AI ALIGNMENT FORUM
AF

Wikitags

Calibration

Edited by Yoav Ravid, gustaf, brook, Jim Fisher, et al. last updated 2nd Apr 2025

Someone is well-calibrated if the things they predict with X% chance of happening in fact occur X% of the time. Importantly, calibration is not the same as accuracy. Calibration is about accurately assessing how good your predictions are, not making good predictions. Person A, whose predictions are marginally better than chance (60% of them come true when choosing from two options) and who is precisely 60% confident in their choices, is perfectly calibrated. In contrast, Person B, who is 99% confident in their predictions, and right 90% of the time, is more accurate than Person A, but less well-calibrated.

See also: Betting, Epistemic Modesty, Forecasting & Prediction

Being well-calibrated has value for rationalists separately from accuracy. Among other things, being well-calibrated lets you make good bets / make good decisions, communicate information helpfully to others if they know you to be well-calibrated (See Group Rationality), and helps prioritize which information is worth acquiring.

Note that all expressions of quantified confidence in beliefs can be well- or poorly- calibrated. For example, calibration applies to whether a person's 95% confidence intervals captures the true outcome 95% of the time.

List of Calibration Exercises
based on this post. Todo: find more & sort & new post for visibility in search engines?

  • WebApp with user created Exercises
  • PDF by CFAR
  • by CFAR
  • based on The Success Equation
  • based on Scout Mindset
  • by Guided Track (Also linked to by 80000 hours and OpenPhil)
  • by Outside the Asylum
  • by Peter Attia
  • by Quantified Intuitions
  • by 2pih

Exercises that are dead/unmaintained

  • https://www.metaculus.com/tutorials (dead link)
  • http://web.archive.org/web/20100529074053/http://www.acceleratingfuture.com/tom/?p=129
  • http://credencecalibration.com (dead link)
  • https://calibration.lazdini.lv (dead link)
  • http://web.archive.org/web/20161020032514/http://calibratedprobabilityassessment.org/
  • https://predictionbook.com/credence_games/try (deprecated; see also this Github Issue)
  • https://calibration-training.netlify.app (dead link)
     

 

Subscribe
2
Subscribe
2
Discussion0
Discussion0
Posts tagged Calibration
33Paper: Teaching GPT3 to express uncertainty in words
Owain_Evans
3y
0
36Behavior Cloning is Miscalibrated
leogao
4y
3
19Do LLMs know what they're capable of? Why this matters for AI safety, and initial findings
Casey Barkan, Sid Black, Oliver Sourbut
2mo
0
Add Posts