During a discussion on one of the various LessWrong-aligned Telegram chat groups, I noticed how vague the "short" in "short timelines" is, and ran a poll to find out what people mean when they say "short timelines".
%
Range
Votes
0%
<3 months
0
0%
3–6 months
0
33%
6–12 months
8
33%
1–2 years
8
21%
2–4 years
5
8%
4–8 years
2
4%
Something more than 8 years
1
Total: 24 votes
Small n, certainly. But with this broad a distribution even on a time axis that doubles with each line, I think it shows we need to be precise about "how short is short" when discussing how likely we think a "short timeline" is.
It's potentially even worse than this: in the chat group, one respondent wasn't sure which answer to give because their idea of "short timelines" would be a different time bracket for "human level AGI available on the market" vs. "AI running 50+% of knowledge economy".
No, I don't think we can agree on a translation to X years. "Short" is often deliberately vague and relative to an implicit Overton window, and as you point out, "timelines" (to what?) is also underspecified. If we want to be more specific, we should be more specific (e.g., "human expert dominating model publicly available by 2028") rather than try to force vague terms to not be vague.
During a discussion on one of the various LessWrong-aligned Telegram chat groups, I noticed how vague the "short" in "short timelines" is, and ran a poll to find out what people mean when they say "short timelines".
Total: 24 votes
Small n, certainly. But with this broad a distribution even on a time axis that doubles with each line, I think it shows we need to be precise about "how short is short" when discussing how likely we think a "short timeline" is.
It's potentially even worse than this: in the chat group, one respondent wasn't sure which answer to give because their idea of "short timelines" would be a different time bracket for "human level AGI available on the market" vs. "AI running 50+% of knowledge economy".