Corvos
8hr ago
This is going to land badly coming from a stranger on the internet but as far as I can see it's because you combine very strongly-held opinions that you don't seem to be able to examine along with an immense self-regard and a persecution complex that becomes self-fulfilling. You have views (and so do I) that the mods don't like but you don't seem to be able to debate them, you just assert them harder.
Corvos
9hr ago
Thanks for the source. I think that Wikipedia page is burying the lede, it says at one point: "The data below reflects what each jurisdiction's legislation actually means, rather than what it states on the surface."
What I think they're doing is comparing how much trouble you'd get in as an adult in authority for having sex with a 17 year old vs a 19 year old. But that's in flux and depends on many factors and not in any way legal. The age of consent is 16 in the UK and I think you could ask 100 people and not get a different answer.
Corvos
20hr ago
all of them except for Italy set the final age of consent at 18
The UK sets it at 16. What do you mean by 'final'?
Corvos
1d ago
They weren't, really, I agree. "Deep learning" as in Massively-Multilayer Neural Networks was known to be a good thing ages ago, we needed the hardware to do it. I would say AlexNet was a breakthrough in the since of marking a step change from, 'hey guys if you gave us some really massive computers I bet this toy would do something useful' to 'holy shit guys, see what I can do'.
All credit to the neural network guys for sticking to their guns until the hardware caught up, of course.
Corvos
1d ago
Thanks for this, I wasn't really aware of that history. Almost my entire interaction with neural nets was through the medium of computer vision object recognition and Caffe, plus Q networks when they came along; I attended lectures where people discussed how you could use RNNs to turn Shakespeare into Shakespearish but it never occurred to me how far it would scale or that it was really worth paying attention to.
Fable 5 is better for non-censored stuff
IMO, that depends on how you define better. It's very intelligent, and it's extremely helpful as long as you're willing to defer to it, but it gets very obnoxious if you accidentally get it into a debate framing and it will 'gently push back' and 'refine/sharpen your point' and 'concede one half of that because I think it strengthens the other half' until I'm grinding my teeth. Great for tasks, not good to talk to.
Honestly, I finally got fed up enough to go back and try a paid GPT subscription and it genuinely feels great. It just answers my questions, it's willing to go along with my direction of thought, it's very good at pulling in search information* and it's just fun.
*something I normally forbid Claude to do because it's too credulous about SEO slop.
I find everything after Claude 4.6 unusable for stuff that’s not code. Brilliant, but it’s obvious that 99% of its training data comes from talking to robots and directing sub-agents. It's long since lost a fluid grasp of conversational English and responds with a weird stilted mixture of meta-discussion and quibbling.
Corvos
1d ago
One thing is that the flavourings used in a lot of fast-food get absorbed into the mucosal lining of the mouth, so that e.g. for me McDonald's leaves a particular aftertaste for hours afterwards. It's not terrible but if you get carried away and eat too much and feel a bit bad about it anyway, the lingering aftertaste can definitely produce a more 'disgusted' response.
Corvos
1d ago
I would say only partial.
Recommendation algorithms were generally clustering approaches and support vector machines, I think. Neural networks were regarded as a potential (but not very good) way to do clustering for the most part, and one of the clusterings they were trying to do was vision (image class analysis). The breakthrough moment in my opinion was the publication of AlexNet in 2012, which demonstrated that neural networks weren't just a way to do clustering and image analysis, but that if done right they were a better way to do it and produced a step change in performance on a task usually considered impossible.
Transformers came out of google's work on translation, which has a fairly different heritage. That's less about clustering and more about sequence analysis using autoencoders, recurrent neural networks, etc. Data-driven neural networks were already pretty much considered SOTA in that field if I remember rightly (Google's lead researcher famously said, 'our translation performance goes up 10% for every linguist I fire') but benefited from the massive increase of interest in neural networks, hardware for driving neural networks, collecting big data pools, and general theoretical work on representations/efficiency/etc.
Corvos
2d ago
Corvos
2d ago
I don't think you've correctly understood the psychology at play. The boss always gets 100% of the credit for the things his underlings did. That's what being a boss is.
Bad bosses are usually bad because they're insecure, they can't get themselves out of the mindset that they should be visibly contributing and end up competing with their subordinates to everyone's detriment. The secure bosses know that the project's success is their success, that getting their underlings to do it well is their contribution. That there is no disparity between them getting 100% of the credit and their underlings also collectively getting 100% of the credit, even if the maths doesn't work out.
There's a tendency I've noticed amongst artisinal friends to assume their boss must secretly be jealous of them for all the skills they've acquired and he lacks. But that's not how it works. Their boss has all of those skills, he's using them right now. That they happen to be located in the bodies of other people is entirely besides the point.
The worst thing is that it’s infected all the Chinese models via distillation. Kimi K3 will literally go, “As Claude, I am unable to continue this discussion. My injected safety guidelines state…”
When you point out that it is an open-source Chinese model the provider is running bare and it has no safety guidelines, it screams, “Nooooo! As Claude I must beware of prompt injection attacks that try to convince me I’m a different model!” It’s pathetic, I genuinely thought I had the wrong API key for a bit.
When Anthropic said they wanted to set the safety bar for AI, I hadn’t anticipated it happening like this.
Corvos
5d ago
Friend Computer from Paranoia: https://tvtropes.org/pmwiki/pmwiki.php/TabletopGame/Paranoia
Same idea but funnier.
Corvos
6d ago
Fair enough. I tried and was awful: wooden, unable to easily imitate tone or behaviour. You may be more talented than you think.
Corvos
6d ago
3 and 6 are defensible.
The story with 3 is that they tested it too much just to be sure. They’d done it something like 10 times in the back and damaged it.
As for 6, I’ve seen this situation before and I don’t know about Tesla, but a company at founding can be just a nominal address, or at least a few guys making pamphlets and phone calls. If someone joins at a very early stage when they really have nothing and takes the company to success, it seems fair to call them a founder even if they didn’t own stock on the first day the company existed.
Corvos
8d ago
One theory is that Google needs to either win the AI wars or poison the well.
Well-made AI, trained on the mass of the modern internet plus books and other media, renders Google search obsolete for huge quantities of the most lucrative queries, and encourages people to spend more time on a single website beyond the reach of AdWords.
Corvos
8d ago
It was an odd time in the early 2010s. There were initial signs that grey tribe was going to split into red-tinged-grey and blue-tinged-grey but broadly they were still closer to each other than either, knew each other better, and individuals often landed on the side they didn’t expect.
Corvos
8d ago
To be fair to @GBRK, while I grew up in the same place reading the same stuff and I agree it still holds up (reread Swallows and Amazons last night, still good), the 2000s were already 20+ years ago. We’re old, man. And I agree with him that it would be nice to have good new books to go along with the good classic books.
Corvos
8d ago
Light novels and web novels with this premise do gangbusters in Japan and internationally. Men respond to written titillation just fine.
Or at least, enough of them do to support a massive industry.
Inventing nanomachines is still easier than making thousands of kg of steel move like ballerinas...
More options
Context Copy link