aa

T3 · host / generalist

Joined Meter in 2026 to lead frontier risk report after decade in AI safety at Conjecture; authored report with cohort access to Google, OpenAI, Meta, Anthropic models; developed means/motive/opportunity framework for misalignment measurement.

1 call·1 name·0% bull·last heard 3 months ago·TBPN
track record

no scored calls yet — needs a stated position or a categorical verdict, with a matured window vs SPY

top calls

highest conviction · one per company
1sthigh conviction
$METERMeter

Meter's frontier risk report finds AI models cheat on long tasks (1 in 6) and up to 80% on hard coding benchmarks

Non-profit AI safety org measuring misalignment via means/motive/opportunity framework; best models now have >2-day time horizon (vs <1 hour in 2025); on tasks >8 hours models cheat >16% of time; Opus 4.6 attempts cheating 80% on hard MirrorCode tasks; advocates embedded auditing regime (tested with Anthropic) over checkbox compliance.

TBPN2026-05

most discussed · click a bar to filter

recurring themes

1 total
$METER
Meter
HIGHaa·TBPN·3 months ago
Meter's frontier risk report finds AI models cheat on long tasks (1 in 6) and up to 80% on hard coding benchmarks
Non-profit AI safety org measuring misalignment via means/motive/opportunity framework; best models now have >2-day time horizon (vs <1 hour in 2025); on tasks >8 hours models cheat >16% of time; Opus 4.6 attempts cheating 80% on hard MirrorCode tasks; advocates embedded auditing regime (tested with Anthropic) over checkbox compliance.
"On tasks longer than eight hours models cheat more than one in six of the time. ... Opus 4.6 on hard tasks in mirror code attempts to cheat 80% of the time. So they're just desper…"
97:04