MACHINE LEARNING

→ View original post on X — @googleai

Google Distillation Method Trains Smaller LLMs Efficiently

By

@googleai

–

12 July 2023 21h08

Deploying LLMs is memory inefficient and compute-intensive. At the #ACL2023 Google booth at 3pm today, @chunliang_tw will describe a new method, distilling step-by-step, that trains successful small models using fewer training data for fine-tuning and distillation.

Amazing team of scholars creating datasets and visualizing bird data

By

–

12 July 2023 20h25

It's from an amazing team of scholars, coders & artists: -NYT & institutional data: @ananny -How to see birds: @JerThorp -C4 & the making of datasets: @_will_orr
-Unstable categories & biodiversity: @Hamsini_S
-The trouble with ImageNet: @SashaMTL

The Exponential Growth of AI Training Datasets Over Two Decades

By

–

12 July 2023 20h25

Datasets are now impossibly large. Back in 2003, Caltech 101 was 10,000 images. By 2010, ImageNet had 14m images. In 2023, LAION has 5 billion images and text captions – it'd take you 1000+ years to look at it all. The entire territory of the internet has become the map of AI.

See:Set: New Tools for Analyzing Dataset Patterns and Worldviews

By

–

12 July 2023 20h25

That doesn't mean we stop analyzing them – we just use new tools. At Knowing Machines we've developed See:Set, a search engine to look into datasets and track patterns. These 9 essays come from direct experiences using See:Set to understand how datasets encode worldviews.

9 Ways to Study and Analyze Datasets: Research Findings

By

–

12 July 2023 20h25

New release: '9 Ways To See A Dataset" https://
knowingmachines.org/publications/9
_ways_to_see_a_dataset
… This is a 9-part series about how to study datasets – and what we found. Dives into individual datasets like LAION, ImageNet, NYT Corpus, C4, NABirds & iNaturalist. But studying datasets is harder than ever…

→ View original post on X — @xai

Most Fundamental Unanswered Questions in AI

By

@xai

–

12 July 2023 19h17

What are the most fundamental unanswered questions?

→ View original post on X — @_jasonwei

Med-PaLM Research Published in Nature Journal

By

@_jasonwei

–

12 July 2023 19h17

Honored to have played a small role in Med-PaLM, now published in Nature! https://
nature.com/articles/s4158
6-023-06291-2
… My grandpa is 89, and every year he asks if I've published in Science or Nature (he knows I'm a researcher, but he only knows those two journals). This year I can finally say yes!

→ View original post on X — @mit_csail

PhD Student Shares AI Transparency and Interpretability Research Insights

By

@mit_csail

–

12 July 2023 18h34

Hi! This is @aspenkhopkins
! I’m a PhD student in @aleks_madry
’s group. I work on AI/ML transparency, interpretability and fairness. I’ll be taking over CSAIL’s socials for the next couple of days to talk about research, life, and living in Cambridge. Stay tuned ~

→ View original post on X — @hwchase17

PostgreSQL HNSW Extension for Vector Embeddings by Neon

By

@hwchase17

–

12 July 2023 18h20

pg_embedding @PostgreSQL is popular database choice. Excited to share a new extension from @neondatabase to help you use it for embeddings as well (with HNSW)! Blog: