36
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 19 May 2024
36 points (100.0% liked)
TechTakes
1441 readers
49 users here now
Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.
This is not debate club. Unless it’s amusing debate.
For actually-good tech, you want our NotAwfulTech community
founded 1 year ago
MODERATORS
I started pre-hurting in 2018-2019 when I saw SoC vendors pitch things as AI chips because you could natively run accelerated models on them for speech and image recognition. Yes, really, that was the total pitch. It’s AI because you could run models. Where to train the models, what the total scope of capacity was, total execution space, total throughput, etc? No no, not practically touched at all in those docs
And where I said pre-hurting? At the time I thought “oh god”….you can imagine how quaint that memory feels in retrospect right now
oh yeah doing embedded dev that was always the worst. my RISC-V SoC has an AI accelerator on it! cool, what does the datasheet say it does? TensorFlow lite and nothing else. ok maybe I can use that silicon for something else at least? nope, there’s no documentation on its capabilities or how it works at all, it just does TensorFlow lite. shit well ok, what can I do with TensorFlow lite? you can load the example model we’ve provided and make a shitty voice activation trigger or try to train a model but there’s no docs. well fuck me then.
And don’t you dare ask about the intended binary representation, that’s obviously some illegal hacker shit
Just render it out with lite! What do you mean about in 3-5y? Software doesn’t change, you scrub. We have a golden master toolchain right here on our ftp, it’s at
ftp://nonssl.http25.pub.norsou-semi-sometimes.xyz/pub/archive/noindex
feel free to download and develop your own applications!