saranormous

A lot of people who can’t code can now use LLMs to generate usable code, so it doesn’t seem totally wrong? Example: https://twitter.com/emollick/status/1649477099353411585?s=46...

host here. thanks for sharing Tim. highlights for me: Nvidia founding story, his pov on the durability of transformers, the two ways he runs nvidia (shipping reliable chips and exploring new fringe applications), and the apps he’s interested in (eg climate modeling, bio).

feedback and guest suggestions?

what’s the false statement?

discusses the foundation model landscape, how they at nvidia decide to train models (or not), operating model at nvidia, durability of transformers, the h100 chip, what’s next

discusses the foundation model landscape, how they at nvidia decide to train models (or not), operating model at nvidia, durability of transformers, the h100 chip, what’s next

this is awesome. is there good research explaining methodology of feedback collection/desired dataset (beyond just relative human preference?)

is there good background reading on ecDNA somewhere?

nutmeg overdose? where’s the research

do you have to review/publish the “chore” output for this to work? Thinking about how this applies to writing, when I don’t have the creativity every day, but it’s probably still good practice and useful to have the file of ideas

startups aren’t a game of averages. the averages are definitely worse than the FAANGs. choosing well (and getting lucky) are paths to “definitely better.”

It also depends on what your professional goals are (growth, leadership, impact, fun, next opportunities) assuming they are not only monetary.

this area is "strangely underexplored" because it could encourage the exploitation of desperate young people. to be fair, this market already exists (donors already earn money for plasma donations), it's just about whether or not a new use case dramatically grows the market

Agree that everyone's sense of time is changing. I struggle with this - I can't keep myself from speed-reading/skimming longer form text.

I do think this is problematic, and not all good. A great deal of knowledge still requires significant length of text to convey -- sure, the retrieval cost for any individual limited concept on the internet is much lower, but you're not going to develop a rich understanding of, as a hyperbolic example, the history of China, from one article or from Wikipedia.