The Gradient / Index
The reading room.
Search the stories, research, and sources from every issue.
93 editions · 461 stories
93 issues found
The LLM Proposes, the Executive Disposes: A Self-Verifying Agent Instrument that Dissociates Commitment Drift from Binding Drift in Long-Horizon Agents
Researchers unveil a deterministic Executive that forces language models to act only after pre‑registered predictions are validated, enabling structural verification for long‑term agents.
AutoFOAM: The Self-Refining Autonomous OpenFOAM Agent
A new self‑refining AI agent can set up, run, and improve CFD studies in OpenFOAM using only plain English.
How we built a realtime system for responsive voice AI in six months
OpenAI’s new GPT‑Live platform brings turn‑less, low‑latency voice interaction to users, shrinking conversation delays to near‑real time.
LLM Framework for Discovering Major Mathematical Conjectures: AI's Quest for the Next Riemann Hypothesis
Researchers unveil a three‑stage AI framework designed to generate and vet major mathematical conjectures, pushing automated discovery toward breakthroughs like the Riemann Hypothesis.
Advancing the price-performance frontier with GPT-5.6
OpenAI announced lower pricing for its GPT‑5.6 models on Luna and Terra, promising better cost efficiency for large‑scale AI deployments.
Advancing the price-performance frontier with GPT-5.6
OpenAI announced new, lower pricing for its GPT-5.6 models on the Luna and Terra platforms, promising faster, cheaper AI at scale.
Advancing the price-performance frontier with GPT-5.6
OpenAI announced lower pricing for its latest GPT‑5.6 models on Luna and Terra, promising tighter price‑performance ratios for businesses. The move aims to make large‑scale AI workflows more affordable.
How enabling two settings tripled our scores on the ARC-AGI-3 benchmark
A recent OpenAI experiment shows that toggling just two API settings can dramatically boost performance on the ARC‑AGI‑3 benchmark.
Gemini API Managed Agents: 3.6 Flash, hooks, and more
Google’s Gemini API just got a major upgrade, introducing version 3.6 features like Flash and extensible hooks. These enhancements aim to make managed agents more reliable and production‑ready for developers.