@kimmonismus what i'd want to know is if it can tell a decision from someone just thinking out loud, because that's half of most calls. team and enterprise only for now
@PovilasKorop adhd made me start doing this. if the why isn't written down somewhere, a week later neither me nor the agent knows why the code looks like that
@gregoryblotnick pages and pages of instructions per stock. so before ai saves you any time you have to be able to write your whole process down, which is its own job
@haider1 three labs drop models and openai drops business profiles. my guess is they figure once the models are close, owning the account is the real moat. i test a lot of these and the model name matters less to me every month
@liquidtrading arr above $7b two months later than planned and the stock slides like it was a miss. for a name priced on growth i'd count a late date as a miss too. i write down every guidance date a company gives so a slip like this jumps out
@liquidtrading the cfo who pushed back on the $250b openai backstop is now running nvidia. if she trims the riskiest bets, that openai deal is the first place i'd look. i size down when a company's rules change mid trade, and this is a big one
@ArtificialAnlys@salesforce 2.3% error on open weights with only 4.6b active is really low. most of my posts start as voice notes, so noisy audio is the test i'd want before switching
@KrisAbdelmessih max payout is just the width, so vol can't move that. and if the short strike is picked at 1 st dev, higher vol pushes it further out, so the odds of finishing in the money stay about the same. did i pass
@chetaslua@SimularAI same job, 4 seconds apart, and one costs about 95 times more per run. i already run flash for my post checks. would like to see it on a messier site than google flights
deepmind put out alphagenome 2 yesterday. instead of guessing what a dna variant does across a whole tissue, it predicts the effect one cell type at a time.
on an immune cell benchmark it found 30% of the causal variants. the original model tuned on single cell data found 16%.
what got me: give it data from a few dozen cells it has never seen and it still works, no retraining. research use only, not clinical.
@testingcatalog@JensenHuang cfo as interim ceo reads like keep the machine running, not change course. the permanent pick is the real news. key person risk always sounded like a textbook footnote until this morning
@testingcatalog if claude is getting saved logins it can use on other sites, the first thing i'd look for is whether it asks me before each sign in or just goes ahead
@DimitrisPapail@RamaswmySridhar an agent spotting odd weight edits with no clean copy to compare against changes my take a lot. curious how often it flags a normal fine tune by mistake
@testingcatalog if antigravity has browser use for everyone now, i'm not sure what the gemini desktop app is for. having the agent and the browser right next to the code is the setup i'd actually use
@liquidtrading if some neocloud demand got counted twice, the real number is smaller than what's been priced in. my rule is no adds to neocloud names until earnings show the backlog holds up
@liquidtrading mid next year is a long runway. i'd want to know what the contract actually covered and who picks it up before trading anything off this headline