Michael Sullivan

1.7K posts

Michael Sullivan

Michael Sullivan

@msully4321

I think about programming languages and GM tabletop RPGs. Working at Vercel. Core dev (emeritus?) on @mypyproject. he/him

Katılım Şubat 2011
474 Takip Edilen368 Takipçiler
Nick Levine
Nick Levine@status_effects·
New work with @AlecRad and @DavidDuvenaud: Have you ever dreamed of talking to someone from the past? Introducing talkie, a 13B model trained only on pre-1931 text. Vintage models should help us to understand how LMs generalize (e.g., can we teach talkie to code?). Thread:
English
174
380
3K
1.1M
Michael Sullivan
Michael Sullivan@msully4321·
@gro_tsen Though I also kind of doubt that they'd have succeeded all *that* much better if given a working setting. (They could probably do alright if given full reign to write code to generate the unlambda, though, and maybe some prompting to hint at it? I might try that out.)
English
0
0
2
17
Michael Sullivan
Michael Sullivan@msully4321·
@gro_tsen I took a look at their code, though, and they were using their own unlambda interpreter that supported about a third of the language, but not input! So none of the problems were really possible in their setting.
English
1
0
2
17
Gro-Tsen
Gro-Tsen@gro_tsen·
We finally have a killer-feature of the human brain that LLMs are unable to match: programming in Unlambda. I sense a great opening for engineers.
Lossfunk@lossfunk

2/ Our method: test them on esoteric programming languages. Brainfuck. Befunge-98. Whitespace. Unlambda. Shakespeare. All Turing-complete. All requiring identical reasoning to Python. All with 1,000-100,000x fewer GitHub repos than mainstream languages. Same problems. Radically less training data.

English
2
1
11
1.3K
Michael Sullivan retweetledi
Armand Domalewski
Armand Domalewski@ArmandDoma·
The most important thing San Francisco politicians can do to resist Trump is get our own house in order. San Francisco should be a promise of what America can aspire to be, not a cautionary tale for what it must avoid becoming
English
89
159
3K
96.2K
argon
argon@ArgonGruber·
@ericneyman like, this chart says: P(Biden)=60%, P(Newsom)=20%, P(Harris)=10%. how do you get from there to P(Newsom|not Biden)=50%? you can't get there by applying Bayes Theorem, i think.
English
1
0
0
174
Michael Sullivan retweetledi
rntz
rntz@arntzenius·
I'm on the market for a postdoc position! (RTs appreciated!) After two years in industry making data-oriented logic programming practical, I'm hungry to get back to basic research & I've got ideas crawling out my ears. email in bio; research ideas in 🧵
English
1
18
35
4.5K
Michael Sullivan
Michael Sullivan@msully4321·
@lastpositivist "The High Elf's Burden" holds a prominent place on my List of RPG Campaign Ideas I Will Never Actually Run
English
0
0
0
15
shachaf
shachaf@shachaf·
If you write atomic_fetch_or in C or C++ on x86, gcc/clang will compile it to "lock or". But if you look at the return value, it'll switch to a cmpxchg loop! Kind of spooky magic behavior.
English
1
0
1
717
Michael Sullivan retweetledi
Yury Selivanov
Yury Selivanov@1st1·
Who needs in-flight entertainment when you can...idk, port uvloop to Python 3.12...
Yury Selivanov tweet media
English
7
5
132
33.3K
Michael Sullivan
Michael Sullivan@msully4321·
@iquilezles Is using a slide rule allowed? Or log tables? (.... probably that will only make things more complicated and less correct.)
English
0
0
1
118
inigo quilez
inigo quilez@iquilezles·
Hello, I present you with the Human Shader, a project where anybody can participate and help create the first ever human-brain-only powered shader and mathematical image! Please join us and be part of history! humanshader.com
inigo quilez tweet media
English
98
601
2.6K
582.3K
Michael Sullivan retweetledi
Gel
Gel@geldata·
The EdgeDB 3.0 beta went live on Friday, and it includes a boatload of major features. 🚢 You can learn about all of them by reading our changelog: edgedb.com/docs/changelog… I'm here to tell you about just a few of them. More in the 🧵…
English
3
7
16
3.3K
Michael Sullivan retweetledi
Gel
Gel@geldata·
The feature list of EdgeDB 3.0 is going to be massive, and you'll get to try them all out starting… now! EdgeDB 3.0 beta is live! edgedb.com/docs/changelog… We can't wait for you to play with triggers, mutation rewrites, query planning, a redesigned UI, `edgedb watch`… and more!
English
1
9
35
4.1K
colinhacks/zod
colinhacks/zod@colinhacks·
just got this message from my mom i hope the money doesn't change me
colinhacks/zod tweet media
English
3
0
178
23.8K
Michael Sullivan
Michael Sullivan@msully4321·
I added some variants on it; probably "view_patterns_explicit.py" is the best of them but I don't love needing multiple methods on the scrutinee
English
0
0
0
108
Michael Sullivan
Michael Sullivan@msully4321·
But if you could write a helper view, it could be something like... `case alias_view(name)`
English
1
0
0
138
Michael Sullivan
Michael Sullivan@msully4321·
Python 3.10's shied away from adding customizable pattern matching with `__match__`... but I'm not going to let that stop me: gist.github.com/msullivan/7f53… (The big value prop for __match__ IMO is not in defining it on your actual data types, but in defining *helper* views!)
English
1
0
2
228

Keşfet

@status_effects @PradyuPrasad @AlecRad @DavidDuvenaud @matt_levine @gro_tsen @ArgonGruber @ericneyman