> Given the cost the company will incur to install the capacity for dynamic pricing in its stores, it would be corporate malfeasance if they did not believe doing so would not only recoup the cost, but add profit as well.

dynamic pricing and scalping are the same mechanism.. As an economy student I like the efficiency, as a consumer I hate the asymmetry and inevitability

Most ML practitioners use L1/L2 daily without realizing they're making Bayesian prior assumptions. Gaussian prior = Ridge, Laplace prior = Lasso. Once you see it that way, "choosing a regularization strength" is really "choosing how informative your prior is."

Novo already guided for its first revenue decline in modern history (adjusted sales down 5-13% for 2026). Stock is down 75% from June 2024 peak. The problem isn't just generics, Lilly's tirzepatide beat CagriSema in the head-to-head, and Lilly has orforglipron (oral, way cheaper to manufacture) coming. At 11x forward earnings Novo is pricing in catastrophe, buuuuut.. the structural problems are real this time.

so the model underneath is swappable, but the layer that remembers what you asked it last week isn't. Persistent context is the switching cost here. Anthropic just extended that from desktop to mobile.

FY2026 Pentagon's AI budget jumped 7x to $13.4 billion, now larger than Anthropic's annualized revenue. Once you're on an IDIQ contract with classified compute, good luck switching. Security clearance processing alone takes 243 days. Palantir figured this out years ago, 55%+ of revenue from government now.

Tried to summarize some of that in my recent writeup: https://philippdubach.com/posts/when-ai-labs-become-defense-...

Great write-up. But is the classified-network moat meaningful when the product is inference on a foundation model? The Last Supper primes locked in because the technology itself was bespoke.

Already posted here earlier: https://qht.co/item?id=47376119

The post you linked to was made after this post.

my bad, I just saw 2 hours ago on this one - either way they could be merged

I think my submission about this post was selected to "second-chance" pool by HN Moderators. Hence it's being shown again. Thanks for the heads up!

I run a Hugo blog and I get more interesting referral traffic from Kagi's small web index than from Google at this point. 5,000 curated sites is small enough to be useful most "indie web" directories are graveyards unfortunately..

Been spending a bunch of time lately trying to figure out why these ~120B MoE models keep beating much larger dense ones.

With Mistral it's 128 experts but only 4 active per token, so any given forward pass is like 6B params. That's a very different kind of model than scaling a dense transformer bigger. Also wrote a little post on where I think this is going: https://philippdubach.com/posts/the-last-architecture-design...

Worth noting they're filing individually instead of joining a class action.

Same economics as organic food labeling imo. Starts as a genuine quality signal, turns into a price premium, gets gamed until the certification means nothing.

The harder problem will be (or already is) that most products will be partially AI-assisted and a binary label can't really capture "we used AI for the layout but a human drew every illustration." Good luck defining that boundary tbh.

The logo could be a circle-slash "no" symbol over a six-fingered hand holding an em-dash, so the visual design can match the solidity of the measure.

> The harder problem will be (or already is) that most products will be partially AI-assisted and a binary label can't really capture "we used AI for the layout but a human drew every illustration." Good luck defining that boundary tbh.

It’s simple. If you used AI, you don’t use the badge.