Dmitrii Troitskii

Engineering Fellow @NDIF_team | Postponing Robocalypse

[Dmitrii Troitskii @mitroitskii]

@abacaj This is exactly the problem we are aiming to address @ndif_team

[Jaden Fiotto-Kaufman @jadenfk23]

Frontier LLMs have capabilities that smaller AIs don't, but up to now there's been no way to crack them open.

Now that #Llama3 405b is here, what's the most interesting experiment YOU want to do?

🚀 Apply at https://t.co/Zo3ALEt8td to make it happen and read for details 🧵⬇️

[Jaden Fiotto-Kaufman @jadenfk23]

Frontier LLMs have capabilities that smaller AIs don't, but up to now there's been no way to crack them open.

Now that #Llama3 405b is here, what's the most interesting experiment YOU want to do?

🚀 Apply at https://t.co/Zo3ALEt8td to make it happen and read for details 🧵⬇️

[Dmitrii Troitskii @mitroitskii]

> .. these findings carry timely insights for the desirability and feasibility of outsourcing legally relevant tasks to AI models, as well as for the importance for AI developers to implement rigorous and transparent capabilities evaluations to help secure safe and trustworthy AI

[Dmitrii Troitskii @mitroitskii]

Seems like we’re in a new phase of GPT-4 hype-cycle – sobriety.

Re-evaluation of GPT-4’s performance on the Uniform Bar Exam shows much less impressive results than OpenAI’s initial marketing push

https://t.co/eBfcRvPxRe

[Dmitrii Troitskii @mitroitskii]

> … when examining only those who passed the exam (i.e. licensed or license-pending attorneys), GPT-4's performance is estimated to drop to ~48th percentile overall, and ~15th percentile on essays.

[Dmitrii Troitskii @mitroitskii]

Interesting debunking analysis of the new hot paper about gpt-4 ability to solve all MIT exams.

[Raunak @raunakdoesdev]

A recent work from @iddo claimed GPT4 can score 100% on MIT's EECS curriculum with the right prompting.

My friends and I were excited to read the analysis behind such a feat, but after digging deeper, what we found left us surprised and disappointed.

https://t.co/mpDqlenk04

🧵

[Dmitrii Troitskii @mitroitskii]

Amazing QnA with @ch402 Thank you to @davidbau for arranging this!

[Dmitrii Troitskii @mitroitskii]

Enermous thank you to @wesg52 for giving the opportunity to work on this project

[Wes Gurnee @wesg52]

This paper would not have been possible without my coauthors @NeelNanda5, Matthew Pauly, Katherine Harvey, @mitroitskii, and @dbertsim or all the foundational and inspirational work from @ch402, @boknilev, and many others!

Read the full paper: https://t.co/hZkFK6aI38

[Wes Gurnee @wesg52]

Neural nets are often thought of as feature extractors. But what features are neurons in LLMs actually extracting? In our new paper, we leverage sparse probing to find out https://t.co/hZkFK6aI38. A 🧵:

[Khoury College of Computer Sciences @KhouryCollege]

Spatial light modulators. Empathetic robots. Gig worker empowerment. In a recent showcase, Khoury College's research apprentices discussed not only their findings, but their futures, and the problems they're bent on solving. Read more below.
https://t.co/3qVkglE98X

[Dmitrii Troitskii @mitroitskii]

@nbouscal Do you see a world where SBFs behavior that signals sociopathy might instead signal autistic traits?

[Dmitrii Troitskii @mitroitskii]

@profoundlyyyy @SBF_FTX One specific dimension which, it seems, people have missed is the correlation between lust for power and an interest in changing the world. If you believe in the value of the longterm future what could be more powerful than to influence it?

[Dmitrii Troitskii @mitroitskii]

@profoundlyyyy Yeah, that’s the read I’ve got too. Also, I feel like there is too much “@SBF_FTX was X that’s why this happened” where X is utilitarian / sociopath / stupid. Peoples’ motivations are never that simplistic.

[Dmitrii Troitskii @mitroitskii]

@profoundlyyyy @SBF_FTX To iterate on this. I don’t know his intentions, and judging by many reports on Twitter and elsewhere, he might be incredibly overconfident, personally unpleasant and macchiavelian, but it doesn’t actually contradict him caring about EA causes on a personal level

[Dmitrii Troitskii @mitroitskii]

@JMannhart I’ve received a warning about this potentially being the case from an outside observer who closely followed crypto. They also speculated that FTXFF funds might have been acquired in a fraudulent way. Time will tell, - this might be a big painful lesson on trust.

[Dmitrii Troitskii @mitroitskii]

Gathering opinions around the EA twitter-sphere on the whole SBF-EA situation; Feels like a deep dive into a pool of rumors and hot takes. Is Twitter always like this?

[Cate Hall @catehall]

@JeffLadish This tweet feels like the right mix of inflammatory and vague to be uninformative, unfalsifiable, and useful as ammo for anyone wanting to blanket condemn EAs :/

[Geoff Anders @geoffanders]

It's premature, and indicates bias, when people rush to exonerate before all the evidence is in.

We're still getting to the "who knew what, when" part of the scandal.

[Jeffrey Ladish @JeffLadish]

A few thoughts on SBF, FTX, and Effective Altruism:

1) Until the current crisis, I think it was reasonable for EAs to look up to SBF and think FTX was hugely net good. SBF intentionally making a huge fortune to then donate to EA causes was ambitious and awesome...

[Max Alexander @absurdlymax]

"Wait we're funding constrained?"

"Always have been."