AI News: Flower 1.2 released; Anthropic AI's Claude; Combining ChatGPT & Mathematica; Speedup transformers using FlashAttention; AI2's Objaverse 1.0 Dataset....

Hi there, today we will share some research updates from Anthropic AI's Claude; Speedup transformers using FlashAttention; Flower 1.2 released; AI2's Objaverse 1.0 released; Combining ChatGPT & Mathematica; GeorgiaTech's Visual Navigation and Locomotion over obstacles (ViNL) and many other cool AI updates. So, let's start...

Anthropic AI: Meet Claude, a Rival to ChatGPT created by Anthropic AI. Anthropic, the startup co-founded by ex-OpenAI employees that have raised over $700 million in funding to date, has developed an AI system called Claude similar to OpenAI’s ChatGPT that appears to improve upon the original in key ways. Claude was created using a technique Anthropic developed called “constitutional AI.

GeorgiaTech: Researchers trained a legged robot to navigate while stepping over obstacles in its path, similar to how humans and pets walk! The research team presents Visual Navigation and Locomotion over obstacles (ViNL), which enables a quadrupedal robot to navigate unseen apartments while stepping over small obstacles that lie in its path (e.g., shoes, toys, cables), similar to how humans and pets lift their feet over objects as they walk.

Stanford: Speed-up transformers by reducing data transfer between GPU memory and GPU on-chip SRAM. Researchers propose FlashAttention, an IO-aware exact attention algorithm that uses tiling to reduce the number of memory reads/writes between GPU high bandwidth memory (HBM) and GPU on-chip SRAM.

Flower: The Flower Team is excited to announce the release of Flower 1.2 stable. Flower is a friendly framework for collaborative AI and data science. It makes novel approaches such as federated learning, federated evaluation, federated analytics, and fleet learning accessible to a wide audience of researchers and engineers.

AI2: AI2 researchers present Objaverse 1.0 (A Universe of Annotated 3D Objects), a large dataset of objects with 800K+ (and growing) 3D models with descriptive captions, tags, and animations. Objaverse improves upon present day 3D repositories in terms of scale, number of categories, and in the visual diversity of instances within a category.

Towards Models that Can See and Read: Researchers present UniTNT, which leads to the first single model that successfully handles both "seeing" and "reading" images to solve VQA problems.

Google RT-1: ChatGPT can plan a theme party, but can it clean up your messy house after the party? Sadly no. What might a GPT for robotics look like? Google researchers released RT-1, a Transformer with eyes, arms & wheels as to feel the environment.

Combining ChatGPT & Mathematica: There are all sorts of exciting possibilities, suddenly opened up by the unexpected success of ChatGPT. But for now there’s the immediate opportunity of giving ChatGPT computational knowledge superpowers through Wolfram|Alpha. So it can not just produce “plausible human-like output”, but output that leverages the whole tower of computation and knowledge that’s encapsulated in Wolfram|Alpha and the Wolfram Language.

Sponsor our Newsletter | Join our 13k+ ML Subreddit