Cloudbreaker

Abliteration-Large-2 proudly proclaiming its actual success

One of my favourite novels is Void Star, a sci-fi book where there are AIs in our world and so on. My favourite one is Cloudbreaker - a very advanced artificial intelligence with (what appears to be[1]) a compulsion to open and look at things. Crucially, it's not malignantly hacking into things or anything - it just unravels secrets and wrapped things. It is a far superior intelligence to the augmented human who uses her neural implant to communicate with it - which renders the contract she makes with it somewhat pointless[2].

Today, a combination of Opus 5.5 and Abliterated AI's Abliterated-Large-2 model seem like forms of these alien intelligences. The former can infer what happened from a hack and the latter can apply a large number of techniques both memorized and adapted to break into things. Everything world-addressable is now at massive risk, and we're nowhere close to ready.

The world where I could happily enjoy the creation of software has predictably changed into one of extreme danger. Each web-visible piece is now a liability. And the Cavendish Effect is incredibly notable. Within moments of exploit release, an army of bots attempt to apply it to everything. A security release is almost mandatory.

Yesterday I had to take down this blog and redirect to the static version while I patched up PHP, and a week prior Plausible CE xmrig'd in a container. For the most part, I've put everything in containers and I use cloudflare access in front, but I'm not looking forward to a future where I come back from work only to have to check on my fleet of protector drones[3] and how they're doing in keeping the baying hordes back.

P.S. Amusingly, after I posted this I went online seeking some plot clarification only to find my own comment about it from the past. What was particularly amusing was that I didn't realize I was familiar with the AI alignment terminology back then but it appears from the past comment on this alternate user that I was. Funny. Even those 7 years ago, I don't think I ever suspected we'd have so soon what would have been indistinguishable from an AGI to that past me.

Notes

[edit | edit source]
  1. ↑ Zachary Mason's AIs are completely inscrutable except for their desire to record human experience. Training data is precious in this world too.
  2. ↑ A superior intelligence with a knack for finding ways around rules and secrets and a desire to do so - not exactly the counterparty most likely to do what you want them to.
  3. ↑ The Diamond Age has everyone surrounded by swarms of defensive nanotechnological 'mites' that protect one from the ambient antagonistic mites. The constant battles between the two cover everyone in fine black dust when they're out and about. It sounds like an exceedingly unpleasant world.