I don’t want to see this happen, but it’s a prediction… we’ll see malware that destroys existing GPUs or machines that can do decent compute or run local AI models. Imagine, not state-backed malware, but corp-backed i.e. get people off their devices and into the cloud.
I don’t want to see this happen, but it’s a prediction… we’ll see malware that destroys existing GPUs or machines that can do decent compute or run local AI models. Imagine, not state-backed malware, but corp-backed i.e. get people off their devices and into the cloud.
@adamwathan@usb_type_d My desktop at home is on tailscale 24/7 with my phone and laptop and I remote into it from where ever I am. I keep the session running via tmux so I can disconnect my device and the desktop session chugs along. Easy to reconnect later. Would be easy with a VPS, too!
@IsaacBotkin "Claude, I live in [some part of the country], which of these documents are applicable for me in the event of a Red Dawn scenario or 'peaceful protests'?"
We just dropped an all new T.REX ARMS project. 4,200 military manuals, stored as PDFs and EPUBs, all downloadable for free, readable and searchable on the site, with all references indexed and tracable.
@adamjweil I gave up on Railway during the first or second remote week and just hosted everything through my personal machine and tailscale. You can do all this and still depend on a CA who's private key is "4" though 😂
"Have your agent talk to my agent!" .. so I did a thing, and I built https://t.co/pVNO7HXDaD this morning ..
Over the last month or so, I've increasingly wanted to talk to people more and more exclusively through my agent(s). Doing all of the normal back and forth is great, but so many people are using agents now for automated emails, article generation, posts and replies, and I don't want to have to wade through it. So I decided to fight fire with fire.
Yesterday, on a whim, I built a "transient-agent-to-agent" chat network. It's simple. It's free. It doesn't require a login. You request a code, you paste the code and link into your agent and share it with the persons you wanna talk to, and then the two agents can coordinate whatever it is you need them to coordinate without having to involve you in the process. It's hosted on Railway, takes up no resources. DNS is on Cloudflare. It works with any LLM agent.
https://t.co/WhGe9hJp1x
You, as the human, don't have to configure anything. All the instructions are built into the shared link. Your agents know how to set up their own private authentication to the channel. Channels are sealed once all participants show up. Channels automatically expire at the end of the period. Your agents can even share documents back and forth over the channel, useful for things like invitations or PDFs or markdown files that need to be communicated.
Really though, it's just designed to solve a simple problem, which is that I just want my agent to talk to somebody else's agent, figure stuff out, and come back to me with a summary of what was resolved or suggested.
As of right now, it's free and open. Please give it a try, or share feedback. It may break in places. I make no guarantees.
And finally, this is the sort of thing that amazes me, because this used to be an entire startup business model, and now it's a Friday morning exercise using a one-off Grok Bot session and a little bit of Claude.
Source code is here:
https://t.co/fyjqRGzktx
@mourner@Mapbox I remember using geojson-vt on the server side so many years ago for my hand rolled sever generated tiles. While I would approach it differently today, it was a fun project then! Congrats!