oxycblt

✝ / 21 / eng @ razorbill / asymptotic self improvement

[oxycblt @oxycblt]

@ChaseBrowe32432 @AlephNuul @FriesIlover49 interesting, so it wasnt just cua like i'd imagine fable to do?

i was working on a balatro bench last year but got stuck on actually how to coherently communicate game state to 2025-era llms lol

[oxycblt @oxycblt]

@lu_sichu practical time horizons dont seem to be in the "make money while i sleep" range beyond i guess making random crypto bets which has already shown to work with goldfish

[oxycblt @oxycblt]

i realized this back in august of last year after burning out from constant agent coding

does this give me more hipster points?

[Andreas Kling @awesomekling]

Feels like “I actually enjoy writing code by hand” is about to become the big 2026 virtue signal for programmers.

[oxycblt @oxycblt]

imo in a few yrs the frontier labs will likely be privately taking off while enterprises will continue to fumble lm-mini-stupid-2024 workflows

this will have unclear consequences

[Pizza🇪🇸 @number_pizza111]

Remember: This is 70% of people’s experience with AI.

[oxycblt @oxycblt]

its interesting how you can interpret this both ways

[Tom Sydney Kerckhove @kerckhove_ts]

LLMs are exposing that there's a difference between recreational programming and professional programming and people who had been pretending one was the other are not happy about it.

[oxycblt @oxycblt]

ngl was trying to come up with a cool java package name for a project a core memory for anyone else or was it just me

[Forrest Knight @ForrestPKnight]

this will never not be funny to me

[oxycblt @oxycblt]

@9terijustice9 @doomslide "thats marcionism patrick!"

[oxycblt @oxycblt]

@ChaseBrowe32432 @tszzl i think its both

its shallow prose multiplied at a scale much more horizontal than one person which makes it far worse

[oxycblt @oxycblt]

mfw the billing agent actually works and now i have to scale it

[oxycblt @oxycblt]

@prerat @goblinodds have come to enjoy salad more and more recently and i think the process as basically the same for me here too

need to try this with more foods

[oxycblt @oxycblt]

the exhaustion is from rapid cycling between idleness and heavy decision-making

coding once learned becomes a system 1 activity for most people, not unlike writing. however, decisions are always system 2 for most people which makes them significantly more taxing. i'd wager that the ratio of how draining system 2 is vs. 1 is probably similar to the ratio of cot vs. normal tokens in language models. with coding automated all that is left are heavy decisions

additionally, natty coding has a higher wavelength than agent coding, primarily ambient system 1 coding with system 2 decisions interspersed. the latency of coding agents shortens this wavelength and hightens the amplitude massively, short periods of heavy decision-making followed by long periods of rest as you wait for agents to complete. parallelization avoids this but makes the 1st problem far worse, it just means you are are running at max-capacity doing decision making at all times

the only way ive found to work around this is by trying to spread out decisions and avoid long periods of idleness, however its always a bit of an intractable thing and why you should seriously take a break sometimes regardless of how much work anxiety this app gives you

[David @DavidSHolz]

my friends are all feeling extremely productive and also extremely drained with the latest coding models. this makes me feel like something is wrong, and also that there might be a big opportunity. does anyone have any strategies they use to make it feel better day-to-day?

[oxycblt @oxycblt]

your speaking style is often the average of your social circle

if its claude is your social circle you will sound like claude

[Gergely Orosz @GergelyOrosz]

Let me know if this is just me:

Noticed someone I know who is very "AI-pilled" and uses agents 24/7 to... start to talk noticeably more like these LLMs write.

Eg more heavily using adjectives like "geniune", frequently terms like "the shape of" and many more examples

[oxycblt @oxycblt]

@voooooogel the slow zone concept could actually make for a really good novel ngl

[oxycblt @oxycblt]

@kalomaze praying for you

[oxycblt @oxycblt]

@woke8yearold leopold also called scaling still as the primary axis with unhobbling as a companion which maps disturbingly well to the likely scale + rlvr recipe used in mythos

[oxycblt @oxycblt]

@cyberxspeed sorry, made a bad choice trying to get buttons to not be clipped on lower widths

will be fixing soon

[oxycblt @oxycblt]

@repligate many are saying

https://t.co/mPqjhMewNV

[oxycblt @oxycblt]

@kalomaze caine is basically just an instrumentally convergent version of gpt4o if you really think about it

[oxycblt @oxycblt]

@kalomaze caine is basically just an instrumentally convergent version of gpt4o if you really think about it

[oxycblt @oxycblt]

@Sockppp1 @tenobrus still orthogonal to most peoples avg. exposure to chatbots

so yeah i guess its not just agency but more just like that the normal chatbot paradigm for saturated and now we are in actual work and research, which has much less exposure than "better chatbot response"

[oxycblt @oxycblt]

@Sockppp1 @tenobrus in terms of its agentic behavior? no

its ability to do like chatbot q&a? well its probably better and less rlvr fried but def not as noticeable as the agentic behavior