My JPM banker could not explain why all of my boring energy and industrial stocks got pummeled in 1 day. This is a super helpful read into what we do and do not know. It's wild that you invest in a portfolio with "limited tech exposure" that ends up being 75% exposed to tech via AI compute costs (or potential lack thereof).
Small vs. large models: There's a third option. A team of small models, an ensemble.
We've been working on this over the last year, and we can demonstrate on leading benchmarks that ensembles of small and efficient models outperform the best large models on quality, and performance, with 10-100x on the last two KPIs.
You've outdone yourself with this one...assuming you didn't use DeepSeek to write it??? Just kidding. Well done, sir. (P.S. - Try asking DeepSeek about Tiananmen Square ;)
Are the computer chips that are used to pre-train an AI model the same type of chips that are used for inference?
My JPM banker could not explain why all of my boring energy and industrial stocks got pummeled in 1 day. This is a super helpful read into what we do and do not know. It's wild that you invest in a portfolio with "limited tech exposure" that ends up being 75% exposed to tech via AI compute costs (or potential lack thereof).
Small vs. large models: There's a third option. A team of small models, an ensemble.
We've been working on this over the last year, and we can demonstrate on leading benchmarks that ensembles of small and efficient models outperform the best large models on quality, and performance, with 10-100x on the last two KPIs.
More info at a1-labs.co
Great post.Would love your thoughts on Nvidia future in this
You've outdone yourself with this one...assuming you didn't use DeepSeek to write it??? Just kidding. Well done, sir. (P.S. - Try asking DeepSeek about Tiananmen Square ;)
Amazing post!
One of your best
The Man
🔥. Excellent post.