TLDR AI 2024-10-10

BUILD YOUR OWN ADVANCED VOICE IN A FREE 1 HOUR WORKSHOP ON 11/1. LIMITED SPOTS—APPLY TODAY!

ALPHAFOLD WINS NOBEL PRIZE IN CHEMISTRY

Demis Hassabis, John Jumper, and David Baker won the Nobel Prize in Chemistry for their contributions to protein folding with AlphaFold among other innovations.

GENOMIC FOUNDATION MODEL BENCHMARKING

GFMBench is a new framework designed to address challenges in genomic foundation model (GFM) development by providing standardized benchmarking tools. With millions of genomic sequences and hundreds of tasks, GFMBench automates benchmarking for open-source GFMs.

OPENAI REDUCING DEPENDENCY ON MICROSOFT DATA CENTERS

OpenAI is reducing its dependence on Microsoft data centers by securing its own compute capacity. Microsoft is also decreasing its reliance on OpenAI as it competes in the AI product market.

LLMS KNOW MORE THAN THEY SHOW: ON THE INTRINSIC REPRESENTATION OF LLM HALLUCINATIONS

Another study that shows language models encode internally when they aren't factual. This evidence can help us steer them and mitigate hallucinations.

DIFFERENTIAL TRANSFORMER

Transformers tend to over-allocate attention to irrelevant context. This work introduces Diff Transformer, which amplifies attention to the relevant context while canceling noise. Specifically, the differential attention mechanism calculates attention scores as the difference between two separate softmax attention maps. The subtraction cancels noise, promoting the emergence of sparse attention patterns

A SIMPLER CLIP FINE-TUNING

This paper introduces ProLIP, a method for adapting vision-language models like CLIP without adding extra parameters. By fine-tuning just the last projection matrix of the vision encoder, ProLIP achieves strong results in few-shot classification.

SCIENCE AGENT BENCH

Benchmark code for the science agent test. This essentially serves to measure how good models are at producing novel scientific advances.

CONTROLLED VISUAL GENERATION

Controllable AutoRegressive Modeling (CAR) is a new framework that brings precise control to pre-trained visual autoregressive models. The approach allows fine-tuned image generation with fewer resources by progressively refining control representations.

SPATIOTEMPORAL PREDICTION WITH PREDFORMER

PredFormer is a new transformer-based approach for spatiotemporal predictive learning that outperforms existing methods in both accuracy and efficiency.

TIKTOK PARENT COMPANY BYTEDANCE HAS A TOOL THAT'S SCRAPING THE WEB 25 TIMES FASTER THAN OPENAI

ByteDance is aggressively using its web crawler, Bytespider, to amass data, potentially for developing its own LLM. Bytespider's data consumption rate is significantly higher than its competitors, including OpenAI and Anthropic. The purpose of the collected data remains unclear, but TikTok has been integrating AI-powered features like AI-generated ads and avatars.

SONAIR TAKES A CUE FROM DOLPHINS TO BUILD AUTONOMOUS 3D VISION WITHOUT LIDAR

Sonair has developed an ultrasound-based 3D computer vision technology for autonomous hardware. Aiming to replace lidar by offering a cost-effective and low-energy alternative, the system uses AI for object recognition and provides a 180-degree field of view. Sonair is initially targeting autonomous mobile robots. It recently raised $6.8 million to support technology aimed at enhancing robot perception.

AI WILL USE A LOT OF ENERGY. THAT'S GOOD FOR THE CLIMATE.

AI data centers are driving demand for clean, 24/7 energy, pushing tech giants to invest heavily in renewable and nuclear power solutions. This surge in demand could accelerate the cost reduction of clean energy technologies due to their learning rates. In the long run, AI's energy needs may spur policy changes and advancements in clean energy infrastructure.

the webinar by Oxford Languages and Wellsaid

GENSIM SCALING ROBOT DATA WITH SYNTHETIC DATA

A great paper that shows how to scale robotic collection data with an improved high quality physics simulation dataset.

IMAGE RESTORATION

This project introduces a new differential equation-based method for image restoration.

TESLA'S HEAD OF VEHICLE PROGRAMS JUMPS TO WAYMO AHEAD OF ROBOTAXI REVEAL

Tesla's former head of vehicle programs, Daniel Ho, has joined Waymo as programs director.