thoa

doin gs tuff like working on usedul project for people to enjoy when done

[thoa @UltraSuperSilly]

@thsottiaux got this message using chat mode with pro model. if this means i ran out of pro model usage or something then this message makes no sense since i already have pro. @OpenAI

[thoa @UltraSuperSilly]

ive been thinking alot recently trying to figure out and come to terms with how my brain works and its hard because of the stigma around comparing to others.

any time i compare to others i get a stab of guilt like its something i shouldnt be doing, part of it is because i think part of my central belief around how i function is that i am hyper meta-cognitive which causes alot of trouble for me. i wouldnt say its positive or negative but when comparing to others it feels like something i have that they dont which structurally feels like im putting others down.

another thing is that being “smart” is weird to me. every person ive met who talks about how smart they are or thinks they are part of some greater group of ”smart” people kills me and i cant stand them. brainwashed themselves into thinking they are better than others because they studied enough when they are retarded. also contributes to the guilt i feel when i think im different than others. do i really think im special or different than the people i meet to have the privilege to think that way. am i naive to believe im different

i think a big chunk of it is the understanding that there is no real measurement of being “better” than anyone. being a pessimistic overthinker doesnt make you different than a glib retard.

are you better than them though? probably. is letting your thoughts about your thoughts eat you alive good for you? maybe at some points

are you who you are because of how you think or do you think how you think because of who you are?

theres some axis that i cant place my finger on that i value alot in people

[thoa @UltraSuperSilly]

just got downgraded with claude in back to back messages lmao opus 5 -> opus 4.8 -> sonnet 4.6

[thoa @UltraSuperSilly]

@usr_bin_roygbiv i feel like more context would be good for planning etc on certain projects. when im planning milestones context gets compacted usually at least once.

[thoa @UltraSuperSilly]

long path to somewhere

[thoa @UltraSuperSilly]

hey im working on a crafting solver which can currently find a exact optimal strategy for almost any non influenced craft(just havent done awakener orb etc yet which ill do when i do recombs which also need multi item) (some 5 and 6 mod items have alot of mods so states explode and it gets really hard to solve but im still improving on this aspect)

i was thinking of making my own website for it but the engine itself works outside of the current web app i have setup (which also has a emulator and simulator like craft of exile but like 1000x faster, and also has a calculator for exact odds for strategys) but i was planning on looking at if any established tools would be interested in having it incorporated into theirs

solver/emulator/simulator is ~80k lines of c++

would be interested in chatting if you have the time and sorry for lots of (parantheses)

[thoa @UltraSuperSilly]

feeling like ive hit a wall with what current models are capable even harder than a few weeks ago. for example codex keeps resorting to trying to simplify the goal of our solver instead of improving it which ends up faking “better” results at the cost of solution accuracy.

at this point i need to give it 0 freedom of what test cases to run at all since even simple instructions get abused.

for example our solver filters down 16k+ actions based on what is applicable for the specific case, a week or two ago i got it to make a bunch of 4-5 goal test cases for a specific issue we where having where solution compilation and verification was taking 5-10+ min and often failing. during this it manually selected only ~5 actions to be used in these benchmarks. which technically would have been ok for the specific case for iteration speed as long as full action suite runs where also done. it didnt tell me any of this or document this and didnt run full action suite runs.

We have been reusing this set of benchmarks for all of our optimization work in these past week or two and only after i was getting concerned that the solver didnt seem to explore most valuable actions did it tell me this.

Around the same time it fully excluded multiple large categories of actions in the native and wasm portion of the solver without documentation because they where causing it issues(instead of fixing these issues)

The most frustrating part is every issue like this is resolved by me noticing that something is off usually a few days later.

It is unable to find these kinds of issues even when i tell it directly that i found an issue like i mentioned before, and to look for other mistakes that its made that are similar.

Part of the blame might be on me but i think this is going to be a recurring issue for anyone working on novel problems with large codebases in abstract areas.

[thoa @UltraSuperSilly]

@yacineMTB notice how the three subjects are US, China, AI. sounds to me like US and China are going to be teaming up against the ai overlords

[thoa @UltraSuperSilly]

@mil000 @grok please look at the picture and find out what is worth flipping. after that please purchase and sell those worth flipping, and mail me the profits in a envelope to my p.o box

[thoa @UltraSuperSilly]

#philosophy #pondering #thinking

[thoa @UltraSuperSilly]

@boneGPT thank you for the free startup idea. they will be seeing machine elves soon. or i suppose they would see biological elves or something

[thoa @UltraSuperSilly]

i hate when irl people bring up how “ai” is “bad” because it will take peoples “jobs” and people will have nothing to do. maybe those people with “jobs” should look into doing things instead of getting really good at circularly emailing eachother

[thoa @UltraSuperSilly]

may not be alot for some people, but this was yesterday from sol ultra/xhigh working on abstraction refinement and policy verification for some ssp mdp stuff. very slow progress today but it looks somewhat close to getting things working again

[thoa @UltraSuperSilly]

reset limit and got it to continue and it finished, and it managed to break at least a few things at some point and now sol xhigh is 3 hours into testing trying to figure out what the issue is

[thoa @UltraSuperSilly]

had some shit running on ultra last night. asked before i went to bed how much it had left to do. it said it was like 70-90% done and at that point had been running for 2 hours ish. woke up and it ran for 12 hours and cooked my home limit lmao

[thoa @UltraSuperSilly]

is it worth using ultra for planning?

[thoa @UltraSuperSilly]

was going to watch gifs earlier this evening for a few hours as a treat.
night ruined when discord embeds broke. how am i supposed to be productive someone is out to get me

[thoa @UltraSuperSilly]

@lyc_aon i have never EVER looked at my bank account or any representation of how much money have and i NEVER will.

MY cash reserves are UNKNOWN and IMMUNE to mobile BANKING.

if i ever need “reserves” whatever i need will be there or it wasnt meant to be

[thoa @UltraSuperSilly]

something something moving forward. unstoppable. going to make it?