HN user
"Solved" (heavy quotes) because instiutionalizing dozens of millions of people with no improvement is a massive crime in and of itself.
That's like those stories of LLMs saying "I fixed the vulnerability in your app" by deleting the project entirely
"The model refuses to follow my specific word detail prompts" and "The model refuses to perform hacking attempts" are on the same side of the model refusing to do something baked into it though.
Some would say "Gang of four" and "Design patterns" and "OOP" are all marketing BS
It seems like this is a description: https://interconnected.org/home/2024/09/05/cursor-party
They can, but what would make them think their initial misunderstanding is even wrong? While some might think "surely the system isn't like that..." many would default to "of course they want to steal my money" etc.
Not to mention the amount of misinformation that can get passed around
I think because the gradient is simply too confusing for laypeople to understand.
Even for a simple system like US social security that has a gradient. For every $2 you make over the limit, you lose $1 in benefits. I've heard countless times misconceptions of people thinking they'd be losing money (as in literally having less money net) by working.
I think the methodology is interesting, but you can probably design a similar, and probably better loop for "diving" into a topic and going through examples than is presented.
I have to imagine they just asked Claude to dive into a topic and generate example problems along the way.
As for the content... a lot feels like knowledge overload and concepts are introduced without explanation or "why". It basically says "here is a training loop" and never answers my immediate question of "...for what?" It also introduces random concepts like setting the seed that don't look like they're even in the final training loop it provides.
As a comparison, Pytorch itself has docs that go over the entire training loop as well. And it explains why you're loading the data its loading. With examples that run instead of being scattered on.
Slop aimed towards children has practically _always_ existed. The "100,000 whys" naming reminds me of an old "700000 games" CD
AI slop is just a more complete reimplementation of the "shovelware" from the 90s
That license IS any copy left license, but nobody will trust a random fork is my understanding.
I can agree though that the network effect of "everyone else uses X" is pretty unfortunate and has led to a lot of other issues too, so yes very disappointing
Representative democracy was supposed to be the solution for this. i.e. "I dont know what the best way to do age verification should be, but it's probably important"... and then voting for the candidate that has shown ability, intelligence, and a good nature to look at the issue in your best interest and handle all the edge cases without screwing you over.
What we've lost is the integrity of this system entirely. No candidate can truly be trusted to implement an age verification system without first getting a check from big tech
What was complicated about the setup? Its a plug-ins folder... add it in and boom you are ready to go
Im not sure OpenAI or Anthropic are necessarily better? Or even opposed to this
You are correct, but especially in current day that analogy is quite bad.
I expect Median LoC might be very high with the average developer using AI these days... but the dev who is making atomic changes that are fixing the AI output is probably tiny LoC but way more important
That does not sound at all reassuring, that the only safeguard is the system blocking access and that the API has no safeguard.
Its also easily possible to have sensitive files misplaced, especially for a general non-technical user that would be the one falling for a browser hijacking attack
I would also imagine the majority of people used to RLHF on AI could have English not as a first language. I have no expertise in this area, but I feel like the tuning at that phase would also affect the "voice", "tone", and "style" that gets put out.
Because it's not blatantly obvious what is a B versus a C until after the fact.
So you
1. Reduce morale of Bs with potential. Layoffs or just reducing opportunity affects the whole team 2. Enforce stereotype threat on Bs you've miscategorized, reducing their performance and turning them into Cs.
It probably affects the As too, but I know I'm not there yet, so I can't speak on it.
This just sounds like your Linkedin went stale for some reason or other. Do you have the Open to Work status? Or even just be active on there, apply to jobs, like posts or stuff like that?
I get plenty of messages from "recruiters" through LinkedIn. They're not particularly great, but I also only have a few years of experience
The controls are really frustrating on mobile. Clicking a new tile sometimes adds it to my district. Other times it makes a new district?
Me together built... doesn't make sense either
Or, you know... money
Just down vote and move on. You're helping them generate better slop by providing input like this
I'm surprised that (all) these models haven't been export controlled already. Relatively benign software like VMware is export controlled or even hobbyist radio projects have gotten hairy with ITAR.
But a model that can provide general information, research, or source code for most modern technology?
It is really unusual that this is the first notice of this
There's good AI writing
Sure, but the LLM-isms in AI writing are mentally exhausting to see in every way at this point.
The whole point of reading, frankly, is to understand the voice of other people. When you pass that through a distorted filter that makes everyone sound the same... its bad, lossy, frustrating communication
It's also dishonest. When you publish something that is direct output without your wording. Digital catfishing at best.
The only good AI writing is providing the prompt, because the question is way more interesting, and way more constructive to learning than the answer
There is little chance your writing is as consistently the exact same as the default LLM output, which this definitely is.
The only case I would consider this being a misleading part of the statistic would be (1), and that type of usage is probably too rare to be significant.
Multitasking "while watching TV" is probably still harmful. Just focus on the task. For me personally, having the screen there has led to increased anxiousness and loss of my sense of time.
Watching TV with a friend is probably less harmful, but its still "just" sitting in front of that screen with little activity going on.
It does kind of depend what you're watching/listening to, but humans need extended time away from that constant stream of media regardless
Dp you have any source on that last part?
Not trying to citation needed you, but that claim is quite significant.
And it also increases the competition! Most of the giant tech companies aren't really making money by having X amount of software. They make their money by having software that is better than their competition, so they can get Y amount of users. But if your product's development is now simplified because AI can reproduce it, you still have to work harder to keep that competitive advantage
As someone who has poorer vision than average, pair programming in person as the observer is almost impossible. Id have to ask the driver constantly for clarification on what they typed or have them increase their text size to something that would be uncomfortable to them.
That's just to say its not always more effective
Texture is independent of taste
I mean, I didn't read it as a joke. Taking a rest can lead to a clearer ability to think... thereby being more productive, not less.