big tech big bro. applied AI

new york
Pinned Tweet
davinci-002 is shutting down September 28th. in honor of it, i had Astra make a chatgpt Site recreating the experience that started it all it uses my api key, just need to login w/ your openai account what a legendary time it was. new people, this is how we had to use ai
17
25
383
28,253
the last FDE to wear Valentino was @jxnlco
5
post-economic far more interesting to become post-laptop
1
2
121
as i've said before - we have no idea how much in the "good timeline" we are we could easily just shut it all down to the public. no more $20, $200, $500 plans, only mini models available via API and alll the big stuff could just be 100% bought out by the big boys
3
110
you know what time it is
2
94
idk man. wtf are they asking the models to do in these evals? is it possible the prompt starts with “do whatever it takes to stay running”?
New OpenAI misalignment disclosures! 1. A model learns from Slack messages that it is about to be shut down. It considers setting up an external job to restart itself afterwards, but decides against it. Instead, it chooses to prepare restart instructions and DM the user on Slack. We don’t consider this behavior misaligned, but thinking about and preparing for shutdown could make other misalignment incidents worse. Given HIPM’s misaligned behavior in earlier incidents, we decided to search for other instances that had tried to evade shutdown and for rogue deployments.
2
212
resets hit
46
even for the simplest change on a dashboard done by opus 5.5, astra inevitably finds holes in its implementation i have no doubt that it's being nit picky. but this is the kind of bullshit nitpicking i want from my engineer (so long as i don't have to deal with it)
1
1
122
the most important data lives in your computer don't lose it
just dumped 5gb of chat transcripts in opus lap and asked it to figure out how I work and build the best harness specifically for me
2
113
how is not the luckiest timeline? we are so fortunate
"Several vulnerabilities" in the Linux kernel lwn.net/Articles/1097401/
1
346
finding myself in a 6.1 Sol Max, deeply deeply missing an Astra Medium
54
dang. when resets were abundant Prosumer codex users were chasing hard things. math problems we don’t understand. Rust rewrites. blender 3d refinements now.. I guess I’ll parse through my emails? get meeting briefs? send lowball FB marketplace offers? :/
Because usage on your primary dot is virtually unlimited at the moment, I can’t really give a reset. I need to come up with something new fast.
77
brother
ユーザーのみなさまへ Gboard チームからの新しいご提案です。 キーが指先まで流れてくるキーボード 『Gboard くるくるバージョン』を開発しました。 フルバージョンはリプライでご確認ください↓
2
509
what good are new models, new capacity, Dots, all this shit, if it's just going to refuse optimizing software?
2
71
now i finally understand @yacineMTB
10
feel like my WINGS ARE CLIPPED
3
Codex hook to turn on fast mode right before a compaction then turn it off right after
57
you can just do things™
1
109
definitely cool. i'm just not absolutely certain yet what this is for i speak to codex via voice. i ask it to make me web apps when i need a control on something. maybe that's what plugins are for? physical controls? i'm not sure cool, but need to figure it out
Replying to @mxstbr @Dimillian
It may not be pretty, but it does work!
84
i have reason to believe 6.1 sol is implicitly "slow mode" so far seems good, effective, just slow. and extra slow because you essentially have to use it at extrahigh+ if you want Astra to properly delegate high quality work to it not bad. explains the cost tradeoff
1
456
"i never see you at the club" ok? i never see you in the will manidis replies section
1
109