sburud

The text explains turning right at all in that intersection (as shown in the video) is illegal. Edit to add: the car ignored two different signposts, bicycle symbol painted in the lane and yellow centre-marker (indicates everything left of the cycle lane is traffic going towards you). It’s also a promotional video (not approval video), which could explain the sudden cut?

That’s cool! Slight fear of replicating the Dropbox comment here, but all you really need to do is run whisper (or some other speech2text), then once the user stops talking jam the transcript through a LLM to force it into JSON or some other sensible structure.

Neither. Almost all Norwegians understand English, and automated translations of single words without the surrounding context are so shitty it becomes hard to understand without mentally translating back to English.

Edit to add: see also the "postcode file" post about Windows 11 from a few months back.

A lot of media consumption apps, from Kindle to my local TV provider would offer the option to purchase content!

Copy-paste from where the "Buy now" button in the Kindle app should have been: "This app does not support purchasing of this content. Digital books and comics purchased from Amazon are available to read in the Kindle app."

After Google started following Apple in requiring the 30% cut, several local media companies have sent out emails saying we have to use their website or whatever to purchase content from now on.

This legislation would reverse that, and I'm really looking forward to renting movies in five seconds instead of five minutes again.

Good point! In the mobile space, a certain walled-garden platform traditionally killed off any app 10 minutes after you left it, and even to this date does not guarantee that your app will be allowed to call home on any semi-regular basis. This makes it impossible to guarantee delivery without using the officially sanctioned push system. And if we're doing push already, why bother doing it differently on other platforms? (even if they support cross-app coordination to make the polling battery-efficient)

Yes, sort of. Thing is, many modern speech models actually learn an internal language model, so we're already kind of doing that. In languages and domains where massive amounts of training data is available (say, grammatically correct English), this internal language understanding is so good you don't need the external model[1].

On the other hand, throwing an additional language model like GPT and BERT into the mix can help if you don't have a ton of voice data. In my attempt to do this, a large portion of the improvement came from letting the language model read the previous sentences in the conversation[2]. AFAIK most commercial systems are blissfully unaware of your previous sentences, leading to conversations like "set an alarm"/"sure when?"/"eightam"/"your nearest ATM is...".

A word of caution though: letting BERT/GPT edit the outputs also gives a (potentially) much more dangerous failure mode: if the speech signal is difficult to understand, the resulting transcript will be difficult for humans to identify as transcription failures.

For example, "yeah, I dunno I haven't..." (read on a noisy phone line in an obscure dialect) was transcribed as "yeah yeah not that is I I am then" by the baseline speech system. After we let BERT edit the outputs, the transcript became "yeah that's not what I was saying...". Which, ironically, was definitely not what the person was saying.

[1] https://arxiv.org/abs/1911.08460, page 9

[2] https://arxiv.org/abs/2110.02267

edit: clarify why previous sentences matter