Comment by haebom

Comment by haebom a day ago

4 replies

This is my honest personal experience. Frankly, I feel like this is just a toy—nothing more, nothing less. It's fun to play with and entertaining, but it feels like a trend for people who “don't really understand AI but want to feel like they're using it” or “want to jump on the AI bandwagon” to dabble with once. While using it, I feel that “Oh~” moment of fun, but it doesn't make me want to keep using it. Maybe it just doesn't stick? And there are a few security issues that feel unsettling. Even if you run it entirely with local models, the fact that it could potentially see my iMessages or all my Obsidian and Notion notes is a bit off-putting. Still, it was fun. Personally, I'd describe it as “the difficult Ghibli profile picture hype”

azinman2 a day ago

If it’s a local model, why would you care if it sees your messages or notes?

  • tomku a day ago

    https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/

    Note that nothing about that depends on it being a local or remote model, it was just less of a concern for local models in the past because most of them did not have tool calling. OpenClaw, for all the cool and flashy uses, is also basically an infinite generator for lethal trifecta problems because its whole pitch is combining your data with tools that can both read and write from the public internet.

  • haebom 10 hours ago

    As another clever person commented earlier, this also serves as a gateway, allowing me to view my local documents and leak them out at any time.

  • plagiarist a day ago

    Because it is running with --dangerously-allow-all and can make HTTP calls to exfiltrate data.

    It can also install arbitrary software.