LLMs are a distillation of human understanding across all of humanity. I like this analogy.
@yitayml
-

Med-PaLM-2 Achieves Impressive Results in Healthcare AI
By
–
Amazing Med-PaLM-2 results! Great job! Great to see LLMs making great impact on health and glad to have contributed to the underlying improved & new PaLM model!
-
Google Launches PaLM API: Architecture Improvements Shared
By
–
New PaLM API launched! Feels amazing to be able to share your work with the world! Glad to have contributed to this massive team effort by helping to lead architecture and objective improvements for this latest generation of PaLM!
-
Has Model Distillation Received a New Name Recently?
By
–
Did model distillation get a new name recently?
-
Flan-T5-XXL Performance Compared to Decoder-Only Models
By
–
people also forget that the speed of flan-t5-xxl is also equivalent to a ~5B+ decoder-only model because it's an encoder-decoder model.
-
Custom Fine-tuning Recommended Over FLAN-UL2 for Specific Tasks
By
–
+1 thanks. highly recommend the custom fine-tuning on top for whatever you want/need. flan-ul2 is not built for general chit-chat.
-
Academic AI Tasks Bias Toward Short Length Outputs
By
–
Read the flan limitations section of the blogpost. Most of the academic tasks are short. It creates a bias for short length outputs.
-
UL2 Model Impact Greater Than Expected
By
–
On the surface, UL2 seems great. But it is much greater and has made more impact than people would expect.
-

Google Worker’s Health Toll in AGI Research
By
–
Finally after close to 3.5 years of being a Googler… I finally managed to use some massage points Didn't realize how wrecked my neck & shoulders were and how many neck/ shoulders have to be collectively sacrificed to get to AGI lol
-

Jerry Wei’s Research Continues Brain’s AI Excellence
By
–
Even though @_jasonwei left Brain for OpenAI, did you know that Brain still has @JerryWeiAI
, Jason's exceptionally talented brother! Research ability clearly runs in the family! Check out this amazing thought-provoking paper led by Jerry Wei: https://
arxiv.org/abs/2303.03846