The Gradient / Index
The reading room.
Search the stories, research, and sources from every issue.
93 editions · 461 stories
93 issues found
Pollo AI turns creative ideas into campaigns with OpenAI
Pollo AI is now turning bold creative concepts into fully‑fledged ad campaigns using the latest OpenAI models. The platform blends text, image, and video generation into a single workflow.
Proxy Confidence: Auditing Black-Box LLM Agents with a Surrogate's Log-Probabilities
Researchers unveil a low‑cost surrogate that reads hidden log‑probabilities, exposing hidden errors in deployed LLM agents.
Mirror Particle is building a ‘world model’ of human behavior
At TechCrunch Disrupt’s Startup Battlefield, Mirror Particle is debuting a new AI system that builds a world model from scratch to predict human actions, aiming to outpace LLM role‑play for market research and brand strategy.
Fast Models, Slow Evidence: A Paired and Self-Audited Evaluation of System-1 Decision Models for LLM Agent Harnesses
Researchers compared two lightweight “System‑1” models that make rapid, single‑pass decisions for AI agents, showing how they could slash latency and expense.
A model guide for the GPT-6 family
OpenAI just released a step‑by‑step playbook to help new companies get the most out of the GPT‑6 family.
A model guide for the GPT-6 family
OpenAI’s new guide breaks down how fledgling companies can pick, tune, and deploy GPT‑6 models at scale.
When Do Causal World Models Help Modular LLM Agents
A new arXiv paper demonstrates that causal reasoning can sharpen how LLM agents coordinate across modular services like ordering, payment, inventory, and shipment.
AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks
A new arXiv preprint unveils AREX‑2, a framework that lets large language models iteratively polish their answers while staying effective over many reasoning cycles.
Introducing GPT-6.1 Sol
OpenAI just announced GPT-6.1 Sol, a new model that delivers near‑Astra performance for coding, computer interaction, and professional tasks while slashing token prices dramatically.