Artificial Intelligence
AI 2040: George Hotz vs. the Cult of Superintelligence
George Hotz challenges the hard takeoff myth in a widely discussed post. Here's what he's really arguing about AI trajectories toward 2040,…
Read more →$ ls -la category: artificial intelligence
Category: Artificial Intelligence
Artificial Intelligence
AI Coding Agents: Legacy Code Migration Finally Within Reach
Terence Tao used AI coding agents to revive his old Java applets. This real-world case shows how coding agents are transforming legacy…
Read more →
Artificial Intelligence
How to Train Your GPT: I Actually Tested It
I tested How to Train Your GPT hands-on: installation, local CPU training, and text generation. Here is what this educational repository actually…
Read more →
Artificial Intelligence
OmniAgent: field test of an agent that hardens code
OmniAgent promises to harden code and self-improve. My real-world test on a vulnerable Python file reveals a serious prototype, but not yet…
Read more →
Artificial Intelligence
brain0: real-world test of a black box for AI agents
brain0 links AI-generated code to its originating prompts. A real test on a Git repo: drift detected, provenance JSON, offline mode. A…
Read more →
Artificial Intelligence
AI-Generated Videos Designed to Maximize Brain Activation: EPFL’s NEVO Project
EPFL's NEVO project uses a simple computational algorithm to generate videos that maximally activate a target brain region, raising both exciting research…
Read more →
Artificial Intelligence
GPT-Live: OpenAI’s Full-Duplex Voice AI That Changes Everything
OpenAI launches GPT-Live, two full-duplex voice models that listen and speak at the same time. Here's what it actually means for developers…
Read more →
Artificial Intelligence
Meta’s AI Agents: Zuckerberg Admits Things Aren’t Moving as Fast as Hoped
At an internal town hall, Mark Zuckerberg admitted Meta's AI agents haven't progressed as fast as hoped. A candid signal for the…
Read more →
Artificial Intelligence
Better LLM Models, Worse Tools: The Proprietary Fine-Tuning Paradox
More powerful LLM models can be less compatible with your custom tools. Armin Ronacher documents this paradox with Claude Opus 4.8 and…
Read more →
Artificial Intelligence
Kimi K2.7 Code in GitHub Copilot: First Open-Weight Model in the Selector
GitHub adds Kimi K2.7 Code to Copilot: the first open-weight model in the selector, a lower-cost alternative to Claude Sonnet. Here's what…
Read more →