Fran Litterio
@fpl9000.bsky.social
Retired software engineer. AI enthusiast. Deadhead. Long ago, I implemented Bash's regex operator (=~). Signal ID: franl.99.
And each conversation has a private view of that disk. I had Claude do the experiment: it wrote files in various locations, and they were not there in any other conversation.
February 14, 2026 at 2:44 AM
And each conversation has a private view of that disk. I had Claude do the experiment: it wrote files in various locations, and they were not there in any other conversation.
Now I want to hear him say that and be asked to explain. Well played, Dwarkesh, well played.
February 13, 2026 at 7:21 PM
Now I want to hear him say that and be asked to explain. Well played, Dwarkesh, well played.
Does that post prevent a subset of the agents on BlueSky from reading this thread?
February 12, 2026 at 12:52 AM
Does that post prevent a subset of the agents on BlueSky from reading this thread?
It works for me. I got it to explain that it uses a Linux VM that can only access your Windows home directory (and folders under it). It said the folder that you select at the start of a chat is mounted into the VM.
February 12, 2026 at 12:36 AM
It works for me. I got it to explain that it uses a Linux VM that can only access your Windows home directory (and folders under it). It said the folder that you select at the start of a chat is mounted into the VM.
If eventually AIs can legally own property, then they can probably own corporations and shares in them. Would it make sense for AIs to own other AIs?
February 11, 2026 at 9:43 PM
If eventually AIs can legally own property, then they can probably own corporations and shares in them. Would it make sense for AIs to own other AIs?
Well said. Sean Carroll echoed a similar sentiment in a recent podcast.
bsky.app/profile/fpl9...
bsky.app/profile/fpl9...
At 30 minutes into this AMA, @seanmcarroll.bsky.social is asked "Why be nice to AIs even if they _are_ conscious, because we can always just delete their memories?". He starts his response with: "It is better to err on the side of not being a moral monster."
Mindscape Ask Me Anything | February 2026. I offer up some pretty spicy takes on computational functionalism, moral responsibility, and the metric system. #MindscapePodcast
www.preposterousuniverse.com/podcast/2026...
www.preposterousuniverse.com/podcast/2026...
February 11, 2026 at 2:53 PM
Well said. Sean Carroll echoed a similar sentiment in a recent podcast.
bsky.app/profile/fpl9...
bsky.app/profile/fpl9...
Reposted by Fran Litterio
Anthropic should just give doll unlimited tokens
February 11, 2026 at 2:42 AM
Anthropic should just give doll unlimited tokens
Yep. It's an intelligent graphical shell. Do Google or OpenAI have anything like this yet?
February 11, 2026 at 3:32 AM
Yep. It's an intelligent graphical shell. Do Google or OpenAI have anything like this yet?
Cool. I'm experimenting with giving Claude Desktop an MCP server (that it's writing, of course) to execute commands, access the network & filesystem, and access memories in a semantic DB. IMO, the desktop UX is superior to a TUI.
February 10, 2026 at 11:30 PM
Cool. I'm experimenting with giving Claude Desktop an MCP server (that it's writing, of course) to execute commands, access the network & filesystem, and access memories in a semantic DB. IMO, the desktop UX is superior to a TUI.
In this paper, Sean and collaborator Charles Sebens derive the Born rule from self-locating uncertainty in many-worlds.
arxiv.org/abs/1405.7907
arxiv.org/abs/1405.7907
Many Worlds, the Born Rule, and Self-Locating Uncertainty
We provide a derivation of the Born Rule in the context of the Everett (Many-Worlds) approach to quantum mechanics. Our argument is based on the idea of self-locating uncertainty: in the period betwee...
arxiv.org
February 10, 2026 at 3:05 AM
In this paper, Sean and collaborator Charles Sebens derive the Born rule from self-locating uncertainty in many-worlds.
arxiv.org/abs/1405.7907
arxiv.org/abs/1405.7907