Open cases

Weird, but true. Cases, evidence, and verdicts you can trace.

• Did OpenAI's o3 score over 25% on FrontierMath?

  At the December 2024 o3 announcement OpenAI reported over 25% on the FrontierMath benchmark. Epoch AI's independent test of the released model scored about 10%, and Epoch later disclosed that OpenAI funded the benchmark. Which number does the evidence support?

  AI capability claims · 0 claims · 0 evidence · 2026-09-22

• Did Llama 4 Maverick really rank #2 on LMArena?

  Meta's Llama 4 launch cited a #2 position on the LMArena leaderboard. The model on the board was an unreleased experimental variant; the public weights, re-tested, ranked 32nd. What does the evidence say about the launch claim?

  AI capability claims · 0 claims · 0 evidence · 2026-09-22

• Are orbital data centers really 10x cheaper on energy?

  Starcloud's white paper claims space-based compute can reach an energy cost around $0.002/kWh, roughly 10x below terrestrial data centers. Independent analyses in 2026 put total cost of ownership 2.5–3x higher than on the ground. Which claim does the evidence support?

  Compute & energy · 0 claims · 0 evidence · 2026-09-22

• Odyss N1: ¥200M raised, but where is the product?

  Odyss (OdyssLife) announced a large round led by Sequoia China and Monolith for an AI food-tracking necklace, with a shipping plan for Aug–Sep 2026. At IFA in September the device was still in beta with no price or ship date. Does the delivery claim hold?

  Consumer hardware & startups · 0 claims · 0 evidence · 2026-09-22

• Does Erly really make $50K a month?

  A viral story says a 24-year-old with no coding background built an alarm app with AI that now makes $50K/month. All figures trace back to one podcast.

  Solo founders & indie revenue · 0 claims · 0 evidence · 2026-09-22