Really cool idea for speeding up LLM inference by a lot! Auto regressive doesn't have to be a barrier anymore 🙂

Felix Draxler@drrelax.bsky.social · 4mo ago

LLMs are autoregressive and slow? No! Parallel Token Prediction decodes multiple consistent tokens in one model call. PTP allows arbitrary dependencies in one call, unlike discrete diffusion. Practical: 2.4x speedup github.com/mandt-lab/ptp ICLR: Apr 23, morning poster P3-#608

LLMs giveth and LLM taketh away

Published some notes on Microsoft's phi4-reasoning model, an 11GB download (via Ollama) which may well overthink things... it produced 56 sentences of reasoning output in response to my prompt of "hi" simonwillison.net/2025/May/6/p...

meet Chris J Li - this titan of thought has single-handedly conquered the fields of machine learning, optimization, statistics, reinforcement learning, and federated learning. he's not the visionary we want, but judging by the current state of affairs, he may be the one we deserve

Defended 🎉🎓 Big thanks to @royf.org, @sameer-singh.bsky.social, and labmates for their mentorship and support over the past 5 years!

How do LLMs interpret expressions of linguistic uncertainty such as "highly unlikely"? Short answer: pretty well .... unless they have relevant prior knowledge. Details in our EMNLP paper aclanthology.org/2024.emnlp-m... (with Kat Belem, Markelle Kelly, Mark Steyvers, @sameer-singh.bsky.social).

Turned @dippedrusk.bsky.social 's amazing Vancouver list dippedrusk.com/posts/2024-0... into Google Maps pins: maps.app.goo.gl/nGBbcUAMMixC...

Vagrant's Vancouver | Vagrant Gautam

A non-comprehensive list of places to go and things to do in the Greater Vancouver Area as curated by yours truly over 6 years. Might be outdated so please double-check!

dippedrusk.com

Excited about #NeurIPS2024, my 15th one I think! Eager to meet everyone & hear abt your work! But if you want to hear me, there's an exciting panel tonight lu.ma/v7oohp0u Also SpiffyAI is hiring ML engineers & UCI CS is hiring AI faculty, pls reach out to chat! 🧵

​From Research to Commercialization: A Fireside Chat with Senior AI Leaders · Luma

From Research to Commercialization Join us for a conversation with speakers who made the leap from top research institutions to industry and are shaping how…

lu.ma

If you use SHAP, LIME or Data Shapley, you might be interested in our new #neurips2024 paper. We introduce stochastic amortization to speed up feature + data attribution by 10x-100x 🚀 #XML Surprisingly we can "learn to attribute" cheaply from noisy explanations! arxiv.org/abs/2401.15866

Read only the first 1-2 sentences of each and go with your gut. You'll likely get the perfect score! Kind of thing where I probably prefer an unaligned model output to an aligned one..

Matt Clancy@mattsclancy.bsky.social · 2y ago

I made a Google quiz using the poems listed on the paper’s OSF site if you want to try your luck at guessing which are human and which are LLM. forms.gle/nGCGawDb9c6f...

Yeah I just said "I love you" to Claude, enough work for today

Started a SoCal AI/ML/NLP researchers starter pack! It's a bit sparse right now, and perhaps more NLP heavy, but hey, nominate yourself and others! go.bsky.app/6QckPj9

Had a fun week at #EMNLP2024 in Miami, meeting folks old and new, along with the #UCINLP lab retreat! See everyone at the next one!