montag

Ha ha. If anything, HN skews quite high.

Disabled YouTube comments is turning out to be a strong signal that I'm not going to like what I'm about to watch.

> "how much do we actually need to fetch and render HTML?" and the answer is in the ballpark of single digit FTEs

Sorry, but we're not going back to Lynx...

Why don't they determine aid with a formula, instead of using these thresholds that set up weird incentives? (Same question applies to tax brackets)

I don’t think Spotify was ever a Qt application. Back in the day, it used a custom cross platform UI framework they called Spider. But it was XML based, similar to QML.

The bookmarking feature is crucial. And Bluetooth? It’s an instant buy for me. Great project!

It would be amazing if Yubico offered a physical button option.

I really hope it’s not the case that capacitive sensors are somehow preferred from a security standpoint (i.e., harder to trigger with a robot).

“Breathing on” works as you intended (like breathing on a cold window to make it foggy); the alternative “blowing on” does not.

Haha. Well besides that, it’s not really a font generator but an image generator

It's also easily misused. Take the regular expression validator for passwords on the California DMV website, for example. The website states "Must include at least 4 alpha characters". But the validation pattern

    ^(?=(.*[a-zA-Z]){4,})(?=.*[0-9!#$%]).+$

requires that these characters appear consecutively.

Thanks Peter

This sounds just right.

Achieves 98% on MNIST handwritten digit recognition.

I think this effort is positive, but a bit misdirected. Think data breach liability. Facebook and YouTube are willing and capable defenders of sensitive customer data. Watch the AshleyMadison documentary. Arrogant disregard for customer privacy and almost no culpability. These smaller, irresponsible players are where consumers are most vulnerable.

Mr. Beast is FAR from the most pathological content on YouTube.

He is a Savant.

Why does OP urgently need to change the name?

Thanks! Didn't know about this cool feature.

Louis Vuitton Medallion Blanket $1510.00 - Purchased

Really can’t make this stuff up

Haha. Cutting rate limiting to speed things up

Hmm. It would make a lot more sense to me if it were called semantic saturation, not satiation.

The more benign version happened to me. Over $1k for an Illustrator subscription that I forgot to cancel and didn't touch for 2 years. No email from them that might have reminded me. When I went to support, they offered me nothing. Say what you will about personal responsibility, but I won't pay for an Adobe product again.

edit: Also want to mention that I used a couple different subscription trackers like Trim, and they didn't identify the Adobe charges. Weird.

> In China you cannot not be explicitly a CCP data-collection front

Unintelligible.

Rewrite as “in China it’s very hard to avoid turning over data to the CCP.”

This is true but I think it would still be nice for VCS to have a first class concept of "peek through" changes (whitespace, formatting, etc.) for the purpose of blame.

Imagine how many millions of “shut up!” “F—- you alexa!” Etc.

Just quoting again from the guide:

3. DIRECT LLM PROMPTING In this method, contestants use a traditional LLM (like GPT-4) and rely on prompting techniques to solve ARC-AGI tasks. This was found to perform poorly, scoring <5%. Fine-tuning a state-of-the-art (SOTA) LLM with millions of synthetic ARC-AGI examples scores ~10%.

"LLMs like Gemini or ChatGPT [don't work] because they're basically frozen at inference time. They're not actually learning anything." - François Chollet

Additionally, keep in mind that submissions to Kaggle will not have access to the internet. Using a 3rd-party, cloud-hosted LLM is not possible.

What do you mean it can 'generate python programs that solve the tasks'? I can't find any mention of that. I only see hand-coded solutions.

I don't think that's breaking any rules, and in fact it would help to expose a whole class of weaknesses in the test.

> submissions to Kaggle will not have access to the internet. Using a 3rd-party, cloud-hosted LLM is not possible.

https://arcprize.org/guide

Curtis Roads, in the latest edition of Computer Music Tutorial:

> A new industry has emerged around AI services for creating generic popular music [...] This is the latest incarnation of a trend that started in the 1920s called Muzak, to provide licensed background music in elevators, business and dental offices, hotels, shopping malls, supermarkets, and restaurants.