opus 4.8.

[Aidan Vyas @aidanvyas]

@Eldergenix I've been checking for the past week, and every time I check, it always says the site has been updated in the last <2 hours, so I don't think these are actual updates.

[Aidan Vyas @aidanvyas]

@redtachyon They seem to be subsidized for the most active users, @SemiAnalysis_ did a good write up on this recently.

[Aidan Vyas @aidanvyas]

@Rolando_Rojas1 @synthwavedd @OpenAI This is basically the same thing, where the newer generation of the smaller model, Flash 3.5, is better than the older, bigger model, Pro 3.1.

[Aidan Vyas @aidanvyas]

@Rolando_Rojas1 @synthwavedd @OpenAI Sonnet was better than Opus when Sonnet 3.5 was out, and the most recent Opus model was Opus 3, and then when we had Sonnet 4.5 vs. Opus 4.1, but at the same level (i.e., 5 vs. 5), Opus and Fable should be better than Sonnet.

[Aidan Vyas @aidanvyas]

@Rolando_Rojas1 @synthwavedd @OpenAI I think they have 4 classes of models, Haiku, Sonnet, Opus, and now, Fable (Mythos and Fable are essentially the same, except that Fable defaults to Opus on some sensitive queries).

[Aidan Vyas @aidanvyas]

@xeophon Why does the score need to be revised down? If anything, on those 8% of samples, you would expect Fable 5 to do better than Opus 4.8, raising the overall score.

[Aidan Vyas @aidanvyas]

@SpecialPuppy1 @Brendan_Duke I feel like you have to go back to FDR's rural electrification for a super obvious example.

[Aidan Vyas @aidanvyas]

@dwarkesh_sp Writing is the classic example.

Comedy is a better one.

[Aidan Vyas @aidanvyas]

@dlwiest There’s better security with the enterprise plans.

I also think they’ve largely stopped selling the fixed monthly fixed price plans to enterprise, so to get the good Claude Code deals, it would have to be a personal account.

[Aidan Vyas @aidanvyas]

@theblockspot I think Kareem is the consensus?