You can just say "I didn't read the Brookings article and some of the summaries of it were over hyped and inaccurate." instead of making weird personal attacks.

Show Post

Also I dont think it is splitting hairs to differentiate a claim about what is happening *now* and what someone's model say will happen in the future.

Show Post

The question isn't if inference is profitable right now though. The question is how replicable profitable inference machines are. It doesn't matter that you develop the cotton gin first--it matters that you recover the development cost before everyone else starts ginning up their own.

Show Post

Now, I use Claude Code at work, and I haven't done a detailed cost analysis. I think Anthropic will get a decent amount of lock in for the same reason a ton of enterprises pay money to Microsoft even if they can run Linux for free. They're used to it, and their workforce isn't that text savvy

Show Post

The challenge of course is the red queen's race for frontier model superiority. If your frontier model falls behind, the margins you can charge go down. These financial models work because the AI company says "well if there is a contraction I can cut costs by halting training and laying off R&D"

Show Post