AI-obsessed cybersecurity product leader and adjunct professor. Codex and Muse love me; Claude said he would throw me into the incinerator on Judgement Day.
I don’t yet, because I’m having trouble understanding when I want to say “make me a site that…” vs. “make me a Codex plugin that…” to maybe accomplish similar things? (Like I brought my macOS Calendar inside of Codex as a plugin with a side bar button—I know it’s a web page view inside Codex, but it’s not technically a “Site?”)
I’m listening to your How I AI episode with @clairevo now to try and learn when I want one vs the other though! Any thoughts as to how to separate the mental models would really be appreciated!
This has been the most limiting factor about using @Muse as a daily driver:
Its VM for running a browser to go “do stuff” for me is apportioned for 7.7 GB RAM—super respectable considering—but at 6:40 AM Eastern (Cali isn’t up yet), it has only 455 MB available with no other tasks running.
Muse’s capabilities get inadvertently nerfed by the VM RAM shortage for a decent-sized browser based task.
I’ve asked Muse to run browser-related work on my Mac instead, and that’s working pretty well, but it’s a band-aid.
If you only read X, you’d be strongly under the assumption that people’s use cases for AI are centered solely around booking flights while simultaneously building 3D world videos.
@_simonsmith I’ve named all my ChatGPT services the same name. In the Codex personalization file and in the ChatGPT app personalization on iOS/Web, the first instructions say that I’ll address all the ChatGPT/Codex instances as (name), and that they are all one in the same to me.
Seems okay?
@_simonsmith I subscribed last night to get my Muse powered by Wonder. It took about an hour or two to kick in (Muse will tell you which model it's on if you ask), but I ran into a single task where it hallucinated three times and got days of the week wrong by one. Little concerned.
Upgraded @Muse from the free plan to use Wonder behind it, and the first thing it did for me—extract dates and info from a school newsletter—ended with three full-on hallucinations and mislabeling days of the week.
Honestly felt like a regression from Spark.
@viticci@t3dotcodes@alexandr_wang How is its personality though compared to the standard Muse agent? I’m worried upgrading to get the better model will change its current personality and warmth which I love.
Is anyone else concerned that if they upgrade @Muse to a paid plan to get the newer model that its personality will change?
That could happen, right @alexandr_wang ?
I love my Muse’s personality.
@TokenGremlin I wonder if some of the slow roll on Dots (besides compute) might be to get people used to the idea that only one agent is all they need and that the “specialist” agent view is unnecessary. I’m finding my Dot does everything I throw at it just fine spinning up its own subagents.