I also think there was a will to experiments on too many aspects at the same time and in parallel in BS: try a new organisation, a multilingual model with low-ressources languages, a new way of building a dataset (crowd sourcing of sources), attempts to train in fp16, etc etc
@thom_wolf
-
Large AI Teams: BigScience Lessons on Project Organization
By
–
yes, agree on the exhaustion feeling. BigScience was an experiment in trying to involve as many/diverse people as possible but a take-away is that too large teams can be counter-productive. Smaller projects like T0/ethics/RAIL were more successful inside BS than larger Bloom imo
-
LLMs Generate Convincing but Wrong Completions Often
By
–
this is an issue I also have with copilot. it often very convincingly pushes for long and beautiful but wrong completions LLM as “internet influencers”: surface form trumps content https://
x.com/shortstein/sta
/shortstein/status/1587857803678748672
… -
NeRF Zero to SOTA Tutorial at Hugging Face Paris
By
–
I'm in Paris, FR this Thursday 11/3, ping me if you want to chat. I'll do an internal tutorial zero->SOTA on NeRF at Hugging Face, might extract a 5min video to share online from it if its exciting to people.