TLDR AI 2024-08-02

The creators of VQGAN, Stable Diffusion, Latent Diffusion, and other startups have raised $30m+ dollars and started a new company. They have released new flagship image generation models which are extremely capable and come in a variety of tiers.

OpenAI CEO Sam Altman announced a collaboration with the U.S. AI Safety Institute to provide early access to the company's next major generative AI model for safety testing, aiming to address concerns that OpenAI has deprioritized AI safety.

The EU's risk-based AI regulation began on August 1 with staggered compliance deadlines categorizing AI applications into low/no-risk, high-risk, and limited risk tiers, imposing transparency, risk management, and penalties for violations. Standards for high-risk and powerful general-purpose AI models will be finalized by April 2025.

ARCLE is an environment to aid reinforcement learning studies using the Abstraction and Reasoning Corpus (ARC).

Stability AI has released a great new fast and powerful 3D generation model. It uses the same commercial license as the company's previous models.

MMTrail is a vast dataset that features over 20 million trailer clips with high-quality multimodal captions.

Integrate ChatGPT with Stream Chat to make your in-app messaging experience smarter and more engaging. Follow this step-by-step tutorial (with usable code samples) to learn how you can use Stream and the OpenAI API to have ChatGPT enter the chat.

DeepMind has released a library for studying language models via MI. The library helps with running models and functions from complex codebases without tons of importing headaches.

The PyTorch team has released a great sample library for chatting with language models locally. It includes a robust sample system and can run the latest Llama 3.1 models.

Engine Core demonstrates a pattern for enabling LLMs to undertake tasks of a given scope with a dynamic system prompt and a collection of tool functions.

Synchron has integrated OpenAI's ChatGPT with its brain-computer interface (BCI) technology to aid paralyzed patients in faster communication. Its BCI, called a stentrode, can interpret thought commands. It now offers AI-generated response options, potentially expanding to include multimodal inputs. Synchron is aiming for FDA approval and envisions evolving its AI integrations based on patient needs.

ChatGPT's new Advanced Voice mode and the new AI-powered Siri have contrasting philosophies. Copilot systems focus on safety and privacy with small models while Agent systems like ChatGPT Voice use powerful, multimodal models for more natural and dynamic interactions. This highlights the tension between lower risk, less capable systems and those offering more control and potential benefits.

Heeyo, a startup founded by Xiaoyin Qu, has launched an AI-powered app that offers interactive games and a chatbot for children aged 3-11. The app aims to provide personalized learning experiences while prioritizing safety with features like data protection and content designed by child development experts. Despite potential concerns about AI for kids, Heeyo has secured $3.5 million in seed funding. It is positioning itself as a safe educational alternative to popular gaming and video platforms.

Cerebras Systems announced a proposal for IPO to the SEC.

Meta's recent release of the open source LLM Llama 3.1, with 405B parameters and trained on 3.8x10^25 FLOPs, sparks debate on the effectiveness of FLOPs as a regulatory threshold.

Paris prepared for the 2024 Olympics with heightened security, employing AI to analyze CCTV footage at train and metro stations for potential threats.