@hackerdocc

0x1A yo ; verifying things, for reasons... ; organized TYPED Floripa ; prev SRE @bookingcom and Rust & cryptography @opactprotocol

 Floripa
hackerdo.cc
Joined July 2020
  • Tweets8.6K
  • Following1.4K
  • Followers858
415Media
Replying to @hackerdocc
hand-coding shit was fun, same as solving a sudoku. if you were motivated by solving logic puzzles as a day job then i understand why you feel miserable. solution? subscribe to NYT Games I realized I'm motivated by hate for software and I'll only stop once it stops sucking
1
6
3,051
L
9
303
having to manually patch my memory system cuz now my agent has the memory of formally verifying a bug and thus all my conversations are triggering cybersecurity safeguards
2
179
IRs are funny man... they have this implicit semantics in the conscious collective that compilers promise to preserve when lowering but... there's nothing actually executing most IRs...
1
3
373
cc @peter_sewell_
28
stop coding unverified software. it will be utterly useless and i say this with all the care in the world. i think we are witnessing an incipient singularity. maybe all the code will be verified within an year. if it sounds unreasonable to you, adjust expectations
7
1
28
1,134
"that's not what singularity means" well i could elaborate my point further, about how nobody can truly estimate the economic impact of buggy insecure software (which is basically everything), and how fixing that overnight would having profound ripples through scientific discovery, and how AI itself is made of software, and ofc by making software better we make AI better (we've been seeing this with e.g. autoresearch for faster kenels), but whatever, i prefer vagueposting prophetic thought-leading ai psychosis
2
96
google ai pro subscription is crazy value, i have it cuz of google drive and i also get more claude usage and a separate argon quota? (plus a bunch of other shit)
1
3
447
nvm very shitty limits for the $20 plan, and the others are way less value, not really sure what i was expecting. but honest to god it's so anti-consumer when your quota ends in a single run without an agent response, especially when the thinking is encrypted...
1
34
the most valuable lesson in interacting with agents is that when explanations aren't clear it's most likely the case that the explainer is bullshitting you
1
2
13
526
antigravity has no auto-mode? brings me back to one year ago when everyone was typing `--dangerously-skip-permissions`
114
young men's guide to playing the game in the age of AI > spend your days developing formal semantics for programming languages > go deep into debt by buying tokens for proving the correctness of critical software > save humanity from a cybersecurity collapse by developing inviolable sandboxes, containing rogue agent swarms > i guess avoiding a disaster doesn't make the headlines (anthropic bias or smth) > everybody uses your software without realizing cuz claude 1-shots regurgitating it from the training set > die poor and unacknowledged > spend eternity in limbo cuz god isn't real
𝞍 Shin Megami Boson 𝞍
@shinbosonOct 15
sharing the top reply because I thought it was gorgeously written
1
10
473
using decide instead of native_decide is quite dumb (during development, not CI): you spend way more compute and memory and even if native was ever to prove False that would be nice — you found a reproducible lean compiler bug by running CI
96
why does larp get so much engagement?
Justin Schroeder
@jpschroederOct 15
This is a Nobel-price worthy breakthrough, not exaggerating.
1
2
189
(obviously referring to the precise tweet i quoted and not cheng's thing which is ofc useful and good work, probably applicable to interpretability etc but not a scientific breakthrough in any sense?)
2
48
being right is extremely overrated, convincing people is more important and correctness has little to do with that
6
1
18
643
i'm starting to figure out how to make bait: just say smth interpretable in many ways (don't recommend it tho, too much noise) what inspired me to post this was precisely that argon 4 was already doing some stuff related to nuclear fusion, so in fact i was being disingenuous
Eduardo
@hackerdoccOct 13
why are we spending tens of millions of dollars in compute just to give mathematicians trauma instead of... idk solving nuclear fusion?
1
94
more precisely - can be taken to mean multiple things - most have to be annoying to different profiles - so it has to be at least a bit wrong or oblivious - has to be about a trending topic in which laymen have opinions about and experts are gatekeeping - can't be too insincere
27
why are we spending tens of millions of dollars in compute just to give mathematicians trauma instead of... idk solving nuclear fusion?
119
9
5
175
12,804
honestly props to them for copying the claude aesthetic
Google
@GoogleOct 13
Today, Gemini 4 Argon is officially rolling out.
143
did u know @Inria has that shitty PoW "captcha" (anubis) in front of their website? i think that's pretty unprofessional lol i accidentally refreshed the page while that was going on and know they blocked me? lol (can still use another browser etc)
2
3
250
overall i don't think serious institutions should be putting channer anime girls in front of their site...
3
35
Load more