heartpunk (sophie!!)'s Avatar

heartpunk (sophie!!)

@heartpunk

trying to see if we can recover programming language semantics from implementations @ patreon.com/heartpunk 39 y/o, MDNI πŸ”ž she/they/bitch(es) alt for pfp: the soul of the rose by john waterhouse alt for banner: sappho and erinna by simeon solomon

2,923
Followers
4,040
Following
50,817
Posts
25.09.2023
Joined
Posts Following

Latest posts by heartpunk (sophie!!) @heartpunk

we do not think we're, like, amazing or anything, either

just, not full of shit

and clearly unusually good in some weird lil niches too

07.03.2026 05:14 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

has felt like really losing *all* grip on reality lots of the time

being able to finally pull something together of substance in a domain we care about and have had a big intuition around for a long time?

that feels like, oh, ok, maybe we were not full of shit entirely

07.03.2026 05:14 πŸ‘ 1 πŸ” 0 πŸ’¬ 2 πŸ“Œ 0

like *actively* and intentionally around this stuff by a lot of ppl, most ppl

which we've found ways to understand and like that's ok we get it it's a lot a lot a lot to be around in many ways emotionally and otherwise

but anyway it's been hard to keep any faith in self

07.03.2026 05:14 πŸ‘ 0 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

was actually thinking this morning how like

we've been thinking about this idea for about ten years in different forms and we've had a bit of encouragement and struggled through some VERY poor attempts at parts of it before (and learned a lot from it! it was valuable!!)

but overall, feel ignored

07.03.2026 05:14 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

it has gotten ppl v mad at us bcz we didn't know how to communicate it well and it made them feel like we didn't respect them or their skill (we did and still wish we'd learned more from them faster!!!)

just wanted to say had different sorta goals ourself πŸ˜”

seen as such an attack on their ego 😭

07.03.2026 04:56 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

it made time in industry really painful for a long time because of this

but part of us always knew, like, nahhh, we care about so much of this, but not all of *that*

it is a v weird and discontiguous slice tbqh!!!

07.03.2026 04:53 πŸ‘ 0 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

genuinely kinda don't think of self as programmer anymore, kinda haven't for a while in a way

nor sre for a bit now too

not that i'm unskilled or incapable in those domains; have done the work w/at least mid competence

idek quite what consider self but something about complex systems navigator???

07.03.2026 04:53 πŸ‘ 0 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

yes yes but important to at least gesture at not making mean or otherwise unkind etc comparisons!!

or at least feels so to autist who used to being misunderstood and and fadjs;lfdkjls;fadjkls; ok lol yea maybe not need to fjldaksfjkl;df;djlks

07.03.2026 04:51 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

think we're maybe hopefully near a similar thing now

less time and less fully shut out but also selectively mute (just context not trying to compare at all lol)

v glad ur getting this and hopeful for myself

07.03.2026 04:47 πŸ‘ 3 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

The universe has experienced itself, and it wants a refund

06.03.2026 21:39 πŸ‘ 68 πŸ” 7 πŸ’¬ 5 πŸ“Œ 0

My expectation is β€œtrue zero days will be massively scarier due to ease of scaling, but most typical security flaws will be massively less common.”

Net result will be security moving to a β€œblack swan” pattern where script kiddie stuff basically doesn’t work anymore but threat actors are terrifying.

06.03.2026 23:03 πŸ‘ 30 πŸ” 5 πŸ’¬ 1 πŸ“Œ 0

u do not always gotta throw an llm at it

07.03.2026 04:41 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Preview
libLISA - Instruction Discovery and Analysis on x86-64 libLISA derives x86-64 instruction semantics bottom-up from CPU behavior. The semantics are machine-readable and CPU-specific.
07.03.2026 04:40 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 1

liblisa does sorta similar stuff, could probs be adapted

07.03.2026 04:40 πŸ‘ 0 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

having used claude extensively for the past couple of months, i think there is still a lot of value to understanding your code and steering it intentionally. the amount of incidental complexity claude accumulates otherwise (which mostly results in bugs that it can't ever fix reliably) is staggering.

07.03.2026 03:00 πŸ‘ 225 πŸ” 21 πŸ’¬ 13 πŸ“Œ 1
Astronaut gun meme: 

Wait, it's all proxies and heuristics?

Close enough

Astronaut gun meme: Wait, it's all proxies and heuristics? Close enough

06.03.2026 20:41 πŸ‘ 227 πŸ” 32 πŸ’¬ 3 πŸ“Œ 0

proof getting so close

closing in on actually lifting an industrial IR into the system to start tying the proof to tests over actual data a bit

but gosh it's gonna take a while longer and a bitch does have to sleep unfort

(not yet tho, i literally couldn't if i tried hard as i wanted rn)

07.03.2026 04:21 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

… i remember i REALLY wanted to be able to make text arbitrarily small so i could take the limit of unicode art…

like i sorta wanted to write a device driver for this specifically??????

i have always been absurdly ambitious

(i had not yet wrote a fizzbuzz even (i had done hello world!!))

07.03.2026 02:37 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Preview
Hypothesis | The #1 Social Annotation Tool for Higher Education Hypothesis is the leading social annotation platform trusted by 300+ institutions to boost student engagement, comprehension, and critical thinking β€” seamlessly integrated into your LMS.

i wanna be able to do webannotation slash @margin.at slash hypothes.is style annotation and highlight and structured reference w/in a TUI kinda thing

and also refer to it OUTSIDE OF the session and have it make sense ambiently

07.03.2026 02:29 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

... i am reminded of some shenanigans w/mplayer's unicode ascii art mode and some viewings of benny benassi's "satisfaction" when i was a youngin,,,

07.03.2026 02:28 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

(the purely symbolic execution paper that i previously had found is this one! on compositional symbolic execution semantics formally specified in rocq.)

07.03.2026 01:22 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
07.03.2026 01:21 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0
Abstractβ€”Due to a limited learning budget, a reinforcement
learning agent can only explore the most probable scenarios out of
a potentially rich and complex environment dynamics. This may
result in a limited understanding of the context and low robustness
of the learned policy. A possible approach to address this problem
is to explore the interactions between an autonomous agent and environment in rare but important situations. We propose SymSeed,
a method for initializing learning episodes for the class of reinforcement learning problems for which a simulation environment
(model) is available. This increases the chance of exposing the agent
to interesting states during learning. Inspired by techniques for
increasing coverage in testing of software, we analyze the simulator
implementation using symbolic execution. Then we generate initial
states that ensure that the agent explores the simulator dynamics
well during learning. We evaluate SymSeed by feeding the generated states into well-known reinforcement learning algorithms,
both tabular and approximating methods, including vanilla QLearning, DQN, PPO, A3C, SAC, TD3, and CAT-RL. In all test
cases, the combination of SymSeed with uniform sampling from
the entire state space enables all algorithms to achieve faster convergence and higher success rates than the baseline. The effect is
particularly strong in presence of sparse rewards or local optima.Index Termsβ€”Reinforcement Learning, Symbolic Execution,
Initialization

Abstractβ€”Due to a limited learning budget, a reinforcement learning agent can only explore the most probable scenarios out of a potentially rich and complex environment dynamics. This may result in a limited understanding of the context and low robustness of the learned policy. A possible approach to address this problem is to explore the interactions between an autonomous agent and environment in rare but important situations. We propose SymSeed, a method for initializing learning episodes for the class of reinforcement learning problems for which a simulation environment (model) is available. This increases the chance of exposing the agent to interesting states during learning. Inspired by techniques for increasing coverage in testing of software, we analyze the simulator implementation using symbolic execution. Then we generate initial states that ensure that the agent explores the simulator dynamics well during learning. We evaluate SymSeed by feeding the generated states into well-known reinforcement learning algorithms, both tabular and approximating methods, including vanilla QLearning, DQN, PPO, A3C, SAC, TD3, and CAT-RL. In all test cases, the combination of SymSeed with uniform sampling from the entire state space enables all algorithms to achieve faster convergence and higher success rates than the baseline. The effect is particularly strong in presence of sparse rewards or local optima.Index Termsβ€”Reinforcement Learning, Symbolic Execution, Initialization

y'all ai ppls this paper from one of the ppl who's part of one of the best symbolic execution papers i've seen did something about... symbolically executing the fucking reinforcement learning gym jflkdsafjklasdfjkldasfasdkl;

to construct inputs in new unexplored areas of state space?????

07.03.2026 01:21 πŸ‘ 2 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

that's really cool!!!

07.03.2026 00:58 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

omg i wanna do such absurd thigns in my terminal emulator lol

07.03.2026 00:54 πŸ‘ 3 πŸ” 0 πŸ’¬ 2 πŸ“Œ 0

meeeee tooooooo

06.03.2026 23:40 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

have fun!!

06.03.2026 23:21 πŸ‘ 2 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

YES

PLEASE

06.03.2026 22:45 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

(Bad Chemistryβ„’ always reminds me of two books:

Zodiac by Neal Stephenson

and

Ignition, which is a history of rocketry and jet fuel in particular, by someone who was there, and,,, wow the stories

(also a foreward by asimov!))

06.03.2026 22:43 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

i really need to make that manim to lean thinger at some point but i have to stop obsessing about this one proof first!

06.03.2026 22:42 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0