I’m not very convinced by the article. It introduces a method of obfuscation that could be solved once and for all by a mapping built in like 1h (which is probably also part of the software used for obfuscation itself).
I wouldn’t be surprised if some NLP libraries already normalize this kind of text?
I don’t get why some investors still think it’s a good idea to make a voice based interface. It always looked like a bad idea but I thought that the relative failure of home assistants proved the point ?
TLDR : the user database was saturated because of 60k accounts created in the previous 24h for scrapping purpose.
Especially the part where they drop that they expect to reach 1 watt before the end of 2025 (4 orders of magnitude higher than current 100 microwatts)
That makes a lot of sense👍