Boy, do I have a few!
A monitoring system that releases a tiny bit of candy for me when I’m doing things I approve of on reflection bc. rewards > punishment.
Live dialogue state tracker that shows you the dialogue tree. Works IRL, in voice calls or in digital chats.
Textual semantic mute. It can serve as a firewall between you and all the ragebait of the internet.
Proper speaker transcription that solves the issue of tracking who’s saying what.
A webpage that aggregates all my messages together in one place. It is very annoying to use Slack, Discord, Signal, Twitter, Bluesky and so on. If only there were an everything app of some sort…
Community notes for the internet. Someone has probably done this already, but I’m sure the implementation sucks.1
Exercise generator for arbitrary maths/physics skills, drawing on every (good) textbook ever written, adaptively tayloring them to suit the users knowledge and skill.
LLM multi-agent harness with good LLM team composition presets.
A tool that makes a heavily tagged list of observations in all the books I read, fictional or not, so I can theorize w/ all the relevant info instead of whatever pops into my head at the time.
Something to log all the actions I take on the computer, in the context I take them in, for later perusal, in the vein of Croissanthology‘s efforts in preserving his digital self.
A view of LLM chats that lets you swap between a linear thread, a reddit like indented tree and a loom style graph, and lets you toggle between styles in a branch independently of what precedes it.
Rating re-scalers that integrate w/ your browser to take any N/M rating of some X and transform it into “(thingy rating - average rating of things like X)/\sqrt var (ratings of things like X)” where “things like X” is computed using a cached list of items found by an LLM. Maybe we could modify this to use a re-scaling method that’s better suited to the vagaries of ratings, but I feel like this’d automatically save me a few seconds-minutes of searching each time I want to get a sense of what numerical ratings mean and how they vary. Heh.
Did I ever mention I spent a few weeks trying to write an agent harness a couple of years ago after I realized how horrifically bad all of the public ones were? They wouldn’t even give the agent memory! And somehow they got 5k stars on Github. LLM slop is not an issue because artisanal we were already inundanted with artisanala human-slop for our entire existence.
4. Vibevoice ASR seems pretty good
5. Beeper.com is pretty good. I've been using it for a while now.
6. That would be hypothes.is