The biggest problem with AI today is that it can't learn in realtime.
I've been exploring ways of creating a neural net without "weights" or back prop. Instead, there's is constant branching of synapses and pruning. This plasticity I hope is at the foundation of enabling a net to learn.
In the demo, we only have 145 neurons and only a couple of feedback loops and yet, it's enough for basic steering and seeing an entity move faster and faster towards it's goal.
@HughHopkins I think this is a bit of a BS statistic, it's a bit like GDP or stock market performance but that is not to say that uk has problems (mainly caused by brain damaged civil servants)
I built a Ruby CLI for @Superthread_ that's designed for AI agents (and humans).
In addition to regular views, every command outputs JSON. It ships with a Claude Code skill so agents know every command without running --help.
v0.7.3 out now: https://t.co/F9krPu4jew
@chrija People have tried building/running software on prem. Basecamp guys built a Slack replacement that you you host and run your self. To my knowledge it was a complete failure. That said SaaS pricing is extortionate, this is why we are going all in on PWYW.