Tag: #research297 results
Distinguishes rising LLM inference compute into larger tasks (human-like linear scaling) vs. true inefficiency beyond human cost fractions. Uses Pareto frontier of budget vs. 50% reliability time-horizon. Argues much progress is bigger tasks, not unsustainable scaling.
A 1 GW orbital data center would cost $42.4 billion. This is nearly three times the cost of equivalent ground-based facilities. Highlights severe economic challenges for space-based AI.
Agentic development accelerates code writing, review, and shipping, outpacing traditional testing. Meta introduces JiTTesting to enable faster, just-in-time bug detection as code lands. This evolves testing frameworks for modern development speeds.
GPT-5.2 proposed a new formula for gluon amplitude in a preprint. OpenAI and academic collaborators formally proved and verified the result.
OpenAI blog post explores harness engineering using Codex in an agent-first world. Written by Ryan Lopopolo, Member of the Technical Staff. Focuses on leveraging code generation for AI agent development.
OpenAI blog post explores harness engineering using Codex in an agent-first world. Authored by Ryan Lopopolo, Member of the Technical Staff. Focuses on leveraging Codex for advanced AI agent development.