Quad9 has horrible latency and frequent query failures, I hope Mullvad encourages them to improve their routes

Their DoH going down for hours multiple times is what forced me back to cloudflare, i don't trust cloudflare more but at least it works consistently.

Additionally the default of blocking 'malware' doesn't jive with uncensored internet - that should be an opt in and not a default on their flagship address if they want to be taken seriously as unfiltered provider.

If the encrypted endpoint can go down for extended periods and they curate list of 'malware' they are not something that should be considered a gateway to uncensored and open internet.

I frequently get mini failures with quad9 and my DNS client is too dumb to retry with another server (or quad9 returns a valid but bogus response that prevents my client from hoping to the next server) so I had to ditch quad9.

Quad9 also has worse latency but I could live with it, I just can't live with web pages failing to load several times per day(especially hacker News that has their DNS TTL set to 1 second)

It doesn't happen with my ISP's servers nor with CloudFlare or google or even good ol' L3.

I OCR'd my Chinese textbooks and made a stateless MCP that allows me to ground my Chinese language studies according to the textbook only. With this I can start a quiz, understand differences between words that have similar meanings knowing no extra grammar is fed when reviewing. I specifically use it with glm 5.3 as it is the most language specific LLM that understands nuances.

Here's the repository: https://github.com/iodize6399/xuexi-keben

And here's the server itself: https://keben.555420.xyz

That's an awesome example! I'll definitely make use of your repo and idea for my German classes. Did you use a specific OCR method? I'm trying to get something better than tesseract :(

I used Mistral 3. It costed me literally pennies. My workflow was kinda rudimentary but it worked really well. I suggest take a look at this https://github.com/docling-project/docling as well.

I shed many tears after I read the article even though my mom is still alive, I shudder to think of the day I lose her

One can argue, this would be an unintentional tracking of the population by government

One could argue that, but with a properly designed API, it could preserve anonymity while still providing businesses with the necessary validation.

I'm surprised some banks don't provide a CSV export for statements, even if some do, they format it a XLSX file with images and meta descriptions in as header

I don't think OP meant any non-negative reaction towards Raymond but MSFT.

Since you liked Orbis Pictus, you'll definitely appreciate The Way Things Work

Thanks! I'm a big fan of David Macaulay's other books, but I've never liked _The Way Things Work_.

_Castle_ and _Umderground_, though I love.

There are some impressive British children's picture dictionaries, especially the DK Children's Encyclopedia, but also some others that I can't quite put my finger on. A little closer to Comenius compared to that particular Macaulay: more direct and broader in scope and less jokey.

What I most appreciated in The Way Things Work is that he gives a complete working description of the boolean half-adder built out of logic gates, and I think how to connect them to make larger adders. Remember, in our very recent past, it was actually hard for little kids to find explanations of how to do arithmetic with logic circuits? Macaulay did his bit (so to speak) to fix that amidst tons of weird mammoth jokes.

By the way, I've been to one of Comenius's old schools (Herborn, Germany) and grave (Naarden, Netherlands).

What stops the LLM not to iteratively think and expound upon before emitting the final tokens?

Nothing at all. They're not designed to, so they don't. Change that, and they would.

The question is the wrong one. The right question: why aren't frontier models designed to work that way? The answer: it's slow and expensive.

The other answer: that's basically what you're selecting with "Medium", "High" and so on, how many tokens they'll blow on muttering to themselves before they get back to you with an answer. There's more to it, but not that much more.

Reasoning models will frequently backtrack and re-assess what they've said so far. That's one reason test-time scaling is so powerful.

At one point the amount of human induced accidents will increase since no human can account for sudden erratic brakes of a self driver, even though it predicted a correct incoming accident with another V2V connected car.

This made me realize all those benchmarks which unlike loosely acts as a person being awarded with a certificate after his doctorate is rather worthless, for you never know when it bullshit-ted.