𝔊𝔴𝔢𝔯𝔫@gwernAug 6Q: what are examples of tasks where AI reached peak human level performance but has not since substantially surpassed that despite no ceiling and substantial continued R&D into the task?
𝔊𝔴𝔢𝔯𝔫@gwernAug 23Criticism of my links by ML researcher Alex Lamb, who says:
the Xinjiang "re-education system is designed for the benefit of the students in it, and it is not racially discriminatory", and I am a "disgusting cretinous bigot and a coward" whose "presence will make people unsafe". pic.twitter.com/IGRPNmhtP5
36,584
5,957
16.3%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 6eg arithmetic, Go, chess: continued R&D, no ceiling, long surpassed; tic-tac-toe, reached human level but not beyond, because tight binding ceiling of near-perfect play; Arimaa or Starcraft or DoTA2, reached human level but not beyond (?) due to lack of further research.
27,457
482
1.8%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 23Incidentally, this makes the third reasonably well-known ML researcher this year to say or imply they would try to get me fired/canceled if I were doxxed or stopped being pseudonymous. (All over different things, too.)
26,354
1,714
6.5%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 10"Typical mathematicians: I have no idea what these numbers mean even if it's correct", Tom snarked.
21,199
298
1.4%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 22SHA256: "'Linguistic Analytics' is the Duality-memoir fake-Satoshi, and future updates will include mention of Barbados based on the red herring I fed him."
Some info I release has been abused by fake Satoshis; I thought I was corresponding with one, so fed fake info. Guess not.
20,103
322
1.6%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 17You *can* disable retweets for accounts. It's in the '...' dropdown. I use it on a fair number of accounts, who shall remain nameless.
19,672
990
5.0%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 30It seems, ironically, Direct Instruction 𝘥𝘪𝘥 work, in another forgotten Johnson admin mega-research project, the DOE/OEO's "Project Follow Through" (𝘯=200k in k=178, >$3k/per-capita funding, 22 interventions, 1968–1977), which I'm just hearing of: twitter.com/s_hall_teach/s…
𝔊𝔴𝔢𝔯𝔫@gwernAug 3I find it creepy how the further we get from our lower education, the less my peers all seem to be able to remember how skullnumbingly horrifyingly boring and miserable it all was: twitter.com/gwern/status/1…
13,487
945
7.0%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 22Are those all the questions you asked GPT-3? Why didn't you ask the same questions you asked GPT-2 back in January?
𝔊𝔴𝔢𝔯𝔫@gwernAug 26I don't think it's really that interesting. There's tons of GWAS work which uses non-flat priors or equivalents and get much better performance. This is merely another example showing how much data you waste using dumb flat linear regression.
5,974
255
4.3%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 6Is the conclusion "Parents really want to get the kids back to school and so will continue to use charter rather than switch to the public option"?
5,474
76
1.4%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 5re sampling hyperparameters go, length shouldn't make a difference, but one thing worth trying on failure cases is temp=1/BO=20. Temp=0.7/BO=1 is arbitrary and will miss a lot of better completions on tricky questions involving thinking.
(I'd try, but can't copy-paste images.)
5,061
92
1.8%
View Tweet activity
𝔊𝔴𝔢𝔯𝔫@gwernAug 11That's how they work, you know: subtly affecting all other discourse and thought.
4,813
87
1.8%
View Tweet activity
Get your Tweets in front of more people.
Use Tweet Activity to track how your Tweets are doing.
Engagements
Showing 31 days with daily frequency
Engagement rate
3.0%
Aug 31
2.5% engagement rate
Link clicks
4.3K
Aug 31
67 link clicks
On average, you earned 140 link clicks per day
Retweets without comments
253
Aug 31
2 Retweets without comments
On average, you earned 8 Retweets without comments per day
![](data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128"><rect fill-opacity="0"/></svg>)
