pocksuppet

The product is not the attention.

We've transitioned from a product economy to an attention economy. Attention = money.

The entire process of AI training is weight editing. You're simply asking "Why didn't Qwen include more training cases to prevent looping?"

One common trick in Shorts is for the topic of the video to not be revealed until the end, while leaving enough clues that you think the topic will be worthwhile. "So anyways, I was reverse engineering this project with Claude and I came across this little detail that I think some of you might be interested to know. This IoT product that I've got in this box right here is a smart doorbell that quacks when you step on it. But actually by holding this button for five seconds you can make it meow instead."

If you swipe away when it shows it's about an IoT doorbell you watched half, and if you swipe away after seeing the button makes it meow you watched the whole thing. In either case the algorithm thinks you liked it more than average.

In long-form videos this doesn't really work but you can use vague titles to achieve higher click rates. Taken from my actual homepage: "DON'T GIVE ME HOPE, DON'T LET ME HOPE THIS MUCH" "Did you know about these...?" "Calcium Carbide Trick" "The WORST skill to train on OSRS" "They Were Busy With Their Big Plans" "If you were trying to destroy the economy..." "What happens when we don't need each other any more?"

You know which one I'm not going to click on? "FRACTRAN: 14 fractions that produce infinite primes". That's because I already know FRACTRAN is an esoteric programming language written in fractions.

Lawful use isn't a clear line.

They want them to steal everything, they just don't want us to.

Oftentimes the government drops the case. When people get convicted for things they did on Tor, and don't settle, the government often drops the case so they don't have to reveal how they got information about activities on Tor.

When Germany, which doesn't play by the same rules, got DiDW and then Boystown, that was quite revealing on how authorities can break Tor.

Airlines can refuse to carry people whose name starts with a vowel, if they want to. That's a principle of free business.

Is Elon evil?

Trust... and encryption. That second one's quite important.

Ironically this page has censored me, it shows "Please wait while your request is being verified" indefinitely.

Not true, never was true. You can't write real code for either Linux or Windows without venturing into what C calls undefined behaviour, but Linux and Windows don't.

It did. Programming used to mean manual register allocation, and we don't do that any more.

Don't let the perfect be the enemy of the good

no, I'm sure many people are. Great powerful interests want that site shut down and there were many attempts to doxx and swat the owner. It's basically another Anna's Archive.

no info provided and nothing interesting leaked

A social contract towards a profit-maximizing corporation is not real. They are amoral entities with only one goal. The real implicit contract between you and a business is simple: they will do everything they can to make you give them money, and you will do everything you can to not give them money, or else you're a sucker. "Everything they can do" includes convincing you there's a social contract involved.

The centralized platform in open-web shopping is Google, which is how you discover shops.

technical solution to a social problem

Or make users scan their passports to log in.

git hosts get the same effect - ddosers request a diff of every file between every combination of first and last commits.

Everyone is for their own automation and against their competitor's automation. It's just how business works.

Average log size is about 300 bytes. 1GB of logs per day is approximately 40 requests per second average.

Make your site performance fast enough that bots don't bother you.

Cloudflare doesn't even stop bots very well. Many bots are using real browsers, which pass all fingerprint checks, and scripts that can click the checkbox. What Cloudflare has is mindshare, and being a checklist item and an accountability sink.

It's called GEO

You always have an embedding matrix. The embedding matrix specifies which token vector is 'AL', which vector is 'AK', and so on, and which one is 'Cheyenne'. You could do it trivially and say that 'AL' is 1,0,0,0,0,..., 'AK' is 0,1,0,0,0,..., and 'Cheyenne' is ...,0,0,0,0,1, but that's still an embedding matrix - one which doesn't scale.

I didn't really follow the article to its conclusion. Surely the logical conclusion of "higher powers are better" is to take the limit as power goes to infinity - which is equivalent to just finding the most similar key and then returning the corresponding value. Why would you take a combination of all powers, when you know that higher powers work better?

Nothing in the article explained why we need "a memory" to work in this particular way that leads to it becoming a kind of transformer.

Talking about context tokens storing memories makes no sense. By definition, they are not memories, because they are all from the current prompt!

The whole blog post feels like a just-so story designed to make you think transformers were obvious.

Or - a wacky idea I know - it could just show the status all the time. Maybe in another window pane, which isn't a terminal emulator. Or there could be a button you click on to show/hide the status.