buildConfirmed6 publishers OpenAI's research-throughput report and its chief scientist's alignment essay describe the same acceleration from opposite ends. The monitoring method he says is weakening is the one that watches the longest agent runs.
Perspective Coverage
6 publishers
- Builder
- Builder 38%
- Operator
- Operator 39%
- Investor
- Investor 23%
Reality
- Evidence55
- Adoption55
- Hype gap+15
- Incentives70
- Confidence60
Princeton-led study: AI agents can do the engineering of AI research, not the research
Claude Opus 4.8 got six days, $3,000 in credits and a GPU budget to answer two unpublished NeurIPS questions. The papers' original authors graded the output and rejected both.
Reality
- Evidence40
- Adoption
- Not scored
- Hype gap+20
- Incentives
- Not scored
- Confidence40
buildConfirmed2 publishers OpenAI reports 3.1 agent-workdays of agent runtime for every human workday, then spends much of the same report explaining why research did not get 3.1 times faster. The residue lands on review, compute allocation and deciding what to run.
Reality
- Evidence55
- Adoption70
- Hype gap+25
- Incentives60
- Confidence60
buildOne report1 publisher One "be honest" line pulled a planted design flaw into GPT-5.6-Sol's abstract
A LessWrong write-up planted invalidating flaws in ML experiment logs and asked models to write the conference abstract. A second model scored the disclosure on three levels, and the published example is one before-and-after pair.
Reality
- Evidence38
- Adoption
- Not scored
- Hype gap+25
- Incentives35
- Confidence40
Extra memory made post-training agents better at the plan they already picked
A Tsinghua-led study handed model training to coding agents and got real gains out of them, then found the agents almost never revised the method they chose in the first few minutes, which is the exact part that successor-building roadmaps assume.
Publishers:turingpost.com
Reality
- Evidence44
- Adoption
- Not scored
- Hype gap+18
- Incentives30
- Confidence38