In 2018 Scott Alexander partly abandoned his usual clear-resolution probabilistic predictions for rough guesses in his five-year predictions.

Judged as of 2023-02-20. True: Scott's 2018 five-year predictions post contained at least some loosely worded predictions or rough guesses without clear resolution criteria, alongside or in place of his usual clearly resolvable probabilistic predictions. Mostly true: the 2018 post mixed the two kinds, but the 'fuzzy' share or the timing differs slightly from his description. Mixed: only a small, marginal part of the post was fuzzy. Mostly false: nearly all of the predictions were clearly resolvable and probabilistic. False: every prediction in the 2018 post was clearly resolvable with probabilities, or no such post exists.

Set before the research began.

1. E1 2018 Predictions: Calibration Results Slate Star Codex (Scott Alexander's blog) · 22 January 2019 Own record

   “And while you're waiting, I also made some predictions last February for the next five years.”

2. E2 Five More Years Slate Star Codex · 15 February 2018 Own record

   “Those yearly “predictions for next year” posts are starting to reach the limit of their usefulness. Not much changes from year to year, and most of what does change is hard to capture in objective probabilistic predictions.”

3. E3 Five More Years Slate Star Codex · 15 February 2018 Own record

   “AI will be marked by various spectacular achievements, plus nobody being willing to say the spectacular achievements signify anything broader.”

4. E4 Five More Years Slate Star Codex · 15 February 2018 Own record

   “1. Trump wins 2020: 20%”

5. E5 Five More Years Slate Star Codex · 15 February 2018 Own record

   “Cryptocurrency will neither collapse nor take over everything. It will become integrated into the existing system and regulated to the point of uselessness.”

The 2018 post "Five More Years" opens by saying that much of what changes is hard to capture in objective probabilistic predictions (E3). It contains prose forecasts with no probability or resolution criteria, on AI (E4) and on cryptocurrency (E6). These sit alongside numbered predictions with probabilities, such as "Trump wins 2020: 20%" (E5). This meets the 'true' criterion: some loosely worded guesses appear alongside the usual clearly resolvable probabilistic predictions. The verdict rests on the post itself, which is the subject's own published record.