One Step Closer to the Star Trek Voice AI Assistant!

[@MachineLearningStreetTalk]

DO YOU WANT WORK ON ARC with the MindsAI team (current ARC winners)? MLST is sponsored by Tufa Labs: Focus: ARC, LLMs, test-time-compute, active inference, system2 reasoning, and more. Future plans: Expanding to complex environments like Warcraft 2 and Starcraft 2. Interested? Apply for an ML research position: [e-mail]

[@Charles-Darwin]

There's this game dev conference talk with the Zelda TOTK devs. It's super interesting in its entirety, but there's a segment in the latter half where they discuss how they achieved the sound system in the game. Since the SFX are dynamic, I wonder if the algos and approach could translate over to ASR. kzbin.info/www/bejne/hF7HgXepgbmliqcsi=ZtrDryCvvzU5J0RA

[@AeroMagic_Official]

What if I didn’t wanna and instead instead responded to a comment on KZbin … and end up six months later as a intern who can’t afford coffee but loves his non-caffeinated AI DRIVEN life ! because he works for the next open AI so it doesn’t matter he’s got his cameras and is not afraid to work for free if it means a lottery ticket style gamble could be at the other end of the prompt chain.

[@ateammate223]

The demo at 3:30 is super impressive-easily the most realistic voice simulation system I’ve heard

[@ateammate223]

5:07 “Ah yeah it’s quite impressive isn’t it, Tim” lmao

[@hayder978]

You can notice how the transcription accuracy of speechmatic is better than Google's ASR at 3:00

[@kensho123456]

We need to accept a lag because nobody can ever foresee the future so instead it's all about correcting the past (utterances) and this means the hearing person needs to be patient and wait for the lag - because the deaf person needs it, in order to get the full message.

[@iAkashPaul]

With WebRTC you do not need to turn off the mic. There are underlying DSP components which don't let speaker input feed back.

[@ginogarcia8730]

Generalizing an AI even more would make it conscious man

[@simonstrandgaard5503]

Many awesome insights

[@lucamatteobarbieri2493]

Since llms models are redundant and tollerate quantization one could make the quantization dynamic, allowing fast responces only when low latency is needed and slower more demanding reflective responces in other contexts.

[@oncedidactic]

Need a cursor integration, then we’re off to the races

[@LatentSpaceD]

who are all those people walking by in the background?

[@_ARCATEC_]

💓

[@grumio3863]

Latency is totally a bottleneck