Following this. my plan has been to start simple and eventually build out the inference stack piece by piece, however simple. Take some open weights model and from first principles build kernels and a harness for it.
Given that AI is already so good at swe at this point, it feels harder to get a callback because they're probably super AI pilled and I have no idea how good AI is at writing the things that these guys need for performance (I mean, judging from my local work, they seem pretty good at what I'm trying to learn, but I am a beginner in the field). I've received no callbacks either though so I don't know.
How GPT, Claude, and Gemini are actually trained and served – Reiner Pope: https://www.youtube.com/watch?v=xmkSf5IS-zw
Following this. my plan has been to start simple and eventually build out the inference stack piece by piece, however simple. Take some open weights model and from first principles build kernels and a harness for it.
Given that AI is already so good at swe at this point, it feels harder to get a callback because they're probably super AI pilled and I have no idea how good AI is at writing the things that these guys need for performance (I mean, judging from my local work, they seem pretty good at what I'm trying to learn, but I am a beginner in the field). I've received no callbacks either though so I don't know.
Since Anthropic trains on Sagemaker on AWS, what about some Cloud Certifications?