GaggiX

24 ms·

391.
by GaggiX 2y ago
>I'm not sure if that's truly AI It has always been a LLM. There is no human typing at insane speed to the TTS.
392.
by GaggiX 2y ago
>they could make themselves into an AI easily since it's a rigged 3D asset but of course it would be boring compared to the real thing The most popular vtuber on Twitch is an AI tho
393.
by GaggiX 2y ago
It's the only reasoning model they evaluated to be honest.
394.
by GaggiX 2y ago
The API pricing is almost exactly double the deepseek ones.
395.
by GaggiX 2y ago
The model looks quite a bit better in the benchmarks so unless they overfit the model on them it would probably perform better than deepseek.
396.
by GaggiX 2y ago
Hopefully they will finetuning it using RL like DeepSeek did, it would be great to have more open reasoning models.
397.
by GaggiX 2y ago
They are two different reports, and I imagine that the concept of "revealed preference" exists outside of the US too ahah
398.
by GaggiX 2y ago
>80%+ of teens own iPhones In the US*
399.
by GaggiX 2y ago
Or because the results of these models are so transformative that you could pass it off as fair use.
400.
by GaggiX 2y ago
>I don't have to obey the current law, There is currently no law that states it is illegal to train a model on copyrighted work.
401.
by GaggiX 2y ago
>"Can't" or "don't want to"? Can't. Even if someone has the money (I truly doubt), you can't contact millions of copyright owners (as you report).
402.
by GaggiX 2y ago
You can't license 20T of tokens, I guess it's hard to grasp how big these datasets are.
403.
by GaggiX 2y ago
>Either they should make it illegal and establish precedent At that point we just wait for the next Chinese open source model.
404.
by GaggiX 2y ago
DeepSeek R1 is still censored offline, you are probably talking about the llama distilled version of Deepseek R1.
405.
by GaggiX 2y ago
From what I see, the Deepseek R1 model seems to be better calibrated (knowing what it knows) than any other model, at least on the HLE benchmark: https://lastexam.ai/
406.
by GaggiX 2y ago
Now they need to finetune it like R1 and o1 and it will be competitive with SOTA models.
407.
by GaggiX 2y ago
DeepSeek R1 is o1 but free to use, open source, and also distilled on different models, even the ones that could run on your phone so yeah.
408.
by GaggiX 2y ago
AI slop is not a neologism, slop means low quality content, and it's often used when talking about drama youtuber or brainrot content, "AI slop" is just low quality AI content, not a new word. I say this because some people s
409.
by GaggiX 2y ago
I wonder if the decision to make o3-mini available for free user in the near (hopefully) future is a response to this really good, cheap and open reasoning model.
410.
by GaggiX 2y ago
Don't they have already achieved mass adoption? And I'm talking about LLMs in particular, because AIs in general like the ones used by Instagram filters and the TikTok recommendation algorithm are already use by billions.
411.
by GaggiX 2y ago
Oh I see, the paper is out, I read "(arXiv coming soon)" and though it wasn't released yet.
412.
by GaggiX 2y ago
>Is there a text-only evaluation of the non-Deepseek models? Not that I can see but it would be cool to have, maybe the paper will a more complete evaluation.
413.
by GaggiX 2y ago
It really shows how good Deepseek R1 is (even though it was evaluated only on text-only questions). The results are shown here: https://lastexam.ai/ EDIT: the text-only evaluation of the models shown in the paper gives o1 a
414.
by GaggiX 2y ago
Their paper goes into the details: https://arxiv.org/abs/2501.12948
415.
by GaggiX 2y ago
>when that is not possible with LLM. According to who? You?
416.
by GaggiX 2y ago
> and the simple fact that it is incapable of making anything good. Weird cope about AI, not a very smart one.
417.
by GaggiX 2y ago
This model is kind of the bottom of the barrel. Also there are models that specialize in pixel art: https://www.pixellab.ai/
418.
by GaggiX 2y ago
You can do it on one single GPU but you would need to use gradient accumulation and the training would probably last 1-2 months on a consumer GPU.
419.
by GaggiX 2y ago
I guess that depends on how the webspider is configured, I doubt the curation is done in real-time while scraping.
420.
by GaggiX 2y ago
The model can be tested here: https://build.nvidia.com/igenius/colosseum_355b_instruct_16k

More ›