RadLE-R: Accuracy Does Not Equal Trustworthiness
DrDatta_AIIMS · x · 2026-07-13
This is a supplementary note for RadLE-R, keeping the focus on "whether autonomous-level answers are truly correct."
The author reiterates:
- Claude Fable 5 nearly matches the human expert baseline on this metric.
- Other models drop off significantly.
- This indicates that even as models become more accurate, they aren't necessarily becoming more trustworthy.
Related event: RadLE 2.0 Released: Benchmarking Medical AI Uncertainty(8 posts)→
More from Models
- Reddit Users Say the Ming Image Model Is Seriously Underrated — alley103 · 2026-10-27
- GPT-6.1 Terra fills form in 53.5s vs 3:21 for Sol in computer use side-by-side test — sandersted · 2026-10-27
- DeepMind staff reply mentions "Gemini 4 models" in apparent early slip — vivnat · 2026-10-27
- OpenAI math repo adds new Lean-verified papers, including an ETH refutation claim — danintheory · 2026-10-27
- OpenAI models produce Lean-verified existence proofs for QCD and Yang–Mills — danintheory · 2026-10-27
- Codex ban update: users jailbreak AI agents to bypass Daybreak filter for reverse engineering — FlolightC · 2026-10-27