Aditya Kompella

[Aditya Kompella @sakompella]

@RigoLigoRLC @BrodieOnLinux its doing fine, theyve gotten m3 booting since

[Aditya Kompella @sakompella]

@pwnies @beffjezos it's entirely correct, just not nice to hear spelt so unkindly of appearantly

[Aditya Kompella @sakompella]

@mrcupidxo @pwnies @beffjezos beff has personally advocated for successionism

[Aditya Kompella @sakompella]

@DanielSmolsky @opencode :)

[Aditya Kompella @sakompella]

@matthewktromp @TetraspaceWest have you read scott alexander's "against SURA" (and the associated responses people made?) I think there's a very strong argument made that "firebombing the walmart" wouldn't be very effective for even the shortterm, and detrimental for the longterm

[Aditya Kompella @sakompella]

@0xleelu @forloopcodes haiku is just RL trained for being a subagent similar to luna

[Aditya Kompella @sakompella]

@viemccoy @Sam_McCandlish @METR_Evals what is the plan at openai if not similar? i have always understood oai's and ant's endgoals to be extremely similar ideologically

[Aditya Kompella @sakompella]

@xsphi crate2nix also will work

[Aditya Kompella @sakompella]

@xsphi nix is probably not what you want for this; kind of just a different thing but may be of interest. see https://t.co/FxoLc9mJvk

[Aditya Kompella @sakompella]

@xsphi buck2...? nix can be used to cache at the crate level at least

[Aditya Kompella @sakompella]

@ccccjjjjeeee orbstack / linux vm

[Aditya Kompella @sakompella]

@neoraed @Nikuw2137 consider:
no AI -> good output
AI; 10x faster -> sloppy output
AI + strong effort + 2-3x fast -> good/great output
also notable is that much reverse engineering is a lot of writing throwaway code to understand the system; being able to that fast / sloppily is a big boon for free

[Aditya Kompella @sakompella]

@DeepDishEnjoyer artificial analysis v5

[Aditya Kompella @sakompella]

@hecubian_devil seems to me mechinterp is in some ways capabilities research https://t.co/JG3Bw2M27f

[Matthew Tromp ⏹️ @matthewktromp]

AI "safety" research has two uses:
- Making AI more powerful and dangerous
- Giving people who work on AI an excuse to make it more powerful and dangerous ("look at all these safety breakthroughs! Alignment will totally be solved in time. Everything will be okay")

[Aditya Kompella @sakompella]

@stupidtechtakes we had like 10 critical CVEs in the last week

[Aditya Kompella @sakompella]

@thorstenball I see, thanks!!

[Aditya Kompella @sakompella]

@thorstenball I assume "thinking in systems, a primer"? First thing that popped up
Any other cases in particular?
Also --- why business/consulting slant, considering you operate a saas startup, that (I maybe misreading) is pretty consumer focused?
Also typo 💔 I meant "I appreciate this"