Passkeys are similar to client carts but they aren't quite the same thing.

It doesn't work that way for everyone. Personally, I found that trying to write down everything fast enough took so much of my focus that I didn't have any left to THINK ABOUT what was being said. I did much better without note taking, in classes where that was possible. There were a few where that would not work.

You do realize that the whole point of on-device models like Nano is to give you the capability without the privacy problems of sending your data to Google, right?

It's true that with additional layers such as TLS passwords can be protected from MITM, but you can't always control how the other end of a connection behaves. It's better to use a mechanism that's stronger from the ground up.

Passwords are subject to Man In The Middle. Passkeys are not. A passkey handshake verifies both you AND the server you are connecting to.

Don't worry about multiple rounds. Use a long enough password to get high entropy and use SHA512. You'll be fine.

That's amazing, I've got the same combination on my luggage!

Just because these systems don't work exactly like us doesn't mean we don't have similar processes. It just means there's more to it.

Just use passkeys. Problem solved.

For me the issue isn't determinism but persistence. No matter what the computer is doing, it cannot be said to be conscious when its experience keeps getting reset every few minutes. It would be like dying st the end of every conversation and being reborn for the next one. So what about long-lived agents? Well, they tend to go off the rails eventually if you don't keep resetting their context. Is that just a curiosity or does it point to some deeper truth? Who knows.

And if they had this feature at the time, it would have said so.

I was skeptical of AI coding a year ago. Right up until it started working.

In my experience they are quite good and getting better. A year ago they were a better autocomplete; now we can point them at a design doc and expect a reasonable result most of the time. A year from now maybe they're WRITING the design docs, and then what?

A discriminative net becomes generative when run in reverse. Approximately. It's a little more complicated.

Reasoning models just mean the weights get converted back into natural language at intermediate steps. The training can then reward "good" reasoning paths and the prompts can focus on narrower increments of the problem. They can also use tools to error check and course correct and whatnot. It's a strategy. It works well. Is it reasoning in the human sense? Hard to say. It does look similar because it's trained to.