The day in AI, in five minutes.
OpenAI reset Codex and ChatGPT Work usage limits across the board, while Meta released code for developers to build Muse AI gadgets and NVIDIA said DGX Spark will be available this month. Claude and Google Gemini entered the US App Store top 10 free apps at #8 and #10, then dropped to #11; Gemini 3.8 Flash reached #8 among OpenRouter models by spend, and OpenAI's share of OpenRouter spend rose 1.9 points to 28.1%. MiniMax reported elevated errors on its Large Language Models, and its status page returned to normal after 10 minutes and after 11 minutes.
What mattered
- 1
Amazon's $1B data center community plan draws criticism
Why it matters: Amazon reportedly announced a $1 billion plan to address community backlash over data centers, with critics saying it downplays pollution and Amazon praised for ending NDAs.
Amazon announced a $1 billion plan to address community backlash over data centers. Critics say the plan downplays data center pollution, though Amazon was praised for ending NDAs.
- 2
OpenAI resets Codex and ChatGPT Work usage limits
Why it matters: The across-the-board reset changes usage limits for people using Codex and ChatGPT Work.
OpenAI staff announced that usage limits for Codex and ChatGPT Work have been reset. The reset applies across the board.
- 3
Meta open sources code for building Muse AI gadgets
Why it matters: The open source code lets developers build gadgets featuring Muse, including reminders on a color E Ink display or output on an HDMI stick.
Meta released open source code that lets developers build their own gadgets featuring its new Muse AI agent. Suggested projects include running Muse on a color E Ink display for reminders or on an HDMI stick for big screen output.
- 4
NVIDIA to ship DGX Spark with 64GB unified memory
Why it matters: DGX Spark will be available this month with 64GB of unified memory from manufacturer partners including Acer, for developers running open AI models locally.
NVIDIA says its DGX Spark will be available this month with 64GB of unified memory from manufacturer partners including Acer. The device is aimed at developers running open AI models locally.
- 5
Google Research publishes work on provably private federated learning
Why it matters: Google Research's post concerns privacy guarantees for learning from federated data.
Google Research published a post titled Toward provably private learning from federated data. The post concerns privacy guarantees for learning from federated data.
The Vikshy Score
The close of 3 Oct, 06:00 UTC, against the previous close. The full board.
The numbers
Read straight from the source. Every signal.