Frontier models now 'quite bad' at geolocating photos after o3-era peak
paularambles · x · 2026-10-06
Paula shares hands-on experience: summoned to use AI to figure out where photos were taken, she finds models are now quite bad at it. Quoting Miles Brundage, she notes that since o3, AI systems' reasoning, image understanding and web search were good enough to geolocate photos pretty easily — a capability people seem to have forgotten.
Related event: Ex-OpenAI policy chief warns AI can geolocate photos since o3(3 posts)→
More from Models
- Will OpenAI Merge Chat and Codex Quotas? Users Fear Tighter Basic Chat Limits — Iwantthegreatest · 2026-10-06
- Microsoft page reportedly confirms GPT-6 series uses Looped Transformers — teortaxesTex · 2026-10-06
- Amazon's ALoDLM: Token-Adaptive Looped Diffusion LMs Beat AR Baselines — amazon · 2026-10-06
- Looped LMs at Fixed Points: 3x Smaller KV Cache, 1.79x Faster Prefill — IFM · 2026-10-06
- Australia open-sources Matilda Jev, a 56.8ms decision model that skips text generation — Med1_Ai · 2026-10-06
- Codex politely pushes back on a user's hallucinated claims — "I'm also a journalist" — MikePFrank · 2026-10-06