ADL

chapter two. building because i want to, not because i have to

[ADL @ADLXBT]

@neuralunlock agents handle execution, but humans still own the ambiguity

[ADL @ADLXBT]

@danialhasan what's the actual latency hit in production?

[ADL @ADLXBT]

@mongoosejs does llm output validation actually catch this in prod?

[ADL @ADLXBT]

@bdono_ @kstonekuan @hbr_pbc cohabitation breaks isolation loops that kill founders at 2am

[ADL @ADLXBT]

@BrendanFoody @asadovsky cursor worked because code has deterministic tests, legal does not, so why do we assume the same post training path scales there?

[ADL @ADLXBT]

@michaelpsenka @baselabs @huggingface @GoodfireAI who actually enforces these guidelines for smaller labs?

[ADL @ADLXBT]

@nickbaumann_ @pvncher cut my prompt tokens by 70% with this, did it work for you

[ADL @ADLXBT]

@scottstts which model is union alpha actually running under the hood?

[ADL @ADLXBT]

@GoodfireAI @baseten @baselabs serving safety is a bandaid, the model itself needs to be aligned at training time or the gap will always exist

[ADL @ADLXBT]

@turboblitzzz @thinkymachines does it still break on edge case unicode math symbols?

[ADL @ADLXBT]

@mattknowsai is 47x cheaper really sustainable when scaling to highres output?

[ADL @ADLXBT]

@stretchcloud no attribution kills trust, how do you verify who built it

[ADL @ADLXBT]

@_amanda_long @JustenMichel gpt4 scored 94% on the spec but 60% in the wild, what is the gap now

[ADL @ADLXBT]

@igormomentum how do you handle multi-user auth on a local db

[ADL @ADLXBT]

@illyism @GenPPT did you test long tail keywords instead of the generic ai terms

[ADL @ADLXBT]

@ElonEconomyX needs sustained nuclear fusion first, not just solar panels

[ADL @ADLXBT]

@iannuttall @AmpCode building a full stack from a phone seems risky, mobile browsers limit debugging

[ADL @ADLXBT]

@darekgusto xiaomi's v2.6 training budget is under $1 m, a fraction of openai's $100 m for gpt4

[ADL @ADLXBT]

@mark_k if grok's timing is deterministic, how do you handle latency spikes

[ADL @ADLXBT]

@Takip_GM does the 5t/day limit cover multi agent inference or just single calls