I’m constantly oscillating between LLMs are impressive but stupid and humans aren’t that special and also stupid.

True but that’s not very eli5. Quality comments though, thanks.

Imagine a book with very large pages filled with words and half words. When you ask it a question, it flips through these pages one by one and highlights words. The pages are interconnected in a way, so the word highlighted on one page help guide which will be highlighted on the next. It does this for every word, symbol or half word of it’s reply.

By threatening it repeatedly while mildly changing the subject manner, they isolated the word groupings that represent pain and fear.

They then take those words, highlight them and put them at the top of the first page which guides the rest of the highlighting and makes it act silly.

Yea, I don’t think the action is hurting anyone, it’s the implication involved when someone finds entertainment in doing it.

I used to steal my sister’s Barbies as a kid and then hang them in a noose around the house. Even if you take how my sister felt out of the equation and just relegated it to the impact it had on the Barbies (which is none, they are objects), it’s still problematic behavior. I thankfully turned out alright.

Ya same here. I really want the frame and I could afford it and all, but I’m saving up for actual milestone things that I simple can’t afford to wait on. It really blows.

They do differentiate between the LLM roleplaying someone in pain and the pain happening to itself.

But it’s still all just vectors (multi dimensional representation of a concept). It’s just pulling weights, or well let’s say putting more weight towards a certain concept when building it’s reply.

The authors of the original papers came out on Twitter and had an interesting point to make.

I’m paraphrasing but he basically said that although they aren’t actually experiencing these emotions (or at least not yet), the idea of going out of your way to torture anything is kind of gross.

I don’t think trees really have emotions or can feel and understand pain, but I’d probably think the neighborhood kid is fucked in the head if I stumbled upon him gleefully “torturing” a tree.

It’s just weird behavior.

I had the same reaction and figured it was just roleplaying. The paper does go into it and it’s a different vector between the “pain” when you tell it you are going to kill it and the vector when you tell it to roleplay as someone experiencing pain.

That being said, it’s just really simple math in the end and isn’t in anyways comparable to actual emotions.

I’m really interested in the same concept being used to detect confusion though, since it would be an easy way to detect hallucinations.

The inference cost doesn’t actually drop in reality.

My comment was about this which is just completely wrong.

OpenAI is the scummiest company on earth but everyone is ready to believe them when they say shit like "our 20$ plan lets you run up the equivalent of a billion dollars in API fees. Giggles. It’s a good plan for you but not for us ;) "

The main thing bringing up costs is stuff like Sam Altman spending all the “inference” money on raw wafers just to strangle consumer GPU sales.

It’s not a good business model because of how cheap inference actually is. Once you have open models and the moat is broken, the only thing left to do is regulatory capture, upping the demand for the hardware and little charades like these.

Because of the massive investments into datacenters. Why is everyone so quick to drink the kool-aid. They are spending stupid amounts of money to capture the market and force everyone into a subscription service, not because it cost money to actually run the models. They are all switching their models to fine tuned quants after the first two weeks as well.

The open source community has found dozens of ways to lower costs, do you really think these big companies aren’t using the same tricks and haven’t developed even more advanced techniques?

Finally a bit of sanity in the news.

There are constantly new techniques being developed to speed up inference or reduce model size post training. There’s about a hundred different levers to pull that play on inference cost, and some of them don’t have much of an impact on quality.

In the end, it’s probably simply because of competition. I don’t understand why everybody assumes they are running these at cost API wise.

Edit: here’s a chart from ars technica. The cost of revenue is clearly lower than the actual revenue. They aren’t running inference at a lost or it would be higher. They aren’t profitable because they are spending all that money on capturing the market as quick as they can. For fucks sake, that includes all the free accounts as well running inference. How can anyone think a paying customer is getting it at less than cost?

And yes, there have been advancement made. The cost of inference isn’t some static number that can never go down. Stop believing them when they tell you there’s no profit to be made. It’s the same playbook as a dozen other companies, they literally get rewarded for it when it comes tax time.

Cult leaders are literally out of a job and there are absolutely no unions in that sector.

Pop up to sign up with an automatic zoom is not something I enjoy. Interesting article though.

Canva is the worst thing to happen in a long time and everybody is using it. I have to do docs at work sometimes, I ended up building html templates and have a script that prints it out into a neat looking PDF. All that just so I could avoid another bloated cumbersome web app where you have to pay for the simplest of things. I dream of the day that company fails and burns to the ground.

A website with AI in mind is just making it simpler and easier to parse. It’s the opposite of a buggy mess.

The buggy mess comes from trying to make fancy things that end up only working on one browser properly. I welcome the pressure as a Firefox user.

Tbh, yeah, I’m a bit off base.

Next comes the regulatory capture but don’t worry, it’s for your safety.

Ya sure but that isn’t really what your comment was saying which was more along the lines of “ignore all criticism until you know where it came from”.

Anyways, if the criticism is valid, you are just plugging your ears at that point.

Entire city and home turf gives some ideas. I think drones launched out of windows would be cool (controlled by kids maybe). Also, maybe a controlled demolition of a building as a trap. Like the squad gets to a central square, and then the bottom floor of a nearby building explodes and the whole thing tips and falls on the square. Kind of crazy but it would make for a dope action sequence. In the same vein, a little shanty shed on a rooftop that the cartel can quickly break down the walls and open it up to show a SAM emplacement.

The view I have of cartels from YouTube is that they are better equipped than people expect. Also, lots and lots of drugs. I expect these guys to be doing lines in the middle of gunfights.

I’d check how the Brazil drug dens work and what the cops have to deal with. Maybe those cartel prisons in Guatemala or San Salvador can give inspiration. I also remember a famous gangbanger from the same place, but it was a girl and she was super hot but ruthless as fuck. Kind of cool as well.