Uses deepgram, the nova-3 model is very reliable for English, nova-2 is pretty good at German. nova-3 multi is pretty bad. But for realtime transcription still better than the smaller whisper models.

great stuff @alexanderobenauer.com

in a world of tech slop, this quote feels relevant as ever:

Ich kann Ängste verstehen, jedoch müssen wir aus der digitalen Steinzeit im Gesundheitswesen raus.

www.theguardian.com/news/2025/fe...

And yes, this blurp was meant to troll (which it successfully accomplished). What makes me really sad is that this attitude is now (world)widely picked up and taken for “what Silicon Valley really is like”. For that reason alone, thank you for going on air and giving a different account.

But by completely disregarding the past, folks miss out on many insights that have not been popularized, and yes, are bound to repeat the mistakes made. If you then layer this amount of arrogance on top, the result is disgusting. Even if money is the objective, this is so harmful in the long run.

I think you touched on this in your conversation quite well. Even in this short time, there has been a tremendous amount of work and innovations that made the modern digital world possible. It’s easy (maybe to a certain degree even necessary) to mask out the details to act in the here and now.

Plus: a number of avenues that have been touched upon in the 70s and 80s but never really followed on thoroughly. I'm sure there are many more good business ideas hidden in the past.

"Dealers of Lightning" and "The Dream Machine" are very enjoyable reads and shed some light on how much work went into all this. I wish more people would be interested in how this all came to be.

Similarly: computational efficiency as in energy consumption. This is not just a PL question but also deployment and runtime behaviors. How to design applications for a low CO2 footprint? There is academic research but little has trickled down so far.