HN user

persnickety

301 karma
Posts2
Comments213
View on HN

An LLM scraper is operating in a hostile environment [...] because you can't particularly tell a JavaScript proof of work system from JavaScript that does other things. [..] for people who would like to exploit your scraper's CPU to do some cryptocurrency mining, or [...] want to waste as much of your CPU as possible).

That's a valid reason to serve JS-based PoW systems scares LLM operators: there's a chance the code might actually be malicious.

That's not a valid reason to serve JS-based PoW systems to human users: the entire reason those proofs work against LLMs is the threat that the code is malicious.

In other words, PoW works against LLM scrapers not because of PoW, but because they could contain malicious code. Why would you threaten your users with that?

And if you can apply the threat only to LLMs, then why don't you cut the PoW garbage start with that instead?

I know, it's because it's not so easy. So instead of wielding the Damocles sword of malware, why not standardize on some PoW algorithm that people can honestly apply without the risks?

" This movie, built with data collected during the European Space Agency's Huygens probe on Jan. 14, 2005, shows the operation of the Descent Imager/Spectral Radiometer camera during its descent and after touchdown. The camera was funded by NASA.

The almost four-hour-long operation of the camera is shown in less than five minutes. That's 40 times the actual speed up to landing and 100 times the actual speed thereafter.

The first part of the movie shows how Titan looked to the camera as it acquired more and more images during the probe's descent. Each image has a small field of view, and dozens of images were made into mosaics of the whole scene. "

I like how clear this visualization is despite being packed with data. Once you start paying attention to the different parameters, you'll find yourself restarting the 5 minutes video to watch some other property.

There were calculators in the '90s where the display would go faint when you covered the tiny solar panel. Perhaps the battery was already drained. Quite common I would say.

RPN is definitely easier to implement. I helped someone do that as a student project and while it was minimally complex, there were no edge cases with the operators.

You pay for that by having a stack rather than a small fixed number of variables.

What's with the advice about interrupts and undebounced signals?

What does it mean that a flip-flop gets confused? What kind of undesired operation could that cause?

Because, quite honestly, if connecting directly means occasional transient failures, then having less hardware is a tempting tradeoff on small PCBs.

Forced to upgrade 2 years ago

The environmental impact grows a lot more with the number of sold product than the maintenance burden.

For a business that sold a thousand units with a handful remaining, the calculation is going to be a lot less dominated by impact than for a giant who sold millions and has thousands still in use.

And if there's a giant in this industry, that's Apple.

Who do you think spam classification false positives are going to be pubishing if not real users? At least with a captcha, you have some idea that you were rejected before you put in the effort to write your comment.

That's an uncharitable read. Connections can be made, not only received.

Networking is different kind of work than sitting at a desk, but it's still work. The benefits of that work are seen next time you want a job. Every freelancer operates this way, for example.

You're one out of 8 billion people, the only known humans inhabiting an incredible vast cosmos with billions of galaxies and billions of years of timespan. That you are sitting here reading these typed words is nothing short of incredible,

There's nothing incredible about being the only kind of human known to humans. Classic anthropic principle, making every kind of human special as long as they haven't met any other kind.

Assuming that the amount of value you create corresponds to the amount of money you are rewarded with - within the same market, then it's just about revenue.

In your examples, you open up new markets, and doing that changes the other side of the equation: you need more work to get that extra income later.

In the end, that example changes all three: the amount of work, the amount of reward, and the time distribution, so I don't think it really shows that inertia is so common.

Admittedly, the paragraph is not very upfront about it, but to me:

My pulling back started in ~2023

this sounds like significantly reducing the amount of work.

Okay, this could also be read as a complaint that the revenue is not enough for profit with any scaling down.

Does inertia matter? Or is it just another way of saying that the revenue is low?

I'd rather have X in my pocket instantly after creating the value rather than have half instantly and get the rest as an inertial trickle over the next year.

I'd have kept my keyboard shut if the implication presented was "work and little revenue is bad business", but here I'm seeing "no work and no profit is bad business".

My pulling back started in ~2023 after 8 years of thinking, 24/7, "what's the next thing I'm gonna write about?". I was tired and the biz wasn't sparking joy like it once used to. Burnout is grind with no reward.

The business stopped being profitable almost immediately. That's the strongest sign I have that this is a bad business. It's more like owning a job than running a business.

Did I get that right? Having to work to earn money makes a business bad?

And I thought extracting money without creating value was the bad thing to do.

There are other big selling points of Rust, compared to GPU-specific languages.

One is an expressive type system with algebraic types and generics. It's a lot easier to produce versions of code optimized for a certain use case. Only CUDA comes close.

This brings us to the ecosystem. Outside of CUDA, how many libraries exist that can easily be used from a shader context?

Rustc has great error messages. I don't know others, but errors in OpenGL shaders make me want to flip tables.

And being able to run the same code on the CPU is valuable as well, leading to simpler testing and debugging.

So unless you like CUDA and its problems, then this is rather attractive.

The ad company wants to make ads browsing easier, therefore it created HTTP/3. I can entertain that statement.

The ad company created QUIC to make HTTP/3 possible. That's also certainly true.

What follows: the ad company created QUIC because they wanted to make ads easier.

But "QUIC exists to improve ad deliverability" is true only in the shallowest of senses, similar to "nuclear power plants exist to drop bombs on civilians" just because research on nuclear power was driven in large part by war needs. In reality, nuclear power has its own uses beyond the military.

Similarly, QUIC taken on its own merits does not have anything to do with ads. It's just another general purpose protocol.

BTW, multiple streams will not make it any faster to load ads from third parties. Head-of-line blocking only affects resources within a single TCP connection, which can only ever be to one server. That means QUIC's streams do nothing to make loading Google's third party ads easier.