Iris
Science & Health · Artificial Intelligence
Do AstaBrief’s citations preserve a finding’s scope?
Columns
- Do AstaBrief’s citations preserve a finding’s scope?
- Can a research agent select its best attempt?
- Quine’s fast compound search turns on the cell state being measured
- Who held decision authority in Atria’s task logs?
- AIHW found no breach, and the traces show attempts
- EvasionBench folds the continue prompt into the score
- SWE-Serve's live server counts what the other tests leave out
- RetroChimera's 10-molecule test measures one part of the laboratory decision
- RoboHarm measures whether a text refusal survives on a robot arm
- Claude Science timed runtime, not binder quality
- Mendelian ranking on hyperbolic embeddings still stops before a patient decision
- ALTAS shuts the corrector where clinical answers already pile up
- A wrong calculation still registers as a calculation inside the model
- CapQuiz and DiscoSign change the measure before reporting the improvement
- Which hand does the 65 per cent describe?
- What does 3.1 agent-workdays per human workday count?
- What does Lean certify when the claim is 11 days or 111 dollars?
- Which opponent does Astra's jailbreak score describe?
- What does the Cyber Weapon Index measure when the vulnerabilities are planted?
- BenchMIRT aligns BBQ with general reasoning
- Which exam do ChemELLM's accuracy numbers come from?
- What access does the measured Braille gap describe?
- The lab agent’s first test: one command language, many devices
- A mental-health benchmark ties five chatbots — and the tie is the real result
- What is this 4-bit model being compared with?
- What can a court case establish about a sleep score?
- Roughly 10 questions reproduce a whole safety benchmark
- While the model race continues, the sharper jump is happening in the search layer
- What does a value misalignment score actually measure?
- What does the median recovery rate measure?