just finished up filing taxes for this year fairly certain this will be the last year i do that manually next year gpt whatever will do it for me believe we're crossing that threshold already

well i guess we can count on now perfectly normal and reasonable people will be able to also have personal assistants and much sooner than i had anticipated and as hackers we'll be able to maintain our own system modifying it as we see fit great day for open source

how many of you were texting your openclaw yesterday to sneak another codex session

what an anthropological study

tpus are about 2-3x more efficient at inference compared to gpus groq has something called an lpu you know the thing we need is purpose built hardware for consumer devices just meant for inference gpus for training and this something else for inference apple doesn't cut it

this is pretty awesome not all material is digitized i needed to get my last year's journal to my @openclaw so did something similar just took picture of every page and dumped it on the disk my assistant was able to digitize this just fine with minimal mistakes, takes 5 mins

gemini models are still goated for video understanding afaik they're the only hosted api where you can send a video file/uri and it just works end-to-end (vision + audio), no frame-extraction + separate transcript pipeline pretty wild how much friction that removes for

could have saved $200 if i'd waited dang that is a good deal a low powered always on assistant that runs in your home on open source software @openclaw mentioned in the copy is wild

My @openclaw assistant texts me blue bubbles, reads my email, knows my message history, and runs on a Mac mini in my house. All open source. @steipete cooked. The fun is back in personal computing and I'm living for it.

https://t.co/chTW20kgAQ

the subscription model for inference is probably going to break once everyone moves to assistants running tasks 24x7 subscriptions only seem viable to me if at least some people are leaving tokens on the table this will change when we all have assistants

psa to every one else out there on the fence, stop reading about it just use the thing and you will see who cares about someone else's use case it's a personal assistant you will learn by doing trust

typing indicators and read receipts are the real bougie ux you want when texting your @openclaw assistant but you have to disable sip which might to be the best idea wonder if you can run bluebubbles inside a vm on host with that instance sip disabled to get the private api

space gpus okurrr

you can use your friendly neighborhood clanker to configure @openclaw and in fact you're going to need it expect near constant breakages and yes it is incredible though build a second brain with open source and with awareness of all your data sources outside of corporate walled