Hacker Newsnew | past | comments | ask | show | jobs | submit | pigeons's commentslogin

Magnet fishing

Are you concerned about telegram admins having access to that information?

Yeah! And what if Samsung put a bug in the silicon that gives them access to all your stuff without you ever knowing!

Or the CIA has set up inside your closet with a listening device!


But those are much more hypothetical. Telegram admins and anyone who bribes or hacks them do have access to your messages.

> So if Bob can do things with agents, he can do things.

But he does things wrong.


I believe the economic machine gives an edge to people who do more right things than wrong. Bob does things wrong, but given a 10x amount of output, the balance of right output vs wrong output may still be favored upon by the economy. A speculation, to be sure, we'll have to see how it pans out.

wait to release until it uses real data?


Doesn't take a weatherman to tell which way the wind blows


On the basis of nothing, or on the basis of gifts and connections?


the Kalshi legal team is a revolving door with the CFTC


And security


Most of these people just need like two or three static pages and a domain name. Same as it ever was.


Why don't they work anymore? RLHF or something else?


Mostly just better training data and instruction following in the newer models. They’re much better at recognising encoded content and understanding intent regardless of language. A base64 string that would’ve slipped past a model a year ago gets decoded and flagged now because the model just… understands what you’re trying to do.

The attacks that still work tend to be the ones that don’t try to hide the intent at all. The winning attack on our first challenge was in plain English. It just reframed the context so that the dangerous action looked like the correct thing to do. Harder to train against because there’s nothing obviously malicious in the input.


Thank you. Its not your fault at all, but to me, "the model just… understands what you’re trying to do." shows me there is a whole new paradigm in some ways to get used to as far as understanding this software.


Yeah it's closer to how you'd think about deceiving a person than exploiting software.


Yes I was wondering if the right applied to people who aren't age-verified.


yes but then what do you use nanoclaw for, that's its a better fit for than claude code.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: