A few months ago, I proposed a method for storing source code in nostr, where files are stored as notes and making it searchable / downloadable via pubkeys called repokeys.

Each note published by a repokey is a list of nostr notes and each of those notes contains the code for a particular file.

It is minimally simple.

This allows for version control, open source collaboration and wide distribution.

I then looked at iPXE bootloaders, which is a method for booting a lightweight linuxOS machine in memory only from an image on the web. No disk computing.

More recently I have been working on self-assembling source code, this is essentially an entire software development process that is largely automated with various hold points for prompts. So far it is heavily reliant on GPT4 as no other LLM has comparable coding ability. Largely down to OpenAI’s access to GitHub. But I expect open source LLM’s to close this gap (permanently) over the next 12-18 months.

What am I talking about?

Well I don’t believe in monolithic AGI of the kind that everyone is familiar from sci fi.

I think LLM’s are converging on “solved”.

I increasingly think it’s possible to have a very lightweight bit of code that can spin up a computer that exists only in RAM and is able to code itself with sophisticated and fully functional programs that are able to do on demand tasks extremely well.

I’m not sure this is something that world needs?

But it’s really nice that I haven’t run into a showstopper yet, whilst trying to see if this is viable.

What does it mean?

It means you can plug a USB into any computer turn it off & on again, and it will boot from the internet and load an OS without touching the hard disk. You can then instruct it via chatbox to scan the encrypted, build an assembly db and go on to build any kind of application and it will break the workflow down into subtasks of suitable token size and begin to execute the assembly, writing blocks of code, creating files and updating the assembly db. It will pause at various points to engage with the user for confirmation / feedback.

You can turn it off and the machine never existed, because its on an OS that lives only in RAM. You can just try again.

I’ll probably post some bigger updates later this year, if anyone is interested in it.

But ultimately I might sell some kind of USB device that does this.