BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//talks.python-summit.ch//swiss-python-summit-2026//speak
 er//LG8M8Z
BEGIN:VTIMEZONE
TZID:Europe/Zurich
BEGIN:DAYLIGHT
DTSTART:20251023T000000
TZNAME:CEST
TZOFFSETFROM:+0200
TZOFFSETTO:+0200
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20251026T030000
RDATE:20261025T030000
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20260329T030000
RDATE:20270328T030000
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
SUMMARY:From Open Weights to Trusted Models - Emmanuel Okedele
DTSTART;TZID=Europe/Zurich:20261023T143000
DTEND;TZID=Europe/Zurich:20261023T150000
DTSTAMP:20261111T040114Z
UID:pretalx-swiss-python-summit-2026-KJJUPP@talks.python-summit.ch
DESCRIPTION:APIs of frontier models are the fast way to ship\, but rarely 
 the cheapest place to stay. Open weights are a real alternative: you pay f
 or compute instead of per-token rent\, the model runs on your own machines
  so the data never leaves\, residency and governance are yours to set.\n\n
 The open question is whether an open-weight model is actually good enough 
 for your use case\, and which one. A leaderboard won't tell you. It measur
 es someone else's tasks\, not yours\, and an agent that runs over many tur
 ns and calls tools can score well on a benchmark and still fail the real j
 ob.\n\nThis talk is about answering that for yourself\, with two open-sour
 ce Python tools: Inspect\, the UK AI Security Institute's evaluation frame
 work\, and Petri\, Anthropic's auditor built on top of it. Between them th
 ey let you judge an open-weight agent on your own traffic and on the failu
 res you'd never think to collect\, so you can put a number on the open-ver
 sus-frontier trade-off and decide on evidence.\n\n**Takeaways**\n\nBy the 
 end you'll be able to decide whether an open-weight model is good enough t
 o replace a frontier API for your use case\, and which one. You'll be able
  to:\n\n- Build an evaluation from your own traffic\, then let Petri's aud
 itor extend it\, so you test the cases you've seen and the ones you haven'
 t.\n- Score candidates against your current model on real agent work\, acr
 oss many turns and tool calls\, not on a single tidy reply.\n- Turn the re
 sult into a decision: a number on the quality-versus-cost\, latency.
LOCATION:Aula 4.101
URL:https://talks.python-summit.ch/swiss-python-summit-2026/talk/KJJUPP/
END:VEVENT
END:VCALENDAR