redox99

39 ms·

811.
by redox99 3y ago
Sure. I don't know enough about those companies so I'm not complaining about them. But I know Samsung has the money and manpower to drastically improve the situation, yet they don't care. With the right abstractions, they wou
812.
by redox99 3y ago
Meh. Samsung (and Huawei) are still the best selling phones in developing countries. Samsung could definitely both reduce the unnecessarily large number of SKUs, and better unify their software stack so they could update for 5 years all of
813.
by redox99 3y ago
That concept looks so much better than the result, it's not even close.
814.
by redox99 3y ago
Hopefully as LLMs get smarter, you'll need less prompt engineering and stuff will "just work" across models.
815.
by redox99 3y ago
I'm surprised you're having such a good time with 7B and 13B models. I find anything below 33B to be almost useless. And only 65B is close to GPT 3.5.
816.
by redox99 3y ago
I think a better (but harder to obtain) metric would be activity. Given that people use their Instagram account, the user account isn't too relevant since you don't really need to sign up.
817.
by redox99 3y ago
They are comparing text-davinci-003 with ChatGPT which presumably uses gpt-3.5-turbo, so quite different models. They are killing text-davinci-003 btw.
818.
by redox99 3y ago
If they finetuned the model on AWS docs, would the embeddings, and thus the vector search improve?
819.
by redox99 3y ago
No web/desktop version is a dealbreaker for me
820.
by redox99 3y ago
The web/desktop version seems to be basically read only. Can't reply, sign up or anything. Can't even view a full user profile I think?. No home page either for me. https://www.threads.net/@hbo
821.
by redox99 3y ago
> A modern CPU should be able to decode 1080p in software without breaking a sweat. Not true for modern codecs such as AV1
822.
by redox99 3y ago
Sometimes during training, fp16 will cause networks that would converge on fp32, to explode to Infs or NaNs with fp16, because of the limited range. bf16 generally speaking fixes that. It's true also that fp16 is often manageable with
823.
by redox99 3y ago
I never mentioned high frame rates. I mentioned uneven frame times, because you can't properly show 24 or 25fps on a 60hz display. The vast majority of videos on YouTube are 30fps so they don't have this problem.
824.
by redox99 3y ago
Why is the video 25fps? To make it more "cinematic"? It isn't even 24fps cinematic. In either case, they both look awful in your typical 60hz container because 60 isn't divisible by 24 or 25. Just make it 30 or 60fps. It
825.
by redox99 3y ago
When you name something, then it becomes a flamewar about that something. But generally speaking, anything that the US/"San Francisco" left wing ideology deems "bad" is generally unwelcome.
826.
by redox99 3y ago
Yes, that's why I said its more permissive, not that Twitter is fully permissive.
827.
by redox99 3y ago
On Mastodon simply discussing certain topics will get you banned, and instances that don't ban those users get defederated. I prefer Twitter in the sense that it's more laissez faire in terms of what kinds of speech get you banned
828.
by redox99 3y ago
> The first crash was caused by bad design That's underplaying it. The design was negligent, and the whole "secret" MCAS is insane.
829.
Volvo Cars Adopts Tesla's NACS Port in the United States, Canada and Mexico(media.volvocars.com)
75 points by redox99 3y ago|167 comments
830.
by redox99 3y ago
Does this create "new" music for the filled parts, like some AIs (MusicLM, MusicGen, etc) do? Or just repeat in a more seamless way?
831.
by redox99 3y ago
It's interesting how much backlash something that was already well understood generated. * FSD Beta is a level 2 system. This means it requires constant supervision. This crystal clear to every FSD Beta driver, the car will both monito
832.
by redox99 3y ago
CNNs are still relatively competitive at vision tasks.
833.
by redox99 3y ago
If it's "absolutely trivial" to show that LLMs don't have the capacity to form thought, then please publish a paper proving that. So all the "stupid" people studying LLMs that can't come up with such trivi
834.
by redox99 3y ago
If you ask for sauna, nudity seems expected to me, not "unsolicited"
835.
by redox99 3y ago
Eating microwave popcorn every single day seems kind of insane lol
836.
by redox99 3y ago
That'll probably really fuck up a coffee grinder, so make sure you don't use it for coffee.
837.
by redox99 3y ago
Latency is mostly a matter of hardware, model size and small tweaks. LLaMa running locally is very fast. And current LLMs could definitely be finetuned to interrupt conversations like you mention.
838.
by redox99 3y ago
That leaderboard is incorrect https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderb...
839.
by redox99 3y ago
They said 1 or 2 bucks an hour. You can get an A100 for that.
840.
by redox99 3y ago
Last year after I posted a innocuous, benign comment on a certain subreddit I was permanently banned from that subreddit because I was allegedly "ban evading" (I wasn't, I have a single account). I contacted the mod from that

More ›