Hi, I’m the author of the article and the person who designed and implemented the specs transforms. At Stainless we’ve been using an internal version since around the end of 2022 that grew organically to 50+ commands. They have been commonly used by our largest customers to workaround various spec issues without necessitating upstream changes. This year we took the time to reflect on that experience and designed a small set of operations that are more flexible, more intuitive, and compose well together. And made them available to everyone. That’s the version presented in this blog post.

Happy to answer any questions you would have :)

It's the personal blog of a famous internet person

VPNs is such a shady industry, I really don't blame companies for blocking them. 99% of people using a VPN do not understand they are giving all their traffic to a random company in a jurisdiction they can't verify, with privacy policies they didn't read, operated by people they've never heard of, based entirely on marketing claims and affiliate-driven "reviews".

There is pretty much zero regulatory oversight. The ownership structure of VPN companies is opaque, often owned by holding companies. For example Kape Technologies owns ExpressVPN, PIA, CyberGhost, etc.

... on the other hand, there's the fact that millions of people can't meaningfully participate on the Internet without one.

Many VPN providers may be shady, but none as shady as the governments and corporations that drive people to using them.

now I feel old...

Do we know for how much that type of content sells? Not that I'm interested in entering the market, but the economics of that kind of thing are always fascinating. How much are buyers willing to pay for AI conversations? I would expect the value to be pretty low

They say they remove information from the collected data. They aren't very explicit about what information they remove or not. They also seem to be feeding the data they collect right into the affiliated data broker company.

There are two different markets for this kind of complex data:

- Aggregate (demographic) data is useful for targeting, not just regular ads but also in-person outreach or even just identifying areas with a high density of potential customers; you can also use these insights to then categorize people in your own data set (e.g. when onboarding a new user you might cross-reference their details and find out they're high value just based on the "non-personal" data from the data broker that matches their profile).

- Specific (personal) data is useful for companies like insurances to flag you for risks you wouldn't otherwise have to disclose or they might not be able to request disclosure of; because direct transfer of personal data is the most likely kind to run into privacy law issues this is now often obfuscated by feeding it into AI models (i.e. the AI learns to match the collected data at the data broker to the input data it receives from the data broker's customer but there is never an explicit connection between the two data sets so the data broker can claim it is anonymized/aggregated when in practice it's still granular enough for the AI model to be able to categorize you based on seemingly spurious associations).

Note that the fist case overlaps with the second because "aggregate" usually still means that when looking at a new dataset (i.e. data collected from one person) you can say with some confidence which pile of aggregated data it fits in even if that pile doesn't contain an exact match due to anonymization/pseudonymization. Also note that this means there isn't really any feasible way to "aggregate" data in such a way it can no longer be argued to be subject to the data subject rights of the GDPR unless the data is fully isolated (e.g. total number of monthly visitors of an entire website by year).

I doubt its the actual conversations but the aggregated insights that are valuable.

Think: is my brand getting mentioned more in AI chats? Are people associating positive or negative feelings towards it? Are more people asking about this topic lately?

Sure, but are they willing to pay and if yes how much. There is a meaningful difference between « could be useful » and « valuable enough that we want to buy »

Let's assume that people are discussing medical conditions in these conversations - I think that insurance companies would be pretty interested to get this kind of data in their hands.

The question isn’t if there is some interesting info in that data but if there are some actual buyers. Lots of interesting data exist, so what’s the value of AI chats?

> That’s an unhelpful take, if you expect everyone to be fluent in the language of the country they’re traveling to

I'm myself native french speaker and do hate the French attitude on language. It's extremely patronizing and do not benefit anyone

Hope you have a good rest of your weekend

I love it! Such a cute application :D

I see Zig, I upvote :) Have fun, that sounds like a neat project

Thanks! Hope you have fun with your projects as well!

Yesterday I released https://npmdigest.com, a micro-SaaS to work around the frustrating experience of getting spammed by npm emails "Successfully published X" whenever I release new versions of my packages.

And I completed a pretty long technical article on my personal blog that goes pretty deep into SSE + Postgres + v8 + some linux kernel stuff: https://sam.elborai.me/articles/how-sse-actually-works-deno-...

Some other projects I'm currently motivated by

- pls, my take on what my ideal release automation tool would be (currently deno only): https://github.com/dgellow/pls

- steady, an OpenAPI spec validator and mock server: https://github.com/dgellow/steady