Good luck, getting rid of PDFs is going to be as hard as migrating from JPEG everywhere.

RLAIF is a good place to start reading.

Claude will also help you with (mostly good advice) if you ask something like “Research and help me make the most effective plan to train a smaller student model to be better from a teacher model”.

I actually was doing an experiment with a GLM->Gemma E4B for fun, and Claude kept on suggesting I should also add Claude Opus as a teacher lol, suggesting techniques I haven’t heard of like thinking inversion (train a small model to deconstruct summarised thinking into detailed native thinking format of the student).

So I can absolutely see and understand the concern around Fable’s frontier LLM development mitigations, but their approach of silently degrading is completely wrong and dangerous.

AI classifiers, like all AI, can make mistakes, and it’d only be a matter of time before it mis-fires and silently sabotaging a university’s HPC cluster for physics simulations or something because the shape looks like DeepSeek or whatnot to a dumb fast classifier.

Why use VPNs? Just use a public cloud like AWS, or something like Linode and Vultr and all that.

Developers use devboxes on these clouds all the time, it’s totally normal behavior.

Most people buying these Chinese resold tokens are probably using it for coding anyway, so you don’t want the Claude.ai chat system prompt.

Residential IPs don’t even matter. Developers use devboxes, use Claude Code CLI on servers from just about every cloud, etc.

There’s probably a decent volume of customers who just buy Claude Max and spend most if not nearly all of their sessions via Claude Code, and it’s not uncommon for power users to be working on multiple concurrent projects/tasks/codebases at the same time.

How do you really block this without also impacting your core market of developers?

If you’re doing evals, you’re basically doing RLAIF without training a model; just looking at the results.

Fundamentally it is very difficult to stop this while still making your AI models useful.

Similarly, if you did a corpus study on bioRvix to summarize recent science findings — you could use the same questions and answers to fine tune a model.

There is no way to communicate information at scale to companies through the API, for anything approaching a real application, without that information forming a corpus another model can be trained on.

But it wouldn’t be the first time they broke a model:

Their “guardrails” that cause it to reject user prompts also means it relies on its pop science summary of medicine to tell you why bioRxiv is wrong rather than accurately summarize the papers.

They’ve successfully created a smug, argumentative average of the internet which refuses to even consider it might be wrong or that it’s reading a science paper which is based on measurements and not vibes — but why would I pay for that?

I get it for free online.

Don’t trust my experiences as fact since it’s a bit opaque, but I believe 20x only offers 4x the 5hr session limits. The weekly limit is still 2x, which is the same as the price increase.

And it’s perfectly normal to be running Claude Code on EC2, a VPS, etc. I do it all the time!

You block clouds, you block devboxes and your customers.

I use my Codex and Claude Code subs on like 4-6 different servers, ranging from AWS to Vultr to Linode etc.

That’s a major and legitimate use case for developers, Anthropic can’t just block data center/hosting IPs because their actual customers use them on data center/hosting IPs.

Now consider what will happen if your pattern of queries and context history triggers a pattern that makes it obvious it's some API key being used by multiple different entirely unrelated people on totally different things, or any other pattern of use that makes it obvious it's being used for distillation.

Two parts here.

First, well-calibrated systems for detecting API compromise is a good thing (or good intent at least). Credential malware is exploding.

Second, the challenge is that significant amount of genuine work — such as evals — seems practically impossible to distinguish from generating RLAIF outputs.

To a large extent, all these datacenters are being built because people want tokens. Do you use something like Opus or GPT5.5 for coding? Well, you're part of the demand.

If you don't, then yes, as a PC enthusiast I'm sad as well. But like, it's a little bit ironic to be complaining about RAM prices if you've got 5 sub-agents hacking away.

If someone is using heavily underpriced tokens I don't blame them much either. They're not funding the hardware purchases, super-investors that want to control the AI world are.

Generally 97.5% token agreement is very positive. Like the article explains, the difference isn’t the model thinking the capital of France isn’t Paris, but rather maybe saying “The capital of France is Paris” instead of “Paris is the capital of France”.