TLDR AI 2023-10-17

AMD has improved support for its RDNA GPUs (including the RX 7900) and Pytorch for more training options for AI projects.

Nanotronics is a New York-based industrial AI company that wants to build an AI-enabled chip factory that can be assembled and expanded modularly with prefab pieces. Its Cubefabs system uses AI to take away the need for the specialization normally needed in a lab and allow people who are not semiconductor experts to work at the facilities. Each facility will need only about 30 people to operate. The bulk of each facility can be flat-packed and put in a shipping container.

Dataset of tabular tokens to encourage the community to build Large Data Models that better understand tabular data. It is the largest publicly available tabular dataset - a compilation of 627 million tables together with 867 billion tokens of contextual information. TabLiB is available on Hugging Face.

Figuring out the direction someone's head is turned is important for lots of tech applications. These researchers have developed a new way to estimate head positions from any angle.

The authors of this paper have created a new technique called SupFusion to make LiDAR and camera systems work together better for detecting things like cars or pedestrians.

The research introduces Ex-MCR, a novel method that efficiently learns unified contrastive representations for multiple modalities without needing paired data. By aligning existing Multi-modal Contrastive Representations, Ex-MCR achieves top performance in tasks like audio-visual retrieval and 3D object classification.

MosaicFusion is like a magic tool that can create new pictures with lots of objects without needing any prior learning. It does this in two steps: first, it makes the picture, and then it creates a mask to show where each object is.

This study introduces a new method that combines digital surface model (DSM) data and aerial images from different times to improve change detection beyond just 2D perspectives.

Libgen is likely a dataset in many closed models. While the legality of this dataset for commercial use is under debate, researchers are still using it to better understand data quality for language model training.

Although today’s AI models aren’t perfect, there is still a strong chance that they will be eventually recognized as the first true examples of artificial general intelligence.

With the dramatic increase in interest in language modeling. Many existing ML conferences have been inundated with papers. This new venue, organized by many prominent scientists, will be a great option for publications.

Although Wall Street investors have a hard time valuing and understanding Meta’s open-source approach to AI, many in the tech industry believe that positioning Llama similarly to Linux as a smart play.

Generative artificial intelligence is poised to boost global economic productivity and create new wealth, but we need to be careful of widening inequality due to mass job displacement.

Open source AI calendar scheduling.

This repository looks at the system messages used for ChatGPT and how they might influence custom instructions.