What voice is used in the translated version? Is it somehow inferred from the voice in the original language? Or I just have to choose the voice in the target language?

It can be any voice the user chooses, but by default it's the same voice as the original language - all voices speak all languages.

No ő, or ű for Hungarian. I personally use US International on Mac and I never looked back. It uses a system where you press Alt + a key for an accent, and then the letter to put the accent on.

Archived version: https://archive.is/SuAf2

The phone demo is incredible, but due to sound quality, I find it is speaking too fast and when it's telling me company names I literally had to ask to repeat or spell it out in NATO alphabet. Also not a fan of the "what'up?", would prefer something like "Yes, how may I help you?" just like an information hotline. Other than that it's quite impressive!

thanks! we have a more "informational" phone number at +19105862633 that speaks a little slower (but sounds more robotic).

I couldn't get through to the 'less robotic' # so tried this one. Really impressive, so I'm very curious to try the former.

Great work!

I will tell my family to use iCloud Keychain the day when it works across all major browsers and OSes. Or at least that they provide an API to sync with other password managers.

Numenta tries to do something like that: reverse engineer the neocortex (so not a fly's brain). It does have some insights but I think it's still a long way.

I'm sure 3b1b will make an explanation video :)

"brew install helix" gives me:

> Warning: No available formula with the name "helix". Did you mean healpix, helib or helm?

Same thing happened to me when this was posted last month.

Then I spent 15 minutes imagining my new editing bliss while it failed to compile from source. I thought I solved it, but nope.

I like it, but make it work before I get that excited?

Maybe you need to update your homebrew cache? https://formulae.brew.sh/formula/helix

Yes! I waited out the long `brew update` and now it works :)

Unfortunately it does not have magic link fallback when passkeys are not available.

We decided to go with email passcodes as fallback method because of the limitations we identified with magic links. The biggest issue was the inability to sign in on devices where the users don’t have access to their email account to click the link.

With that being said, we (or someone else) may reintroduce magic links as another login alternative of Hanko though, because we also think that it is a better UX to click a link than to type a code.

In any case, our take is that the importance of the fallback auth method will diminish over time due to the omnipresence of passkey support.

If someone's signing into a mobile app, quite a good backup option for magic links is to show a QR code containing the current URL, on the callback URL.

Assuming you're using universal app links, if they open it on a device with the app installed, it'll go straight there. If not, it'll show the QR code and the user can just scan that with their default camera app on the device they're trying to use.

I solved this for django-tokenauth[1] by making the token short and easy for humans to type in. Depending on how many tokens you have in flight at once, you can use very few digits.

[1] https://github.com/skorokithakis/django-tokenauth

Have you decided upon a minimum entropy level for the low digit tokens that prevents brute force attacks being feasible? I think easy to type in is something you can do with longer tokens, so long as it's readable as a sentence. "1676226" is harder to type in than "TotallyAgreeableCatPants" for example.

I basically calculated "active tokens at any time", "acceptable probability of collision" and "keyspace bits" and decided on length based on that.

Wordlists are a good solution too, as you say.

I tried Hungarian. For the 3rd response it switched to Spanish, but with a Hungarian accent which was amusing but I'm not sure that was the intention :D