I haven't kept a list.
it's quite a few, mostly in the last 2 or 3 years because FAIR is still young.
Also, many have done the opposite: quitting a tenure faculty position to come to FAIR.
Lots of folks have accepted junior faculty positions and delayed the start to spend a year
RESEARCH
-
FAIR Researchers Career Transitions: Faculty Positions and Hirings
By
–
-
Non-obvious AI developments beyond major corporate shake-up
By
–
For Friday’s issue, a recap of some of the non-obvious stuff that happened in AI outside of that one guy getting kicked out of the office for a bit
-
Anthropic Founded by OpenAI Researchers After For-Profit Pivot
By
–
Really?
I mean, Anthropic was founded by people who left OpenAI.
Quite a few research-oriented people left OpenAI when it pivoted to a for-profit development outfit funded by Microsoft. -
NASA Lunar VIPER Mission AI Planning System Success
By
–
During undergrad and after graduation, I spent 3.5 years at @NASA working on the Lunar VIPER mission helping build a system health aware real time planning advisor using the @JuliaLanguage
. It’s so cool to see this coming together for the mission: -
Embeddings and RAG Models Driving Major AI Improvements
By
–
But one of the hottest areas by far is going to be embeddings. RAG is becoming a fixture in AI to improve the quality of model responses, and improvements in embedding models is going to be a big part of that. And after months of crickets, we got three big launches in 45 days.
-
RAG relevance for questions using many distant document elements
By
–
RAG can handle all sorts of different questions, it's really appropriate for extracting information from a document. An interesting question is "when the question requires using more many distant elements from doc, is RAG still relevant?"
-
RAG retrieval failures due to random variations, need more tests
By
–
Well spotted!
Both variations (blue and green ) are also due to random variations. For instance the RAG fails to retrieve relevant snippets on 1 single example. The tests I ran were not numerous enough yet to smooth these variations out. But maybe I'll run more. -
Error Correction Codes in Language versus Video Models
By
–
Good point. But, is the redundancy in language the same as in video? Do both contain the same type of error-correction information? We can design for better error correction codes, so I’m not sure they’re the same
-
Auto-regressive Prediction Limits: Video vs Language Models
By
–
Thanks, Chris. Auto-regressive prediction has failed for video in the past but not obviously for language. @ylecun has emphasised this drift, but I wonder how much (1) redundancy and (2) discreteness can provide self-correction.