Caracas

@HyperTechInvest The only thing that could destroy AMD apart from a bubble popping is if $SMIC and Huawei improve their chips at a cheaper price than AMD. But that isn't very likely

@HyperTechInvest It's a great company, it shouldn't have a big Mrkt cap diference with $NVDA Maybe It keeps going. But I think we are getting into ridiculous valuations, I expected AMD to be at this price when the CUDA moat got conquered. Bit bubbly on my opinion, but fundamentals are still👌

$AMD continues to make progress on the software The company introduced vLLM-ATOM, a plugin designed to make major AI models run better on $AMD Instinct GPUs, including MI350 and MI400 Developers can keep using the same vLLM commands, APIs, and workflows, while ATOM works in the background to improve performance on AMD hardware, requiring no new tools or complex configurations It also gives users instant access to AMD’s latest optimizations, including FP4 support on MI355X, rack-scale inference on MI400, fused attention, custom AllReduce, and other kernel improvements ATOM also acts as an innovation sandbox, where AMD can test new optimizations before they are later added to the main vLLM ROCm backend

@HyperTechInvest I bought AMD at 106 thanks to you dude, thanks, still holding but planing an exit. But really look at that slop, It looks worse than Meta Image gen, so lazy. How a almost 1T company doesn't have more compute to use a better image gen model?

@Metaculususer I could never refer to AI as slop. I have the utmost respect for what’s pumping all my semi stocks

@austinc3301 Claro checa tu Dm

@Metaculususer y no tengo ni idea por que llevamos hablando una hora en inglés

Gemini 3 is disappointing. We are officially in a AI winter and this bubble could pop at any second

@austinc3301 But really Thanks you so much for your time. In fact, Me and My friends have a species of benchmark that the AI fails most of the times, I would love to share it to you if you are interested.

@austinc3301 That's some great information, I think you have better knowledge of the newer benchmarks that I have. I personally don't give the institutional benches that much weight. because I see them as a byproduct of a huge asset bubble.

@austinc3301 I think It's really important, AGI needs to do everything, For example to write fiction books. I have tried so much making them with frontier models and nothing.

@austinc3301 That point in the circle is Llama-4 scout, that attempted 10m tokens and failed. I just baffles me, how little attention everyone gives to the context window.

@austinc3301 @Pirat_Nation Yeah, My bad, I could argued my point better, I'm sorry

Dario is quite clearly not a neutral party here, he's deeply conflicted, which is why I've cited neutral third parties and not the leader of a US frontier lab, but I'm correcting the record since you made it sound like Dario was arguing in favour of the efficiency of Chinese models, when he was doing the opposite

A court in Chile has allowed Amazon to build a new data center in Huechuraba near Santiago after local residents and environmental groups failed to stop the project. This forms part of Amazon’s plan to invest more than $4 billion in data centers across Chile. People were worried about high electricity and water use, plus new power lines that could affect the hills, but the project passed all environmental checks and any new power lines will be reviewed later. Amazon says the data center will use special designs to protect the environment. Chile is becoming a popular spot for big data centers because of its clean energy but buiding data centers are destroying the nature

@austinc3301 If the labs were open we would know that for sure. I know that Anthropic used Claude code to code Claude code. But to tweak the weights, I think not

@Metaculususer it is already the case that frontier labs have offloaded a large fraction of their software engineering work to their own frontier coding agents

@austinc3301 What's your favorite benchmark that benches the context window? How we haven't hit a wall when we haven't even pushed 1m token usable context?

@austinc3301 I will consider your opinion

@austinc3301 I think we need to move past them. It's time. Like Yann LeCun said, If LLM's started to solve by themself the structural problems they have, That would be something that I would be into, not benchmark maxxing

@austinc3301 Bam! We share that opinion. The truth is that I don't see benchmarks as and end all be all. I have seen it too many times. AI has saturated so many benches already.

@austinc3301 Because I remember clearly, Sam Altman like 1 year ago claiming 'o3-pro' conquered ARC-AGI. And everyone was going crazy and all What happen after that, nothing, o3 didn't cure cancer or make any solid scientific improvement

@austinc3301 Can I ask you something? What's your opinion on ARC-AGI benchmarks?

@austinc3301 @Pirat_Nation Yeah, I guess I was wrong to cite it, because if you trust him. I just interpreted the cite as a win of Chinese open source. Is very very very rare for Dario to speak of other models. That's why.

@Metaculususer @Pirat_Nation again, you're the one who cited dario, not me!

@austinc3301 @Pirat_Nation That doesn't change that the Chinese have a great hold on The AI academic space. If you don't like that subjective analysis of mine. You can always look at csranking

@austinc3301 @Pirat_Nation Well, I agree with the quote, I think that maybe the issue is the difference in the amount of trust we put on Dario. If someone trust him blindly, then yeah, Chinese models aren't a big deal. I do not, I see him as a bubble enabler. And this statement is a 'This model is great'