eider abacus 🛡️ Profile
eider abacus 🛡️

@abecedarius

Followers
541
Following
45K
Media
38
Statuses
8K

Darius Bacon

Austin
Joined October 2008
Don't wanna be here? Send us removal request.
@abecedarius
eider abacus 🛡️
2 days
Going through family albums, found self in goblin pose ⁦@goblinodds
Tweet media one
2
1
20
@abecedarius
eider abacus 🛡️
2 months
RT @uberstuber: i judge people based on how they treat waiters and how they treat aella.
0
25
0
@abecedarius
eider abacus 🛡️
8 months
RT @steve47285: Lots of people think human extinction from future Artificial General Intelligence is a stupid thing to worry about. (I thin….
0
5
0
@abecedarius
eider abacus 🛡️
9 months
RT @tracewoodgrains: and yes this essay is an extended form of this but dangit I stayed in the coalition of a candidate I never wanted and….
0
14
0
@abecedarius
eider abacus 🛡️
11 months
RT @regretmaximizer: governments should just have a news outlet where they only publish stuff that's true that they think the public needs….
0
3
0
@abecedarius
eider abacus 🛡️
2 years
(Not sure how codeforces participants compare to working programmers at large, but at first glance it looks legit.).
0
0
0
@abecedarius
eider abacus 🛡️
2 years
This appears to fulfill my 2021 comment that likely in 5 years AI would beat 90% of working programmers at new leetcode-type problems. (Was responding to HN discussion downplaying the new Copilot.) Several years early, as seems typical these days.
@stse
stephen tse 🎏defi wealth 🤟robot fingers ➗crypto
2 years
game over. "AlphaCode 2 sits at the 85th percentile on average – ranking just between the ‘Expert’ and ‘Candidate Master’ categories on Codeforces. In the two contests where it performs best, AlphaCode 2 outperforms more than 99.5% of competition participants!
Tweet media one
1
0
1
@abecedarius
eider abacus 🛡️
2 years
RT @TheZvi: Further amplifying since it is last day for public comments: It is vital that we do not let the FDA regulate lab tests, lest t….
marginalrevolution.com
I have been warning about the FDA’s power grab over lab developed tests. Lab developed tests have never been FDA regulated except briefly during the pandemic emergency when such regulation led to...
0
11
0
@abecedarius
eider abacus 🛡️
2 years
RT @TheZvi: Herein I present, as best I can make out, what happened. And indeed is still happening, this is far from over. .
0
11
0
@abecedarius
eider abacus 🛡️
2 years
RT @RichardMCNgo: I just learned that the RTS,S malaria vaccine spent *23 years* in trials, including a 6-year delay caused by the WHO *eve….
0
91
0
@abecedarius
eider abacus 🛡️
2 years
RT @almostlikethat: Look, when Andreessen wrote in 2011 that "software is eating the world" it was a fully general argument. There's no sp….
0
10
0
@abecedarius
eider abacus 🛡️
2 years
This hits me hard: "appeared to produce clean margins" is exactly what I was told after my own cancer surgery last year. As he says, we need a much stronger “right to try” presumption.
@seligerj
Jake Seliger
2 years
0
2
7
@abecedarius
eider abacus 🛡️
2 years
It's funny, I don't want to disappoint GPT-4. (But it's not often good at names, I must say.)
Tweet media one
0
0
2
@abecedarius
eider abacus 🛡️
2 years
Source for the second sonnet:
@mtyka
Mike Tyka (@[email protected])
6 years
Stanisław Lem's "The first sally (A), or Trurl's electronic bard" (part of "The Cyberiad") perfectly anticipated in 1965 what it feels like to work with generative neural nets.
Tweet media one
Tweet media two
Tweet media three
1
0
3
@abecedarius
eider abacus 🛡️
2 years
Bagatelles to GPT-4, still remarkable to me.
Tweet media one
Tweet media two
1
0
4
@abecedarius
eider abacus 🛡️
2 years
The language it was learning: and my own version of the code it was trying to write in the second half:
github.com
A programming argot. Contribute to darius/cant development by creating an account on GitHub.
0
0
0
@abecedarius
eider abacus 🛡️
2 years
Previous try with GPT-3.5 for comparison:
@abecedarius
eider abacus 🛡️
3 years
Teaching ChatGPT about a Lisp dialect it hasn't seen before, then asking it to code up symbolic differentiation. That was the first Lisp program I ever read, in 1981, and though it took a lot of coaxing, this feels pretty science-fictional.
1
0
0
@abecedarius
eider abacus 🛡️
2 years
After dispatching the familiar symbolic-differentiation problem with only one or two errors, this time we proceeded to writing an interpreter for the new language in itself. This would still be *kind* of familiar in that it's a Lisp, but all the details were new.
1
0
0