Lab Insider: User Chats Are Exceptionally Unlikely to Enter Training, Zero If Opted Out
willdepue · x · 2026-09-09
In a forwarded exchange, tszzl responded to concerns about whether user interactions make it into model training: it is 'exceptionally unlikely' anything users did entered training data, and the chances are zero if they have opted out — which most likely have.
He added that breaking the PII-scrubbing boundary just to round the probability down from near-zero to absolute zero would set a terrible precedent, saying 'we won't do it' — a notable explicit statement on lab data-governance boundaries.
More from Companies & People
- X surges to #1 among news apps in nearly all European countries — Polymarket · 2026-09-09
- VB 1.0 PM reflects 35 years later: VB6 still tops legacy migrations at 67.6% — mkheck · 2026-09-09
- Mistral CEO Arthur Mensch: enterprises won't run intelligence on someone else's grid — open source is the answer — a16z · 2026-09-09
- Orbis hosts SF hackathon one week after launch, real-time video gen — _vztu · 2026-09-09
- OpenAI says its Navier-Stokes project ran ~10,000 concurrent agents under strict safeguards — LatentSpaceLeaper · 2026-09-09
- Meta's AI chief: Muse is Meta's most important platform since WhatsApp and Instagram — alexandr_wang · 2026-09-09