Pinned
I have excerpted the panel discussion from our NeurIPS Tutorial on the Science of Benchmarking to YT! Check it out!
Feat. @WendaXu2 @enfleisig @idavidrein @sainingxie @OfirPress
Michael Saxon
3,650 posts
Michael Saxon
@m2saxon
opinions my own
Research scientist @GoogleDeepMind, Prev: postdoc @uw PhD @ucsbNLP BSMS @asu.
Works on multilinguality, multimodality, and how to evaluate
- The viral new "Definition of AGI" paper has fake citations which do not exist. And it specifically TELLS you to read them! Proof: different articles present at the specified journal/volume/page number, and their titles exist nowhere on any searchable repository.
- What an amazing opportunity for the most well-resourced private school students and children of researchers :DNeurIPS 2024 will have a track for papers from high schoolers.
- If you feel the need you make the "I'm not stereotyping" disclaimer, maybe you should reconsider what you're saying She could have saved a lot of words by deleting "Chinese" and "NOTE: I am not a racist but blah blah blah blah"Using racial labels to describe misconduct is harmful and inappropriate. @NeurIPSConf must not condone speech that targets specific ethnic groups. We urge Rosalind Picard @MIT @medialab to retract and apologize for her statement. Btw, most Rosalinds I know are honest and morally
- If you are a junior phd student, and you like another student's paper, email them! They will almost certainly want to talk about it. Schedule a call! Chat about ideas! Legit have made friends this way
- Replying to @PandaAshwineeWhat's unethical about this? No legitimate review will be affected. Is it just because it confers an unfair advantage on papers that use it?
- 2023 research goals: - Submit to every *ACL and one of {NeurIPS, ICLR} (not counting resubmissions) - Read >=1 paper a day - Usefully annotate 3 papers a week - Sort notes in Obsidian weekly - Check my Semantic Scholar recommended every day - Write a useful blog post monthly
- Replying to @m2saxonDifferent papers by the same authors from the same years do exist. If I had to guess these AI safety researchers used an LM to speed up the reference writing process after reading these real references, and it hallucinated. watch out for misaligned writing assistants!
- Unfortunately I will no longer be able to tell an email is spam because it called me "Dr. Saxon" anymore
- This is what happens when a significant contingent of people studying LLMs don't meaningfully engage with *ACL literature. This is why we need policies that don't drive out ECRs by disadvantaging their ability to preprint and publicize against ICLR/CVPR/NeurIPS-primary ECRs.Do language models have an internal world model? A sense of time? At multiple spatiotemporal scales? In a new paper with @tegmark we provide evidence that they do by finding a literal map of the world inside the activations of Llama-2!
GIF - Super excited that I'll be joining the @UW @TechPolicyLab and @uwnlp as a postdoc in the Fall working with @aylin_cim to continue my research directions in situated evaluation, multimodal/lingual GenAI, and start exploring new directions in safety and alignment! Open to collabs๐
- ๐จ๐ฑObligatory job market announcement postโผ๏ธ๐คฏ I'm searching for faculty positions/postdocs in multimodal/multilingual NLP and generative AI! I'll be at #NeurIPS2024 presenting our work on meta-evaluation for text-to-image faithfulness! Let's chat! Website in bio, papers in๐งต
- Tbc, the reason why pushing high school students to write papers is EXTREMELY bad (and even undergrads to a lesser extent) is that it inevitably creates a new bean-counting rat race, where there's tremendous information asymmetry between the coached and the non-coached!
