Dark Souls 3, but ChatGPT chooses my build... 

Your surprised the ai that pulls from the internet uses the wiki?

yea i mean that is quite literally how that works

I’d be way more scared if ChatGPT saw the questions and did a whole play through.

@@KuriYellow Of course, but the fact that it successfully imitates a person who read the wiki but never played the game is impressive by itself.

Ah so most of the dark souls fandom?

Probably watched too many ymfah videos

@@Genindraz too be fair, it searches parts of information online So it more than likely had taken some - a lot of information from Ymfah

@@Flippirinono it takes from previous users

@@Fallen-Knight As someone who hasn't used it, I have no choice but to trust this

​@Fallen-Knight it definitely doesn't considering someone would have to give the ai bot advice on how to play DS3 lol

Well from what I know, it scours the internet for information so some of it probably is actually from college kids discussion posts lmao

oh god dont remind me about discussion posts bane of my fucking existence

I agree with this comment. I’ve seen discussion board posts and they are like that. You can tell by the way it looks.

Given its basis, that adds up.

Kids at my school use exclusively ChatGPT for discussion posts, so I'm not surprised.

I second this motion, he should do it as an epilogue later on after he's had the chance to play the DLC normally

Fr

@@garbothraxis4201 I agree, it should absolutely not be his first playthrough of it, but it would be nice to see.

Why is it scary?

@@dartoney imagine karma-farming reddit bots that can actually write suitable comments, racking up thousands of Karma, completely automatically, gaining access to larger subreddits which check karma count and account history, so they can be used by scammers, people who intend to spread malicious software or to spread hate-speech

@@karajohn5504 oh my, that's totally some kind of completely new thing that can never happen without using a highly-advanced AI. What kind of terrible thing is ChatGPT going to be used for next, Nigerian letters?

@NessunO no, the slightly unsettling part is that it will give scammers and people who intend to spread hate speech more credibility and access to ban evasion.

it's just a hodgepodge of of bad google advice.

the first dlc kinda sucks aside from the weapons and the final boss

Gugnir : Bro, are you high? ChatGPT : Let me check. ChatGPT : Yes I like to believe Maxor only edits his videos when he's high.

Hahaha, god I need to rewatch that video

I see you everywhere LOL

can also use the pillars too, the trick is running around them while the arrows are happening

It didn't, it just happened to give more coherent answers at the start, which lead to a pattern of coherent answers. Previously, coherency wasn't reinforced, so incoherency was.

It's definitely not the best -- but it's also obsolete, GPT4 is out and it looks like it will be 10x better in every category. I'm quite sure GPT5 will be able to do my entire job. (Sorry in advance for the rant, I just had to get this off my chest) The one thing I'd really like to see is a model that's smart enough to choose it's approach. If you ask ChatGPT-3 to play 20 questions, it'll play along but fall flat on its face. It's DEFINITELY got enough raw power behind it, but it just doesn't know how. You could fine-tune ChatGPT just by allowing it to take notes, instead of the "take everything in" brute force solution it seems to be using. If you ask ChatGPT to list the clues its been given in 20 questions, it'll give you about 4-5. If you ask for all of them, it will chug for a while and then give you 6-8. It's physically incapable of remembering all 20 questions. You can even tell it the answer you're picking before you start, and it will forget that by the end. If it could keep track of its overall task and take notes as it goes, it would hit that out of the park. But it can't. GPT-4 is probably has enough raw power and data behind it to design a whole house. Plumbing, electricity, room layout...but there's no way it could keep track of it all and put it together. But if it could -- if it could actually keep track of data in addition to its (8x larger) context space, that would be a huge improvement. I see a future, with current neural net methods, where I can tell GPT to make one room of the house a little larger and it adjusts the blueprints to accommodate it, while remembering all the other things I wanted. And I see a future where if GPT can't give me everything I asked for, it apologizes and says that problem is too hard, but lets me know that if I drop one of the constraints it can find a solution. I don't think any of this is crazy. And that's if you gave it a reasonable amount of note-taking space. There's no reason why they'll stop there. Why not let GPT-5 write down everything it knows (or thinks it knows), and then have a background process trawling through that database looking for contradictions? Not just a language model, not just a language model that can keep track of things on a notesheet, but a language model that has a long list of "facts". The language model can spout its bullshit about which bosses in DS3 you need to beat, and then the user corrects them. GPT then checks its fact database, checks the internet wiki and forums, finds an answer that looks right, and then asks you to confirm it. Then of course the next step would be to have GPT keep track of what evidence it has to back up each fact. How many people have confirmed it? What other facts does it link up with? What sources did it use to develop that "fact"? None of this required a revolutionary new method of machine learning or neural network architecture. This was all just adding long term memory as a peripherary. GPT3 is just a glorified language model, but it can (and will) become much more capable in the very near future. And I promise you, we aren't ready. Maybe it won't be until GPT-6 or GPT-7, but there will be a neural net that can act as CEO, and perform 98% of the office work, and do 98% of the engineering. And before that, we'll have a neural network that's better at deciding government policy than your average politician. How long will it take? 30 years? 70 years? I don't know. The future's going to be wild though. You don't need AGI to shake things up. You don't need AGI to beat the game of Go, and you don't need AGI to run a company or a country. The tricky part is, AlphaGo was able to crush top players in Go, KataGo could crush AlphaGo, and a decent human player just beat KataGo in 14/15 games because it had an oversight in its training. For good and bad, AI are here and we're not ready. My grandpa used one of the early computers that was just a pre-programmed spreadsheet for insurance rate calculations. I used floppy discs and cleverbot. My kids are going to go through elementary school with a phone in their pocket that can pass graduate school exams. And in 20 years, when I'm about to turn 50, AI will have taken my job. A job that everyone would have thought was safe 20 years ago. Fuck man. It's a glorified next word predictor that we use for silly youtube videos now, but the world is going to _change_

@@tomc.5704 that was amazing

@@tomc.5704 Currently GPT's main problem is in regards to token limitations, but there have been huge breakthroughs lately that will likely find their way into future GPT models, removing that extremely limiting factor. It's also worth noting, it's actually extremely inaccurate to call it a "next work predicter" given how it's neural network works. It's task is to predict the next word in a sentence, but that doesn't mean that's all it does. That's just it's job. It's like equating chefs to microwaves, there's a huge difference here, not only in the fact that a chef is far more complex and capable than a microwave, but also in the fact that a chef doesn't have to be a chef. Microwaves are made to cook food, chefs learn to cook food, this difference applies to AI as well as it does to chefs.

Huh that’s interesting I would’ve thought that it had unfettered internet access and used relevant information from the internet to answer questions

It's actually just coincidence. Unless he went from GPT3.5 to GPT4 between videos, it has no knowledge of the past conversations as it's loaded up anew. It's more likely that it happened to give more coherent answers this time around at first, and due to that pattern of coherency, it continued to give more coherent answers throughout his conversation.

OOH that would be so interesting to watch!

what would you try to let ChatGPT choose? What boss you should try or which armor/weapon you should craft/search for? I am interested into the ideas for using it to play Terraria

@@tikurai4970 probably early tips, class, accessories, and boss tips. hope it knows about life crystals underground and fallen stars for mana crystals. maybe ask what to do after each boss? that's kinda what I'd do

Well like any good lie or someone good at bullshtting there is always some truth in there somewhere.

@@demonintellect9834 It even relies on the default answer if you don't know what you're talking about: just telling someone to get good and dodge attacks more

perfect

gug ngir

Well you are using magic afterall

TLDR: Tell ChatGPT when it gets something wrong so it is less likely to get it wrong in the future

I'm doing that myself RN, getting walled hard by the endgame bosses

@@simplexsage I've been thinking of trying it myself, after watching Iron pineapple do it on elden ring.

@@markabyss1200 fair warning, you actually can't quickstep at that equip load, took me by surprise when I tried it

@@simplexsage oh really? I guess that's only an elden ring thing then?

It also has a lot longer recovery time and lowers your stamina regen