The Daily Index — Monday, September 23, 2024 · TLDR Index

HEADLINES & LAUNCHES

TLDR AI

TLDR AI

PROMPTING O1

This guide was missed in the excitement of OpenAI's new reasoning models. It shows how prompting this new model is different and requires simpler prompts and a more structured input context.

links.tldrnewsletter.com · 6 min

TLDR AI

JONY IVE CONFIRMS HE'S WORKING ON A NEW DEVICE WITH OPENAI

Jony Ive is collaborating with OpenAI CEO Sam Altman on a new AI hardware project. The venture could raise $1 billion by year-end and involves key former Apple designers. Specifics of the device remain undetermined, but it aims to leverage generative AI for advanced user interactions.

theverge.com · 2 min

TLDR AI

FINE-TUNING IMAGE-CONDITIONAL DIFFUSION MODELS IS EASIER THAN YOU THINK

Much work has been done to generalize pretrained image diffusion models into niche depth estimators and other image conditional models. This work found that by simplifying the problem and fixing a small bug, researchers were able to get substantially improved performance with less training compute.

arxiv.org · 24 min

TLDR AI

TRAINING LANGUAGE MODELS TO SELF-CORRECT VIA REINFORCEMENT LEARNING

Deepmind has released a paper that shows promise, even if the actual results aren't state-of-the-art. It shows a reinforcement learning paradigm that can be used to help models self-correct when generating math and code.

arxiv.org · 32 min

TLDR AI

MICHELANGELO: LONG CONTEXT EVALUATIONS BEYOND HAYSTACKS VIA LATENT STRUCTURE QUERIES

Another great Google paper that shows how to evaluate long context models. It is directionally similar to the recent work by Magic.

arxiv.org · 21 min

TLDR AI

MOST POWERFUL LLM ON A SINGLE GPU

Solar Pro is a 22B language model that can fit on a single 80GB GPU. The goal of this project is to make the most powerful model that can run on a single device.

upstage.ai · 12 min

TLDR AI

STUDY REVEALS AI ISSUES IN HOME SURVEILLANCE

Researchers find large language models make inconsistent decisions about whether to call the police when analyzing surveillance videos.

news.mit.edu · 9 min

TLDR AI

SAE INTUITIONS

Sparse Autoencoders are the number one tool used today to understand the internals of language models. This post explores the intuitions behind the models and some good information on how they work.

adamkarvonen.github.io · 18 min

TLDR AI

GENERALIZED KNOWLEDGE DISTILLATION TRAINER (HUGGING FACE HUB)

The TRL library has added GKD to its training procedures.

huggingface.co

TLDR AI

ARCADE AI WITH $17M RAISED LAUNCHES PRODUCT CREATION PLATFORM

Arcade is a new platform where users can go from prompt to product.

prnewswire.com · 2 min

TLDR AI

SALESFORCE TAPS NVIDIA TO DEVELOP AI-POWERED AVATARS

Salesforce is collaborating with Nvidia to create AI-powered avatars designed to enhance virtual interactions and customer service.

links.tldrnewsletter.com · 4 min