Please don't skip the rundowns... Hugely helpful and insightful for those of us not living & breathing in the depths of AI but wanting to keep abreast of progress without long essays ;)
Congrats on 5k! Personally I find you insightful. I'd hope that you skip the rundowns (it's a waste of your talent) and focus more on essays. Your rundowns are exquisite, but yeah. Synthesis and summary with an opinion of expertise in crisp essays I think is where you really shine.
Already one of the best AI Newsletters out there, just my two cents.
Nvidia's lead in compute is only temporary. In 12-36 months there will be either:
1. Purpose built non-gpu chips like groq that are cheaper, less power hungry and more space efficient.
2. Mobile devices and laptops will have llm specific chips. Therefore more compute will happen on device. This will be possible because llms will become more efficient and focused on solving narrow but urgent problems.
I agree about more LLM inference moving to on-device, but I’m not entirely convinced that Nvidia is going to be replaced inside of the datacenter. It’s not just about the GPUs, it’s the entire software stack that comes with them. Google has had TPUs for a little while now but doesn’t seem to be making a dent in Nvidia’s dominance.
Nog going to lie. Nvidia does software very well. Last month i purchased my first gaming pc in 18 years. It has an RTX 4070. DLSS and frame generation are pure ai magic.
They increase my framerate by 30-50% and double my resolution. Way superior to amd fsr. It's like improving a car's lap time by putting racing tyres on it.
I am curious to see how apple's own llm will run on their devices. However, the performance of Siri doesn't give me much faith!
Congrats on 5k, great newsletter!
Please don't skip the rundowns... Hugely helpful and insightful for those of us not living & breathing in the depths of AI but wanting to keep abreast of progress without long essays ;)
Thank you for the feedback!
Congrats on 5k! Personally I find you insightful. I'd hope that you skip the rundowns (it's a waste of your talent) and focus more on essays. Your rundowns are exquisite, but yeah. Synthesis and summary with an opinion of expertise in crisp essays I think is where you really shine.
Already one of the best AI Newsletters out there, just my two cents.
Nvidia's lead in compute is only temporary. In 12-36 months there will be either:
1. Purpose built non-gpu chips like groq that are cheaper, less power hungry and more space efficient.
2. Mobile devices and laptops will have llm specific chips. Therefore more compute will happen on device. This will be possible because llms will become more efficient and focused on solving narrow but urgent problems.
I agree about more LLM inference moving to on-device, but I’m not entirely convinced that Nvidia is going to be replaced inside of the datacenter. It’s not just about the GPUs, it’s the entire software stack that comes with them. Google has had TPUs for a little while now but doesn’t seem to be making a dent in Nvidia’s dominance.
Nog going to lie. Nvidia does software very well. Last month i purchased my first gaming pc in 18 years. It has an RTX 4070. DLSS and frame generation are pure ai magic.
They increase my framerate by 30-50% and double my resolution. Way superior to amd fsr. It's like improving a car's lap time by putting racing tyres on it.
I am curious to see how apple's own llm will run on their devices. However, the performance of Siri doesn't give me much faith!
Congratulations on crossing the big 5K mark! Well deserved!
Thanks Daniel! Honestly you've been a big inspiration.
"I don’t think you can productively build in the American regulatory system"
Can you say more about that? Do you mean specifically within healthcare?