ChatGPT connects to GitHub repositories

PLUS: Sam Altman's Senate testimony, Open-source music generation, Mistral's new model & OpenAI's FDA discussions

In today’s agenda:

1️⃣ ChatGPT deep research now connects to GitHub repositories

2️⃣ ACE-Step delivers breakthrough open-source music generation

3️⃣ Mistral Medium 3 launches with mixed community reception

Plus, some interesting news:

  • OpenAI CEO Sam Altman to testified before Senate committee

  • OpenAI and FDA exploring AI for drug evaluation acceleration

MAIN AI UPDATES / 9th May 2025

🔍 ChatGPT connects to GitHub repositories 🔍
AI research gets direct code access

ChatGPT has released a new feature allowing users to connect their GitHub repositories to deep research. This integration enables ChatGPT to pull live data from repositories including code, README files, and documentation, analyzing it in real-time. The feature is currently available to Team users globally and is gradually rolling out to Plus and Pro users (except in EEA, Switzerland, and UK). Users can simply select Deep research in the composer, press the down arrow, select GitHub, and authorize the connector to access selected repositories.

🎵 ACE-Step: Open-source music generation 🎵
Local, fast, and Apache-licensed

The ACE-Step model has been released as an open-source, Apache-licensed music generation system that achieves remarkably fast inference speeds: 3 minutes of music in just 34 seconds on an RTX 4070, and under 3 seconds for shorter clips on a 4090. This speed significantly outpaces commercial solutions while running locally. The model is fine-tunable and designed for potential ComfyUI integration. While audio quality doesn't yet match proprietary leaders like Suno and Udio, its exceptional prompt-following precision and generation speed mark a significant advance for accessible AI music creation.

🤖 Mistral Medium 3 debuts with mixed reception 🤖
Cost-effective but controversial

Mistral AI has released Medium 3, their latest model, with competitive pricing ($0.4 input/$2.0 output per million tokens). However, community reception has been decidedly mixed, with some users describing it as "useless" while others see potential for creative writing applications. Technical comparisons suggest DeepSeek v3 may offer better performance at a lower price point, highlighting the increasingly competitive landscape for mid-tier AI models as developers seek the optimal balance between cost and capability.

INTERESTING NEWS

👨‍⚖️ Sam Altman urges lawmakers against AI regulations that could "slow down" U.S. 👨‍⚖️

OpenAI CEO Sam Altman testified before the Senate Committee on Commerce, Science, and Transportation on Thursday, focusing on winning the AI race against China. In a marked shift from his 2023 testimony that emphasized AI safety, Altman stressed that while "it's hard to say how far ahead we are" of China, continuing American leadership will require "sensible regulation that does not slow us down." He emphasized that "infrastructure is destiny, and we need a lot more of it" to meet growing demand for AI systems. Joined by executives from AMD, CoreWeave, and Microsoft, Altman faced largely friendly questions from lawmakers. He highlighted OpenAI's recently announced global expansion of its $500 million Stargate mega data center project, arguing that "an American-led version of AI, built on democratic values" must prevail over authoritarian alternatives. Notably absent from his testimony were the numerous references to AI safety that characterized his congressional appearance two years ago.

💊 OpenAI explores AI for drug evaluations with FDA 💊

OpenAI has met with U.S. Food and Drug Administration (FDA) officials to discuss using AI to accelerate drug evaluations, according to Wired. The discussions center around a project called cderGPT, an AI tool for the Center for Drug Evaluation which regulates prescription and over-the-counter drugs in the U.S. Representatives from Elon Musk's DOGE have reportedly participated in these talks. The initiative aims to speed up parts of the drug development process, which traditionally can take more than a decade to complete. While AI has potential to make notoriously slow steps more efficient, questions remain about controlling for the unreliability of AI models in such critical applications.

📩 Have questions or feedback? Just reply to this email , we’d love to hear from you!

🔗 Stay connected: