Scott Aaronson: AI's math progress makes "pinnacle of human intellect" claim untenable

GitHub's tutorial walks through opening a Copilot app session in VS Code so the task, plan, conversation, and code changes carry over without re-summarizing. Steps cover reviewing the agent's diffs line by line, editing and testing locally, then pushing updates to the session's pull request.

Mercury 3 runs at 1,068 tokens/sec on NVIDIA GPUs and claims to outperform GPT-6 Luna and Claude Haiku 5.5, with a 1M-token context (up from 260K). Priced at $0.20/M input and $0.70/M output, 50% off at launch; it is 40x larger than Mercury 1.

Cognition launched three enterprise features for Devin: a 0–100 Efficiency Score grading productive engineering hours per unit of spend, Devin Coach pre-run prompt feedback, and auto-approvals that raise limits for engineers graded Healthy or Satisfactory up to an admin-set ceiling.

Together Post-Training lets developers post-train open models like Qwen, GLM and Kimi via SFT or RL, with Together handling the infrastructure while users define the training loop. The company pitches specialist agents matching frontier quality at a fraction of the cost and runtime.

The C++ authoring layer lets developers build computation graphs op by op and run the same graph on CPU, GPU (via open-source ML Drift) and NPU, or directly on CPU through XNNPACK without a model file or runtime. Google recommends the standard convert-and-run path for most projects, positioning this for custom architectures and new lowering logic.