Davis Summarizes Papers
@dblalock.skystack.xyz
I go through all the machine learning arXiv submissions each week and summarize 10 to 20 of my favorites.
Free forever and read by thousands...
This is an automated Substack Account of https://dblalock.substack.com
Discover more/Create at: @skystack.xyz
Free forever and read by thousands...
This is an automated Substack Account of https://dblalock.substack.com
Discover more/Create at: @skystack.xyz
2024-8-25: Scaling curves for All of the Things
2024-8-25: Scaling curves for All of the Things
dblalock.substack.com
November 20, 2025 at 3:34 PM
2024-8-25: Scaling curves for All of the Things
2024-8-4 arXiv roundup: LLama 3.1, training a 100T biological neural net
2024-8-4 arXiv roundup: LLama 3.1, training a 100T biological neural net
dblalock.substack.com
November 20, 2025 at 3:34 PM
2024-8-4 arXiv roundup: LLama 3.1, training a 100T biological neural net
2024-4-28 arXiv roundup: data and scaling, backlog highlights part 3
2024-4-28 arXiv roundup: data and scaling, backlog highlights part 3
dblalock.substack.com
November 20, 2025 at 3:34 PM
2024-4-28 arXiv roundup: data and scaling, backlog highlights part 3
Research Archetypes: Scientists, Mathematicians, Inventors
Research Archetypes: Scientists, Mathematicians, Inventors
dblalock.substack.com
November 20, 2025 at 3:34 PM
Research Archetypes: Scientists, Mathematicians, Inventors
2024-4-14 arXiv roundup: backlog highlights part 2
2024-4-14 arXiv roundup: backlog highlights part 2
dblalock.substack.com
November 20, 2025 at 3:34 PM
2024-4-14 arXiv roundup: backlog highlights part 2
2024-4-7 arXiv roundup: DBRX, Backlog highlights part 1
2024-4-7 arXiv roundup: DBRX, Backlog highlights part 1
dblalock.substack.com
November 20, 2025 at 3:34 PM
2024-4-7 arXiv roundup: DBRX, Backlog highlights part 1
2023-11-26 arXiv roundup: Big potential wins, 1 bit per parameter, Simplifying transformers
2023-11-26 arXiv roundup: Big potential wins, 1 bit per parameter, Simplifying transformers
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-11-26 arXiv roundup: Big potential wins, 1 bit per parameter, Simplifying transformers
2023-11-19 arXiv roundup: Inverse-free inverse Hessians, Faster LLMs, Closed-form diffusion
2023-11-19 arXiv roundup: Inverse-free inverse Hessians, Faster LLMs, Closed-form diffusion
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-11-19 arXiv roundup: Inverse-free inverse Hessians, Faster LLMs, Closed-form diffusion
2023-10-16 arXiv roundup: Cornucopia of easy (claimed) wins for LLMs
2023-10-16 arXiv roundup: Cornucopia of easy (claimed) wins for LLMs
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-10-16 arXiv roundup: Cornucopia of easy (claimed) wins for LLMs
2023-9 arXiv roundup: A bunch of good ML systems and Empirical science papers
2023-9 arXiv roundup: A bunch of good ML systems and Empirical science papers
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-9 arXiv roundup: A bunch of good ML systems and Empirical science papers
2023-8 arXiv roundup: Look I gave a talk, SILO-ing language models, lots of MoE + tool use papers
2023-8 arXiv roundup: Look I gave a talk, SILO-ing language models, lots of MoE + tool use papers
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-8 arXiv roundup: Look I gave a talk, SILO-ing language models, lots of MoE + tool use papers
2023-7-30 arXiv roundup: Better image captions, Scaling EMA, Chain of thought empiricism
2023-7-30 arXiv roundup: Better image captions, Scaling EMA, Chain of thought empiricism
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-7-30 arXiv roundup: Better image captions, Scaling EMA, Chain of thought empiricism
2023-7-23 arXiv roundup: OpenAI breaking changes, Much better attention and image captions
2023-7-23 arXiv roundup: OpenAI breaking changes, Much better attention and image captions
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-7-23 arXiv roundup: OpenAI breaking changes, Much better attention and image captions
2023-7-16 arXiv roundup: Weird step sizes help gradient descent, Better CPU matmuls
2023-7-16 arXiv roundup: Weird step sizes help gradient descent, Better CPU matmuls
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-7-16 arXiv roundup: Weird step sizes help gradient descent, Better CPU matmuls
2023-7-9 arXiv roundup: LLMs ignore the middle of their context, MoE + instruction tuning rocks
2023-7-9 arXiv roundup: LLMs ignore the middle of their context, MoE + instruction tuning rocks
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-7-9 arXiv roundup: LLMs ignore the middle of their context, MoE + instruction tuning rocks
2023-7-2 arXiv roundup: Self-supervised eval, Prompting text models like image models, KV cache eviction
2023-7-2 arXiv roundup: Self-supervised eval, Prompting text models like image models, KV cache eviction
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-7-2 arXiv roundup: Self-supervised eval, Prompting text models like image models, KV cache eviction
Models generating training data: huge win or fake win?
Models generating training data: huge win or fake win?
dblalock.substack.com
November 20, 2025 at 3:34 PM
Models generating training data: huge win or fake win?
2023-6-25 arXiv roundup: Learning from textbooks, Eliminating transformer outliers, Zero++
2023-6-25 arXiv roundup: Learning from textbooks, Eliminating transformer outliers, Zero++
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-6-25 arXiv roundup: Learning from textbooks, Eliminating transformer outliers, Zero++
Have we hit a statistical wall in LLM scaling? - 2023-6-18 arXiv roundup
Have we hit a statistical wall in LLM scaling? - 2023-6-18 arXiv roundup
dblalock.substack.com
November 20, 2025 at 3:34 PM
Have we hit a statistical wall in LLM scaling? - 2023-6-18 arXiv roundup
2023-6-11 arXiv: Training on GPT outputs works worse than you think, but training on explanations works great
2023-6-11 arXiv: Training on GPT outputs works worse than you think, but training on explanations works great
dblalock.substack.com
November 20, 2025 at 3:34 PM
2023-6-11 arXiv: Training on GPT outputs works worse than you think, but training on explanations works great