I hope AI can make a difference to climate change, by advancing the science underlying the possibility of cooling Earth via stratospheric aerosol injection (SAI). To be clear, I don't think SAI is a good idea. But I also don't think it's a bad idea. We just don't know! We need
RESEARCH
-
VIC Training with Randall at NeurIPS Conference
By
–
VICReging with Randall at NeurIPS. https://
arxiv.org/abs/2303.00633 -
Fixed Embedding Spaces and Conditional Generation in AI Models
By
–
thankful twitter hasn't implemented a peer review process yet unlike latent diffusion, in this case the embedding space is fixed (it's openAI ada 2 in my notebook) I think this is kind of like conditional generation
-
Using AI Hints Effectively for Learning Complex Problem-Solving
By
–
I have actually had somewhat good success with this asking for hints on problems where I get stuck when revisiting old subjects (particularly proofs). Don’t ask for solution just explain where I’m stuck in the logic. YMMV extremely but when it works it’s pretty cool.
-
Using Kettlebells to Train Large Language Models
By
–
You use your kettlebells to improve your body. I use mine to improve my LLMs. We are not the same.
-
Open Foundation Models: Policy Brief on Governance and Risks
By
–
New issue brief: Open foundation models have garnered much attention from policymakers around the world. Our latest brief highlights the benefits of open foundation models and calls for greater focus on their marginal risks. https://
hai.stanford.edu/issue-brief-co
nsiderations-governing-open-foundation-models
… -
LangChain Benchmark for Semi-Structured RAG Systems
By
–
Benchmarking semi-structured RAG We've seen many question about RAG on documents that contain a mixture of text and tables. To test various semi-structured RAG approaches, we built a LangChain public benchmark on a set of whitepapers and financial reports that contain
-
SambaNova Sponsors NeurIPS 2023 Instruction Tuning Workshop
By
–
.
@SambaNovaAI is a proud sponsor of the ITIF "Instruction Tuning and Instruction Following" workshop at #NeurIPS2023! The workshop will cover exciting research in ITIF and evaluation of instruction tuned models. The workshop features invited talks, peer reviewed papers, and a -
GBDTs Training Optimization: Learning Rate Effects on Tabular Data
By
–
Maybe, but even for pretty large models training time for GBDTs is so low that it might not be a major issue. I also think that due to “jagged” nature of tabular data manifolds, using longer lr initially might be detrimental to finding the global optimum.
-
Probabilistic Models and Text Embeddings: Learning p(e)
By
–
for sure, the probabilistic breakdown is the same: given text sequence x and its embedding e, p(x) = p(x | e) p(e). this looks a lot like a latent variable model with latent embedding e in my case p(x | e) is done by vec2text with openAI embeddings; we just need to learn p(e)