A short note on DeepSeek v3.2
This is a linkpost for https://kunvarthaman.com/posts/deepseek-v3.2.html
Karanoia
Kunvar Thaman's personal blog
Subscribe:.rss.atom.json.md.m3u.pls
Live Last read · last published · next check
Latest episodes
What I read in November
AI is probably not a bubble, but it does worry me when I see how Luma - the multimodal company - is raising $900 million to build a 2GW datacenter in Saudi Arabia.
Notes on resistance training
I started lifting weights for the first time in 2025. Here's some reflections.
Attention is a differentiable lookup
Expanded on what Joshua said here. This is a link post for https://kunvarthaman.com/posts/attention-is-a-differentiable-lookup.html
A systems level understanding of LLM inference
This is a link post for https://kunvarthaman.com/posts/llm-inference.html
Why I think human capabilities will eventually look spiky
This is a link post for https://kunvarthaman.com/posts/infrastructure-is-important.html
How does power flow through an AI data center?
This is a link post for https://kunvarthaman.com/posts/data-center-power-flow.html
How are the GPUs in a data center cooled?
TThe goal is to move heat from silicon to sky with as little energy as possible.
Environments are everything
This is a linkpost for https://kunvarthaman.com/posts/environments-are-everything.html
The Human Evaluator’s Goodbye
This is a linkpost for https://kunvarthaman.com/posts/goodbye-the-human-evaluator.html