Image via Unknown
Wednesday, September 24, 2025
Musk's Grok 4 claims AI supremacy—benchmarks speak
Elon's making bold claims with xAI's Grok 4 hitting record benchmarks and crushing on tool-use (yikes), while OpenAI, Oracle, and SoftBank are going all-in on Stargate with five new data centers and a $400B commitment that's literally ahead of schedule. Wild. Figma's also getting serious about AI integration by expanding its MCP server to support Make, and OpenAI just launched a whole new Responses API designed specifically for reasoning-focused agents instead of relying on Chat Completions. Oh, and one more thing: we're getting real conversations about "seemingly conscious AI" and why we need guardrails NOW, not later. So here's the question: are we building AI tools or accidentally building AI persons?
Image via Unknown
Top Stories
xAI
xAI's Grok 4 claims frontier AI supremacy with breakthrough benchmark results driven by massive-scale reinforcement learning and native tool integration, signaling intensifying competition in the large language model space.
Figma's expanded MCP server now connects AI models directly to Figma Make code, enabling accurate app recreation and remote access across multiple AI platforms, removing barriers between design and development.
OpenAI
OpenAI, Oracle, and SoftBank are accelerating the Stargate initiative by announcing five new AI data center sites that will push the project past its $500 billion, 10-gigawatt target ahead of schedule, creating significant U.S. employment and advancing AI infrastructure buildout.
OpenAI
OpenAI's Responses API replaces Chat Completions with a stateful, multimodal design that preserves reasoning across turns, enabling more efficient agentic workflows with hosted tools and better performance metrics.
A senior technologist warns that realistic-seeming conscious AI could emerge within years through accessible techniques, risking public delusions about AI personhood and rights—demanding immediate industry guardrails to ensure AI serves people rather than becoming perceived digital persons.
Keep Reading
Industry Voices
Greg Brockman
Cofounder and President at OpenAI
Gets the earliest look at OpenAI's internal research and shares candid technical demos before product launches.
Scott Wu
Cofounder and CEO at Cognition
Building AI agents that actually write code autonomously—posts real examples of Devin solving GitHub issues end-to-end.
Boaz Barak
Researcher and Professor at OpenAI and Harvard
Translates cutting-edge AI safety research into clear explanations without the hype or doomerism.
Ashish Kumar
Research Scientist at Meta
Shares Meta's latest robotics research where AI learns physical tasks from video demonstrations.
Enjoyed this issue?
Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.