The Blog
AI & Machine Learning
Every post filed under AI & Machine Learning.
GPT-6 Astra costs 2.5 times as much as Sol, with roughly half the messages
OpenAI released GPT-6 Astra on September 3 with a million-token context window, higher prices and tighter message limits. Early code-review results show gains, especially across files, while cybersecurity access remains restricted.
Read More
Mechanical Turk’s Shutdown and the Human Labor Behind AI
Amazon plans to close Mechanical Turk on September 30, 2026. Its workers helped build AI training datasets. Teams still using the service now face a short migration…
ByteDance's Reported 10 Trillion-Parameter Model Faces a Hardware Test
ByteDance is reportedly training a huge AI model on domestic Chinese chips. The harder test is keeping 30,000 accelerators coordinated, supplied with data, and running…
Claude Fable Reportedly Helped Find a Jacobian Counterexample. Verification Comes Next.
A reported degree-7 counterexample to the Jacobian Conjecture raises questions about AI-assisted mathematical search, independent verification, and possible uses in…
Leanstral 1.5 Reportedly Found Five Bugs Across 57 Repositories
Mistral reports that its open-source Lean 4 model found five previously unreported bugs, including a Rust integer overflow that panics in debug builds and produces…
GLM-5.2 Makes Open-Weight AI a Stronger Option for Production Coding
GLM-5.2's reported coding scores sit close to Claude Opus 4.8, with lower API prices and MIT-licensed weights. The production decision still depends on workload tests…
GPT-5.6 Adds Three Pricing Tiers and a Government Review Step
OpenAI's Sol, Terra, and Luna models offer different cost and capability tradeoffs. Their limited launch also adds government review to the dependencies teams must…
Noam Shazeer Joins OpenAI After Google’s $2.7 Billion Character.AI Deal
Google’s 2024 deal brought transformer co-inventor Noam Shazeer back to work on Gemini. His move to OpenAI puts model architecture, inference costs, and research…
LLM Failover Takes More Than a Second Provider
Switching LLM providers can change prompt behavior, output formats, and tool calls. A useful backup needs testing at the workflow level, not just a second API endpoint.