xeckr

≈November 2022.

Literally the first time I used ChatGPT, within days of release. It wasn't so much panic as amazement.

It took HN a surprisingly long time to come to terms with the fact that professional SWE as we knew it was coming to an end.

In 2023/2024 we saw a demo of "denial" being a stage of grief live on this site.

time to vibecode vista from scratch Ig

I remember retracing QR codes on graph paper to pass time in grade 12 physics, this was back in 2013/2014.

Cops bust in searching for his drugs, then accuse him of invasion of privacy and humiliation...

"NOT approved!"

"The user said the exact word 'approved'. Implementing plan."

Bad timing.

Brilliant. They're repackaging the argument governments have long made about E2EE being dangerous to children.

Too bad that would hurt their bottom line.

How do you square (heh) this with Jack Dorsey axing nearly half of Block's headcount?

Blatant strawman.

I think the part about China is just about projecting alignment with the USG in hopes that this will result in Anthropic being treated more favourably by the current administration.

Now imagine you run a mowing service with 4 employees. Suddenly an unbounded number of people appear on the job market who are ready to work for your company at a 5% the cost of your previous employees. Best of all, they become more competent and less expensive over time. You can't yet fire your entire original roster all at once since they need to teach the new hires the specifics of the job, but after that's done, what do you need them for?

Show HN: A RL suite that teaches LLMs to introduce typos at just the right frequency for the output to appear more human.

Wow a human in 2040, very optimistic take.

The strawberry/seahorse emoji meme is like a century old in AI-time.

True, though stochastic gradient descent replaces all that human guesswork with predictable scaling laws. The hyperstimuli of the tomorrow will be nothing like what we recognize today as entertainment.

>I think it reveals that comparisons to heroin use in arguments against TikTok are hyperbolic and disconnected from reality, by empirical data.

I don't think that follows from your premises. Who is overstating the evils of heroin? Plenty of people argue that viewing TikTok (or AI-optimized short-form feeds) has bad side effects, mostly in the direction of eroding your ability to pay attention to anything less stimulating.

One thing that makes heroin more benign is that it's "finished" in some sense. The drug trade will find more addictive substances (e.g. fentanyl), but a vial of pure heroin isn't going to gradually become more addictive over time in ways that are imperceptible to the user but visible on the backend because the loss function trends downward.

There's an AI behind the video feed optimized for keeping your attention for as long as possible. That is quite different from making your media more engaging.

The logical endpoint of optimizing AI for viewer retention is something that you literally cannot look away from.

>only 23%–38% become addicted

Wow, only Russian roulette with 2 bullets odds?

It freezes sometimes, but I love it.

Cool idea! I got 50%. Without looking to deeply into it, it seems to me that the two double jeopardy rulings contradict one another somewhat?

Fun fact: the Chinese central bank is not at all independent, but one-party rule prevents monetary policy from becoming incoherent as a result of short-term political interests tied to the election cycle.

An independent central bank fixes this in a democracy, but now it can exert considerable influence over political outcomes.

Consider that Carter was punished by the public in his re-election campaign for a recession which was ostensibly caused by Federal Reserve policy choices which ultimately benefited Reagan.

Indeed, the singularity has come for the University, and now the University's death rattles are heard in the form of academics writing opinion pieces desperately trying to grasp at anything that might save their privileged position, searching for an imaginary escape hatch that relieves them from the compulsion of joining the rabble in contemplating their professional obsolescence.

>As an example, ask an LLM to do some 10th grade math. Inspect the thinking process. It can regurgitate the process and the rules but cannot perform them.

It seems to me that the solution is just RL to get the language model to delegate the actual calculation to the appropriate tool.

I'll go with <3

Looks like we get these for about 60 days for periods lasting 11 years.

>I enjoy having the worlds knowledge at my fingertips. I enjoy being able to video-call my family from anywhere in the world at any time. I enjoy never being lost cause I always have a map showing where I am. I enjoy having group chats with all my different social groups, big and small.

>None of the above was possible just 20 years ago.

I also enjoy having those things, but we had all of them 15 years ago. Since then we got... algorithmic feeds?

Ha, I'm applying for medical school after seeing that AI will quickly overtake 100% of my software engineering skillset, if it hasn't already. There goes my moat...

If there is a mechanism by which the English language can lose letters over time (such as þ or æ), why wouldn't there be one by which it gains it?

It would make even more sense, after all we lose letters because we write those sounds using other letters or letter combinations, however the "ü" in "Türkiye" doesn't have an analogue in the existing alphabet.