@fanjie
TTFT seems really slow on glm 5.3. even tho the tps is good enough, it just makes it seem really slow. lmk if you need any additional information from me.
“AI is a tool for the rich.”
It’s $20 a month.
Hiring a lawyer is $300 an hour. A tutor is $60. A developer is $10,000 a project.
The rich always had access to expert help.
AI is the first time everyone else does too.
@BearHuddleston Starting new chats without a specific project, sometimes I just want to talk in general/about multiple projects.
Also, if I could have all the chats from all the gateways together in an inbox style, that'd be great!!
@RightSideOfAI I use Hermes as an orchestrator as it has all the memories. Then I just allow it to run threads via t3 code on multiple machines. I have it so it almost always delegates tasks so I can have a very long conversation spanning weeks.
btw, not using deepseek api directly, used @opencode go.
for now it was just screenshot -> act -> screenshot loop but will need better optimizations.
could be a good bench as you need to coordinate strategy + placements + need enough tps to act.