Pinned Tweet
Today I’m launching Embasis—an iPhone assistant with AI that runs on your device.
A Daily Briefing you can question. Help with meetings and writing. Leave-time alerts based on your route.
Local AI. Internet tools when needed.
embasis.app/
ģ̶̥̩̮̊̐͗́͆̑̔̔͊̃̚o̷̩̩̰̔̉͂̇͐͌ǫ̸̛̤̼͓͎̭̖̜͓͎́͛̅̿̔͗͝ͅͅd̷̬̖̆͗͑̿̓̈͒̋͂͘ ̶̦̆̐̿̂̽̀̌͌͘e̷̺̿̅v̴͎͎̬̱̟͕͕̰̊e̸̢͚͍͓̺͉͎̱̾́̔́͑̚͠͝n̵̨̧̡̖̻̘͚̺̟͚͆̈́͠i̶̫̦͒̃̉́ņ̸̻͖̙̜͓͉̲͓͛̏̂̏̍̅̍̑̌͌̉͒̓̑͜͠g̴̢̹̝̻̻̭̪͙̠̈́̈́̓͋ ̵̖͒̔͛̇̈̚f̴͓͕̜̟͇̟̲̯͙̗͒̓o̸̞̮̙̗͇̖̒l̸̢̛͖̭̯͒̍̔̏̌̄̚k̷̠̳̓̑͐̀̃̽̌͂s̸̯̭̮̭̭̮̦̯͙͖͍̑̈̅͘͘͝ͅ
Is it just me or is both Codex and Claude limits pretty bad right?
Some people say claude is still good, and it was for me until last week. it was kinda hard to burn through the pro plan
on codex i legit had trouble going through the pro $100 and now i finish in a day
i got voice generation to 5x real time and somehow the pause before it starts still feels like the slow part
google's planned cheyenne data center is up to 590 mw of grid-connected service. black hills plans $1.8b in new gas generation to serve it, with load ramping to a 2030 peak. the grid has its own build schedule
day 5 of looking for builds
looking for small ai tools for boring work. cleaning a messy csv, sorting downloads, fixing subtitles. what still needs a human check?
swapping to lighter runtimes took the bundle from 60.3 mb to 47.0 mb. the unglamorous packaging work counts too when everything runs on the phone
if an app needs the cloud whenever the small model gets stuck, i want to know before i put private stuff in it
benchmarks are mostly marketing. the only eval that matters is whether u'd actually use it every day

