HN user
Much of the internet from the time the training data was initially collected wasn't much different. Consider how much blogspam and listicle SEO/ad optimized content would've been in the training data that is primarily interested in meandering for as long as possible to milk ad views.
I've used it to help me clean up and reverse malware samples to give the person running it problems. It's very good at taking the obfuscation apart and translating it into something easier to read. I can often get it to completely vibecode string decryption functions for me so I can get what exfils they're using and the like.
Doing the same by hand can be really time consuming and difficult depending on how many obfuscation measures are layered on top.
I have a $20 Claude sub, and the closest I've come to hitting a limit was when I was using it to fight back against some scammers spreading a stealer on discord. I was having Claude decode their webhook strings, and the first time I blew them up with pings to get a job, then just started deleting the hooks from under them when they published new versions.
But for standard productivity, I've never come remotely close to hitting a limit.
I wonder if it's because it got labeled "Mild Violence" whereas the violence in LoZ is considered "Fantasy Violence."
Maybe Mild Violence is more commonly applied to more real world violence and therefore scores higher in rating considerations or something. Because that's the only way I can make sense of that labelling.
In my experience, Kimi just tends to think a lot, with the main thing that takes up a lot of space is it constantly second-guessing itself. I've watched it do paragraph after paragraph of "Wait, actually..." while it stumbled and used a ton of tokens on one small detail of what it was asked to do. Though I also gave GLM 5.2 a task to port some JS code to Python to test it, and in my experience it doesn't second guess as bad as Kimi does, but it really did there. It kept doing web searches and second guessing tons of tiny little things, using $0.25 of API spend in total to port about ~50 lines of JavaScript. It did produce an error the first run, but on second run it gave me a program that ran.
I gave Claude Code/Fable the same task and it took significantly less time, but also stumbled on the same error as GLM. I didn't have it fix it though. I was mostly interested in timing differences.
I do like open models where I can, but I'm really hoping they get trained to second guess less. Or maybe I just need to prompt them differently. I'm not sure.
Looking at what movies got away with in the 80s while retaining a rating of "PG" definitely shows we're sliding on what we call acceptable.
There are numerous lists online that can be found with a search, but here's one by an IMDB user: https://www.imdb.com/list/ls020284330/
You could get away with some stuff in the 80s and keep PG.
I suspect it's just random. I tried feeding one of its pawns my Queen and it completely ignored it, choosing to move a knight nearby instead that didn't take the Queen.
The University of Georgia did a study and came to the conclusion that the Southern accent is fading. Less formally, a word game website did a survey that found the standard New York accent is fadingas well.
The Georgia study largely attributed it to migration and people moving more than they used to, with a focus on growing metro areas in the south. Though apparently some younger generations are trying to shed their accents to sound more professional, and the internet is creating a sort of average American accent.
Though nobody is worried they'll disappear entirely at the moment that I can find.
The only way to verify the presence of rabies is to bring in the animal that did the bite. But per the article, they released the responsible bat. Normally what you'd do in this case is the hospital gives you the rabies vaccine anyway just in case, because the alternative is if the animal was rabid, you die.
Bats are the most common carriers of rabies, and also you should watch out for any wild animal that's too comfortable getting close to a human. Especially more shy ones like raccoons and such.
I was in the same boat as a kid. My handwriting is so bad, for the essay portion of the state test we had to take one year, they got an exemption and let me use Notepad to type mine out because they didn't want to risk my grade if the person couldn't read my writing. This was in the mid-2000s.
These days I just disclaim to people when I hand them anything handwritten that I'm very aware my writing is terrible, and I will not be offended at all if they have to ask what it says.
8 years here. Mint->Arch->NixOS. Basically zero Windows use in that time outside of niche needs.
If a big player in a space is asking for regulation, always treat it as them pulling the ladder up to make things harder for new upstarts.
As an example, look at what Anthropic's response to the US making them pull Fable. They commended the action and said they believe there need to be permanent regulations around safety of released models with approval committees and mandatory testing.
They aren't recommending it purely because of safety, they want to add expenses to their competitors without so much money to burn.
Evolution is just a race to "good enough to consistently reproduce" and everything after the sufficient reproduction is irrelevant. Like the goats whose horns have to be cut or they'll eventually pierce their own brain.
Generally it's more advantageous for your own anatomy not to kill you without intervention, but they reproduce and that checks off the "good enough" box.
That's called a motion to dismiss and it does happen. Maybe it wouldn't in this particular case, but not every case gets the full 9 yards of examination if one side's argument is bad enough.
They don't swear off it, but because they know what it is and is not good for, they are way more cautious in their application of AI.
Like the time I got given a swelling tablet at work to dispose of and had to go through phone tag to get an answer on what to do with it or how dangerous it was. And my coworker asked "if [I] tried asking AI?" I said I am not relying on ChatGPT for something that might explode, I'll wait for the person who's paid to tell me about this thing that might explode.
If she is starting work on a slide-show presentation, the prompt is “Help me visualize.” She shoos away these interruptions, but they persist: “Help me edit.” “Beautify this slide.”
To be fair, making slideshows sucks and I've never met anyone that actually enjoys the experience. I'm sure some people out there enjoy it, but anything that gets me out of PowerPoint faster is a win in my book.
There was already evidence last year[1] that pointed to ChatGPT-specific words like "meticulous," "delve," etc becoming more frequently used than they were previously. The linked study used audio of academic talks and podcasts to determine this.
Those engines serve a different purpose than a library like Raylib. They give you a bunch of stuff out of the box like lighting, raytracing (esp Unreal), pathfinding, and a ton of helper functions used in making a game like vector calculations.
Raylib helps you draw stuff, play sound, and do the basics. But you're gonna be writing your own lighting/raytrace/pathfind/etc functions and it's ultimately going to take longer. The upside is if you need to do something very unique, all of the power to mske it reality is in your hands because Raylib isn't opinionated on how your game logic works and how it's packaged up. It's just the delivery guy to give the result to the user.
Statistically, we live in the safest society we ever have. We see a lot of bad stuff happening because news reporting travels further and faster than ever before, amplifying the perception the world is going to shit.
Plus, now, basically every kid is running around with a phone that gives them access to talk to the police or their parents at any time. So it's going to be a lot riskier for someone to try anything against them. Even then, between 80-90% of sexual assaults are performed by people the victims already know, and around 30% of those are relatives of the victim.
They could start getting some of that goodwill back by not paying their CEO a multi-million dollar salary and opening donations to actually help fund Firefox.
I'm surprised at how few comments there are about just how creepy this is. Going to a university is not implied consent for random people to throw up searchable websites with your name and face, let alone allowing random, anonymous other people to attach anything they want to it in a comment section.
I get it he was copying The Social Network, but just because it's been done before doesn't make it better now.
AI music is generally not going to be copyrightable unless they can show genuine human creativity was involved. So if a song is 100% AI, you could just go around performing it or straight up selling copies yourself and there's nothing* they could do about it. Though I do wonder if a human writes the lyrics, but AI generates all the music parts, if it becomes sufficiently human for copyright. Because the lyrics at that point would be actual creativity.
* I am not a lawyer, and this won't stop them from possibly trying to sue you or even winning depending on the situation. Or trying to prove there is human ingenuity involved. Do at your own peril.
Kinda happens everywhere. "I'll send it to you as an MP4" versus "I'll send it to you as an h264+aac"
I highly recommend using the "Not interested" button on anything you don't want to see. It's actually pretty effective at pruning unwanted things from your recommendations. If I get anything political or slop related, it gets the not interested button.
I also have a second channel for language learning where I used it to prune out any videos in English. It's not perfect and recommends a few still, but they get more rare as time passes.
This was something Louis Rossman suggested at one point. Small claims courts don't typically allow lawyers and require a direct representative. And small claims courts are fairly cheap financially and judicially to file in.
I wager such an attack would be very costly since they'd likely be ordered to pay the court cost of around $100 per case if they left it to default. But if they didn't, they now need to take an employee from somewhere to represent them instead of doing their actual job, which is also costly. So getting even 100 people to do this simultaneously could cost upwards of $10,000 to the target company.
A tale older than the use of GLP-1. People do X to lose weight, they hit a target weight, declare victory and continue the habits that got them in trouble in the first place. You can go a little bit heavier on the meals and loosen the exercise if you desire, but you still have to keep yourself within maintenance threshold or the weight comes back.
GLP-1 masks the problem and people don't realize their actions aren't ideal once the mask is removed.
Combined with cryptographic signatures for humans
What happens when the human gives an agent access to said signature? Then you fall back on traditional anti-bot techniques and you're right back where you started.
Because they pay API costs to send the search to SerpApi. I forget exactly what the cost was for them per-search and I'm having little luck finding it, but I know they've published that cost before and I know it's more than a whole cent. By comparison, running a good but not top-tier model to answer the same question might run a small fraction of a cent. Cheaper than a follow up query by the user.
Similar monoculture of global thought is happening in all fields.
Thereby removing yet more interesting things to see in the world through the spread of hyper-optimized inoffensive blandness. In the same way that restaurants are slowly turning into the same set of grey boxes with little of note distinguishing each.
Yup, see how long it lasts when companies in California can't install anything on their servers because they get Rejected for Legal Reasons responses to their package requests.
Because the "store" never confirmed that Cloudflare is 18.