reasons to be cheerful, part 3
Joined September 2015
- Tweets 3,272
- Following 840
- Followers 395
- Likes 55,450
136 Photos and videos
Jun 12
through no fault of my own, I have angered the neighborhood crows
1
4
165
Jun 10
Impressions of 5.5 VS Fable for math research. Fable is like a very well-read but verbose and handwavy PhD student, 5.5 is more concrete and down-to-earth. Overall Fable seems a bit "smarter" overall. This is a noticable change from previous models where GPT was clearly better.
3
2
20
2,018
Jun 11
update: Fable can in fact oneshot(well, mostly) the result referenced here. this thing is a beast. x.com/an_interstice/status/2…
a bit faster than I expected. welp, guess I'm glad I got it on the arxiv a month ago for signalling value.
May 15
annnd wrote it up. mixed feelings since (a) GPT6.5 will probably be able to oneshot it in a few months (b) it was a distraction from more directly AI-relevant work, but I'm still pretty happy with the result.
5
186
Jun 4
it's been two whole weeks and no dyson sphere, deep learning is hitting a wall
May 20
<trying to revise timelines such that I'm surprised upwards as much as downwards>so dyson sphere by.....tuesday?
2
426
interstice retweeted
Jun 3
Our new campaign in Canada, backed by over 30 MPs and Senators, is cause for real optimism.
For the first time, a cross-party coalition of Canadian lawmakers is calling for an international ban on developing superintelligence, recognizing the extinction risk it poses.
🧵
Jun 3
🚨NEW: We’ve just launched our campaign in Canada!
A cross-party coalition of over 30 MPs and Senators are calling for Canada to negotiate an international prohibition on the development of superintelligence, recognizing the risk of human extinction posed by the technology.
🧵
4
12
44
9,285
interstice retweeted
please retweet this and @ anyone you know in a position to reverse the recent "ban random small accounts with years of good posts" policy
19
17
98
9,959
May 20
<trying to revise timelines such that I'm surprised upwards as much as downwards>so dyson sphere by.....tuesday?
2
7
559
May 20
x.com/an_interstice/status/1…"just fit population size to a hyperbola" seems to be more plausible than it has any right to be at predicting the date of the singularity
20 Dec 2024
3
254
interstice retweeted
May 15
so there's this inscruable sequence of numbers. it turns out it's generated by an automaton. but the automaton is also inscrutable. but the automaton has a symmetry group, and *that* thing is scrutable.
2
3
32
977
May 16
starting a pseudo-reading group. every month we will pick an author and read works falsely attributed to them in antiquity
1
6
379
May 15
annnd wrote it up. mixed feelings since (a) GPT6.5 will probably be able to oneshot it in a few months (b) it was a distraction from more directly AI-relevant work, but I'm still pretty happy with the result.
2
22
1,483
May 15
half the people formerly researching the area work in industry now so I'm harassing them on linkedin with my arxiv links
1
1
8
508
May 15
now that that's done back to more practical endeavors like creating a mathematical pseudo-religion for AIs
1
1
9
444
May 10
proved a nontrivial theorem for the first time ever plz clap
35
151
4,374
63,442
interstice retweeted
May 8
yup, we are still taking off
80% METR time horizons for Claude Mythos Preview:
3 hours 6 minutes
May 8
Claude Mythos Preview's METR time horizons AT LEAST 16 hours
confidence interval: 8hrs 29 mins - 2 days 7 hours
but measurements are unreliable due low number of long-horizon tasks
21
47
820
91,128
interstice retweeted
openai․com was once the personal homepage of a guy named glenn (2001), and tiktok․com was the quaint shared homepage of a couple who went from dating to married with a baby (~1998-2001). THREAD!!!
67
2,791
23,384
769,711