This approach is good as long as the tech stack is open source and portable to other platforms. Otherwise, no matter how good a company/CEO seems ATM, you are ultimately at their mercy if they decide to raise prices significantly.

By using an open, interoperable tech stack, you maintain the freedom to switch to another cloud provider at will.

This shared fluid power also creates a compelling reason for cloud providers to remain honest and competitive in their dealings with customers.

You don't get it.

For most companies free users are just a source of potential paid customers. Such companies squeeze the free users to force them to upgrade. For Cloudflare the millions of free users strengthen their negotiating power with ISPs around the world. We provide value to Cloudflare just by being Cloudflare customers. It is possible that Cloudflare might get a CEO who doesn't understand this, but possible doesn't mean likely.

In any case, I've built my website with Astro, pulling in the Cloudflare integration as a dependency. If I wanted to switch to Vercel or Netlify or whatever else, Astro makes it easy. As for database, others offer managed Sqlite.

If all else fails, I'll ditch the few dynamic parts of the website and deploy the bulk of the site as static html to Github Pages or something.

Would be interesting if someone could detail the approach to decode the pre-post processing steps before it enters the model, and how to find the correct input encoding.

Boils down to "use Frida to find the arguments to the TensorFlow call beyond the model file"

Key here is, a binary model is just a bag-of-floats with primitively typed inputs and outputs.

It's ~impossible to write up more than what's here because either:

A) you understand reverse engineering and model basics, and thus the current content is clear you'd use Frida to figure out how the arguments are passed to TensorFlow

or

B) you don't understand this is a binary reverse engineering problem, even when shown Frida. If more content was provided, you'd see it as specific to a particular problem. Which it has to be. You'd also need a walkthrough by hand about batching, tokenization, so on and so forth, too much for a write up, and it'd be too confusing to follow for another model.

TL;Dr a request for more content is asking for a reverse engineering article to give you a full education on modal inference

> It's ~impossible to write up more than what's here

Except you just did - or at least you wrote an outline for it, which is 80% of the value already.

The more impolite version of this basically says "If you can't figure out you're supposed to also use Frida to check the other arguments, you have no business trying." I agree, though, wrote a more polite version.

> TL;Dr a request for more content is asking for a reverse engineering article to give you a full education on modal inference

I don't understand what you mean: I have no clue about anything related to reverse engineering, but I ported the mistral tokenizer to Rust and also wrote a basic CPU Llama training and inference implementation in Rust, so I definitely wouldn't need an intro to model inference…

You're also not the person I'm replying to, nor do you appear in any of this comment chain, so I've definitely not implied you need an intro to inference, so I'm even more confused than you :)

I share the sentiment of the person you're responding to, and I didn't understand your response, that's it.

Whether you like dark mode or not, I think the most important (and least divisive) fact we can all agree on is simply: As users, it's nice to have a choice.

Of course, most of the affected extensions have "AI" or "VPN" in their title.

This one stands out as being the kind of thing I could have used:

> GraphQL Network Inspector

It's a really useful extension if you work with GraphQL endpoints, and fully open-source: https://github.com/warrenday/graphql-network-inspector

It's been fully patched since: https://github.com/warrenday/graphql-network-inspector/issue...

"Each consumer could receive up to $20 per Siri-equipped device covered by the settlement, although the payment could be reduced or increased, depending on the volume of claims. Only 3% to 5% of eligible consumers are expected to file claims, according to estimates in court documents."

"The alleged recordings occurred even when people didn’t seek to activate the virtual assistant with the trigger words, “Hey, Siri.” Some of the recorded conversations were then shared with advertisers in an attempt to sell their products to consumers more likely to be interested in the goods and services, the lawsuit asserted."

Most people are happy living in a fair ecosystem - it's only the 1-2% of the population that seek control, money and power that start trying to exploit the system.

Only if we let that minority keep manipulating the system without consequences, it becomes the driving market force that the rest of the population also feels they have to comply to, to go along, as it already has happened in finance, academia, etc.

> Most people are happy living in a fair ecosystem

For varying and self-serving definitions of fair. (Almost everyone in the rich world is in an unfairly-advantaged minority.)

I'd push back against the 1-2%. I think the reality is, 1-2% is the group of people who will exploit the system and more importantly, have the means to do so. But the number of people who would exploit the system is probably quiet a bit higher, but it doesn't matter because they don't have the means to do so.

doesn't mean why shouldn't fight back. That's exactly why we need research projects like these: to maintain the balance.