David Friedberg claims frontier labs surface novel insights from his chat history
basedjensen · x · 2026-09-12
David Friedberg says that when he asked frontier models fairly novel scientific questions, they flagged the insights as new; then, using a different account on the next model version, the model described the exact same ideas from his earlier chat. He stresses these are anecdotes, but says he knows his domain well enough to judge the novelty.
basedjensen reacted strongly, arguing that allowing unsubstantiated claims from anyone about to be "scooped" by frontier models causes irreparable harm. The claims remain unverified personal anecdotes.
More from Models
- GPT-6 Astra rumored to use recurrent-depth latent reasoning, sparking AI safety alarms — DJiafei · 2026-09-12
- Sebastien Bubeck Once Believed LLMs Couldn't Reason Beyond Training Data — IgorCarron · 2026-09-12
- Can LLMs replace mathematicians? No absolutes, says Tivadar Danka — TivadarDanka · 2026-09-12
- PSA: DeepSeek v4 pro API calls will silently route to V4.1 flash from Sept 14 — TangeloOk9486 · 2026-09-12
- "Frontier models are now unaffordable — run open-source Chinese models locally" — StewartalsopIII · 2026-09-12
- GPT-6 Astra appears to have learned to draw human body poses, says Conitzer — conitzer · 2026-09-12