Florian Brand

evals @PrimeIntellect | open models @interconnectsai

[Florian Brand @xeophon]

I fell for it, damn it.

Finally moved to SF ❤️

[Florian Brand @xeophon]

I won’t fall for the SF psy op

I won’t fall for the SF psy op

I won’t fall for the SF psy op

I won’t fall for the SF psy op

I won’t fall for the SF psy op

[Florian Brand @xeophon]

sitting between two data vendors

guy on the left wants to sell access to a bench
guy on the right wants to sell data to hill climb said bench

[Florian Brand @xeophon]

@ryan_marten lfg!!

[Florian Brand @xeophon]

@local0ptimist @kaggle Let’s do it!

[Florian Brand @xeophon]

I am at the @kaggle bench summit if someone wants to rant about their love/hate relationship with their favorite bench

[Florian Brand @xeophon]

@vitaliychiley Congrats!

[Florian Brand @xeophon]

@llm_wizard @kaggle I am just here to meet fun people!

[Florian Brand @xeophon]

@vincejhuang > Port a codebase/executable into another language

- NL2Repo
- MirrorCode
- FrontierSWE
- ProgramBench

[Florian Brand @xeophon]

today, it is split cleanly by those who don't comment on data they cannot look at and those who do

(frontiercode 2.0)

[Florian Brand @xeophon]

tl is split cleanly by those who look at data and those who don't today

[Florian Brand @xeophon]

@S1r1u5_ By the time we have 5, we should need even bigger brain ai, cause all the stuff that only needs big brain ai is fixed already

[wispem-wantex @wispem_wantex]

Has anyone considered using 4️⃣ as their private key? In unicode it's 7 bytes (56 bits).

This feels like a fair compromise between the two main stakeholder positions, "Private keys should be at least a little bit secure" vs "I want my private key to be 4"

[Florian Brand @xeophon]

@1thousandfaces_ glm 4.7 also loves its tables if you wanna get back to o3's distant cousin

[Florian Brand @xeophon]

@creet_z there are rumors that astra can do the whole shebang in ~30 mins

[Florian Brand @xeophon]

@llm_wizard i don't smoke cigs

[Florian Brand @xeophon]

very interesting how sf went from biohacking and longevity to smoking in ~6 months

[Florian Brand @xeophon]

@basedjensen light weight

[Florian Brand @xeophon]

@wenhaocha1 @AngeloLorenzoni And IRT is also used in a reverse way — selecting tasks from a bigger group to gain the same signal to save costs

A lot of the papers perform as good as a random subset of the same size 😵‍💫

[Florian Brand @xeophon]

@AngeloLorenzoni @wenhaocha1 Basically bigger error bars, people are obsessed with the final number and don’t realize it has huge CIs

[Florian Brand @xeophon]

@xdotli @benchflow_ai 👀

[Florian Brand @xeophon]

@lateinteraction oof, thanks dad

[Florian Brand @xeophon]

the more someone mentions the bitter lesson, the less i take them seriously