Where the AI market is talking

Research acceleration: The view inside OpenAI

That was Anthropic.

Outrageously Small Neural Networks: 6,616 tok/s on One Intel AMX Core [pdf]

I think we should revisit outrageously small neural nets. I needed a cheap model that runs at over 10k token/sec on a single CPU core for some data processing. So I gave Anthropic claude code a pile of tokens to build one. It made three discoveries that I thought were interesting: 1) One Intel AMX core can train a 3M active parameter MoE foundation model at 6,616 tok/s on 4.91B NVIDIA Nemotron tokens in a few days. 2) That model shows emergent in-context copying, positional analogies, and basic arithmetic after about 250M tokens. 3) The foundation model gives large gains in downstream SFT, and the training & eval loss keep going down all the way through 4.91B (and likely beyond). Claude is not as good as a great MLE at debugging MoE. It made a bunch of bone headed mistakes, but …

I'm a seeing-eye dog for a computer

Sure, they exist, but online they are simulatenously unable to apply any logic and then surprisingly coherent and intelligent in the next reply.

Ask HN: Model sycophancy benchmarks?

Closest I know of are SycEval and the sycophancy evals in Anthropic's 2023 paper, both built on a user pushing back at a correct answer.

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

> Keyword is autonomous [0]. Excel isn't. No, I didn't forget it. Just as OpenAI made up a definition to support their claim of AGI, so did I.

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

Do we actually know that OpenAI is in serious financial difficulty? https://danluu.com/zitron/ HN discussion: https://news.ycombinator.com/item?id=49526069 Shower thought: I wonder if this whole "OpenAI is in serious financial difficulty" thing is a psyop so people will relax and avoid doing what's necessary to stop them.

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

Keyword is autonomous [0]. Excel isn't. And I would argue LLMs aren't either. A lot of economically valuable tasks involve manual labor. Can LLMs fix your plumbing? It would help if OpenAI would define what 'most' means and give some examples. Remember they said their LLMs are like a phd level brains? It's just hype and bullsh!ttery. Don't get me wrong, genAI is great and really useful, but it's not AGI imo. And there is no consensus what AGI means, so the industry should come up with an agreed definition first. [0] To me autonomous means self directed, e.g. we humans look at things and then decide which problem or thing to work on. LLMs don't do that. We have to tell them. Yes they might work on a sub task autonomously, but still require human input later on (review), because we can't fu…

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

doing so well that they need to slow down down.. https://www.businessinsider.com/openai-chief-scientist-ai-ri... At this stage, it feels like OpenAI and nvidia is openly negging and lovebombing OpenAI to ensure the public is in a constant state of uncertainly, thus ensuring paralysis in democracy while still going hell for leather in the background. I suspect their intent is to ensure they're the last man standing.

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

Well it's certainly informed, as he has information on behind-the-screnes data we don't have about OpenAI But certainly it is not unbiased

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

Nvidia is not a competitor of openai, but a partner

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

> OpenAI defines [AGI] as "highly autonomous systems that outperform humans at most economically valuable work." Well in that case: Excel reached AGI decades ago. It outperforms humans at the single most economically valuable task in the world: adding up numbers in a spreadsheet. Congratulations, Microsoft :)

Is mathematics about to enter the conservatory?

Our AI system don't just do things. They do it because someone asks. A lot of the boom in AI maths results are solving human formulated problems mostly for PR benefit. Once the novelty passes, would Anthropic or OpenAI keep spending? And without mathematicians to ask the right questions and able to appreciate the results, why would AI driven research continue?

Is mathematics about to enter the conservatory?

Our AI system don't just do things. They do it because someone asks. A lot of the boom in AI maths results are solving human formulated problems mostly for PR benefit. Once the novelty passes, would Anthropic or OpenAI keep spending? And without mathematicians to ask the right questions and able to appreciate the results, why would AI driven research continue?

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

> OpenAI is in serious financial difficulty Is this based on fact or your speculation?

Ask HN: Fable hacked my piano, can I release the results?

Maybe send this question to the Anthropic legal team. I'd be curious if you get an answer and what it'll be.

Ask HN: Perplexity Pro vs. Claude – what are you paying for?

I don’t think I’d even have perplexity on this list. Claude or OpenAI. I’m not sure what the value of perplexity is these days.

OpenAI 2025 financials $38.5B loss ahead of IPO

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI

In this case, Jensen Huang's statements can't be taken as informed, unbiased opinion. OpenAI is in serious financial difficulty and Nvidia cannot afford for OpenAI to fail. Nvidia has provided significant direct and indirect financial assistance to OpenAI; this is just another aspect of that assistance.

I'm a seeing-eye dog for a computer

> I like to think of training and improving AIs as bringing freedom to the world. Yes, when I think of OpenAI and Anthrophic and Google and Meta and any AI labs and their intentions, i cry a single tear for how these great instituitions are working so hard to bring freedom for humanity.

US gov sides with OpenAI on issue of training LLMs on copyrighted material

As you said earlier, OpenAI can do nothing wrong whatsoever with the data/computation they have for their “greater purpose”. We disagree on this point.