Neekon Vafa

PhD student @MIT | Ex-Software Engineer @Google (@YouTube) | Mathematics + CS @Harvard '19 | 🥌

[Keyon Vafa @keyonV]

Thrilled to share I'll be joining UC Berkeley next year as an assistant professor in @UCBStatistics and affiliated with @Berkeley_EECS.

My research will build methods to test/improve the implicit world models of AI systems, so that they reflect reality and human understanding.

[Seyoon Ragavan @seyoonragavan]

Can you encrypt a program to hide its inner workings while still enabling others to use it? This problem of "program obfuscation" is central to cryptography.

To learn more, see this article! It also touches on a paper by myself, @NeekonV, and @Vinod_MIT from TCC 2024.

[Keyon Vafa @keyonV]

Can an AI model predict perfectly and still have a terrible world model?

What would that even mean?

Our new ICML paper formalizes these questions

One result tells the story: A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws 🧵

[Keyon Vafa @keyonV]

100+ filmmakers and celebrities voted for their favorite movies of the 21st century in the NYT poll.

Whose tastes are most similar to yours?

I made a website that lets you find out: Pick your top 10 movies and see your closest matches.

[Keyon Vafa @keyonV]

Can you guess which of these two questions an LLM needs more thinking tokens to get right? 🧵

[Keyon Vafa @keyonV]

New paper: How can you tell if a transformer has the right world model?

We trained a transformer to predict directions for NYC taxi rides. The model was good. It could find shortest paths between new points

But had it built a map of NYC? We reconstructed its map and found this:

[Keyon Vafa @keyonV]

Belated updates:

1- I defended my PhD in May! Big thank you to my committee and especially to my advisor, @blei_lab.

2- I'm excited to start a postdoc at Harvard! Very grateful to @harvard_data for the opportunity.

[Keyon Vafa @keyonV]

1/ I had the opportunity to join the @DataSkeptic podcast to discuss CAREER, a method that uses transfer learning to model career trajectories. This is a project I've been working on with @blei_lab and @Susan_Athey's group. Thank you for having me on!

https://t.co/K7h3GdRp20

[IACR @IACR_News]

#ePrint MacORAMa: Optimal Oblivious RAM with Integrity: S Mathialagan, N Vafa https://t.co/Q1cNvKrduK

[Keyon Vafa @keyonV]

Excited to speak this Saturday at the #NeurIPS2022 Workshop on Distribution Shifts about CAREER, a transformer-based method for modeling career trajectories from labor sequence data.

https://t.co/zUQnv6TzKH

w/ @Susan_Athey, @blei_lab, @EmilPalikot, @TianyuDu7, @AyushKanodia

[IACR @IACR_News]

#ePrint Continuous LWE is as Hard as LWE & Applications to Learning Gaussian Mixtures: A Gupte, N Vafa, V Vaikuntanathan https://t.co/UhBToU35Ro

[TCS blog aggregator @cstheory]

TR21-166 | Average-case Hardness of NP and PH from Worst-case Fine-grained Assumptions |

Lijie Chen,

Shuichi Hirahara,

Neekon Vafa https://t.co/XRaRd4ubw8

[Keyon Vafa @keyonV]

New paper: https://t.co/8CjhtqB1j8

Consider a sequence generated by a language model. Which words were most important for generating each word?

We propose greedy rationalization: greedily finding the smallest subset of words that would make the same prediction as the full text.

[Lex Fridman @lexfridman]

Here's my conversation with Cumrun Vafa (@cumrunv), a theoretical physicist at Harvard specializing in string theory. He is the co-recipient of the Breakthrough Prize in Fundamental Physics, the most lucrative academic prize in the world. https://t.co/1hGjqBeA85

[Cumrun Vafa @CumrunV]

I’m excited to share my new book, Puzzles to Unravel the Universe, based on a freshman seminar I teach at Harvard. It includes over a hundred puzzles and discussion on how they relate to deep ideas in physics and math.

Available in paperback and Kindle:
https://t.co/DajQjvFgvN

[Neekon Vafa @NeekonV]

Thank you! It's been an honor being part of the club over the past 3 years, where I first learned to curl, and I'm excited to see all that the club can accomplish under the new leadership. Good curling! 🥌🥌🥌

[Harvard Curling @harvardcurling]

Congratulations to our graduating senior @NeekonV! 🎉🎓

[Harvard Curling @harvardcurling]

We've been profiled by @thecrimson! #roadtonationals https://t.co/BT9VnOAIbh

[raph @__embed]

Just had a stunning realization. The SAT went from being out of 2400 back to being out of 1600. That means 3SAT reduces to 2SAT and P=NP. QED

[Lance Fortnow @fortnow]

Bill does his poll again to determine those who say P ≠ NP and those who are delusional. https://t.co/kaZQA53mmZ

[Neekon Vafa @NeekonV]

Spoilers: on the first throw, scissors is significantly better than uniformly random...because people really like paper? https://t.co/Up6Gusj2tK

[FiveThirtyEight @FiveThirtyEight]

The Riddler’s Inaugural Rock-Paper-Scissors Tournament https://t.co/BsUuluDy6I