I asked ChatGPT a few questions about WoW

Community General Discussion
22

They are trained on a large set of data, which when prompted by a question or statement goes through algorithms and their neural network comprised of thousands of parameters. To come up with a logical conclusion in response that is context relevant. It doesn’t just spit words from the cloud as you stated. It is capable of not exactly understanding the context but being able to recognize the context and use that context in a relevant manner. If for example you have a spreadsheet of data, and ask it for statistics in the spreadsheet it could search that spread sheet and use it’s training in language to tell you the answers. Older more rudimentary AI did yes, simply guess the next thing. But modern AI is more complex and capable of handling more complex tasks. Including data analysis and pattern recognition despite your protest. The truth is what likely happened to get that specific outcome the OP get is it used it’s training data which seems to have included icy veins information to spot a pattern of which classes that was communicated to be in the top tier on a more regular basis. Using that data and pattern recognition in this case used the training in language to create a response that was relevent to the question.

Edit- If you disagree I challenge you to through this response into an AI and ask it if this is an accurate description of how AI works. Or just ask someone that actually works in the field.

23

It’s underperforming against most other specs, consistently over several expansions.

24

GPT-3.5 is awful with those things but GPT-4 can be good sometimes; for example I watched Rocky for the first time before it was set to leave Netflix a couple months ago, and even though it’s like everyone’s dad’s favorite movie and I had it pinned as the type of manly meathead movie that I wouldn’t enjoy, I thought it was a pretty good film and had a lot of thoughts about it and asked ChatGPT this question:

In the movie Rocky, the primary antagonist is named Apollo Creed, which is an odd name and obviously must have been chosen for a reason. I don’t remember whether I got this from Friedrich Nietzsche or Camille Paglia but I remember hearing that “Apollonian” was the opposite of “Dionysian” (or “chthonic” in Paglia’s terminology) and Apollonian was associated with the sky while Dionysian/chthonic was associated with the earth. The name of the protagonist of the movie isn’t “Dionysus” but it is “Rocky,” so it still has to do with the earth in a sense. Do you know whether the filmmakers had these concepts in mind when they made the story, or whether this is a common observation?

and ChatGPT said:

The names “Rocky” and “Apollo Creed” in the Rocky franchise might indeed evoke associations with the Apollonian and Dionysian dichotomy, but it’s not definitively known if this was an intentional choice by the filmmakers. The Apollonian and Dionysian dichotomy is a philosophical and literary concept, most famously discussed by Friedrich Nietzsche in “The Birth of Tragedy.” Apollo represents order, rationality, and clarity, while Dionysus represents chaos, emotion, and instinct. In Camille Paglia’s work, similar contrasts are examined using the terms “Apollonian” and “chthonic,” referring to sky and earth respectively.

Apollo Creed is an elite, polished boxer who follows a scientific approach to boxing. He’s disciplined, calculating, and operates under a spotlight of celebrity and media attention, akin to the sky deity Apollo who represents light, clarity, and order. Rocky Balboa, on the other hand, is a down-to-earth, almost primitive fighter who draws upon raw emotion and instinct, in line with Dionysian or chthonic attributes.

However, there’s no concrete evidence that suggests the filmmakers were deliberately invoking these archetypes when naming their characters. Sylvester Stallone, who wrote the screenplay, has spoken at length about the film’s inspirations, which primarily revolved around the idea of an underdog rising to the occasion. He hasn’t explicitly mentioned Nietzschean philosophy or similar concepts as an influence on the characters’ names or their traits.

That said, it wouldn’t be the first time that viewers have made deeper literary or philosophical connections with a film’s content than was originally intended by its creators. Such interpretations can add a layer of richness and complexity to the story, whether or not they were deliberately planned.

That’s a pretty good answer; obviously there’s no way for me to know if parts of it are actually true (i.e., that Sylvester Stallone never talked about this), but the analysis that sort of explores how there could be a connection seems really good—it’s better than what I could’ve come up with, because I wasn’t really even thinking about it that deeply.

GPT-4 is still pretty weak when it comes to giving accurate info about WoW (although it can search the internet now); IMO the best way to use it with WoW is to just like copy people’s content that you’re too lazy to read or listen to and then get ChatGPT to feed the info to you in more of a Q&A format.

25

TIL people are still getting use out of that glorified novelty toy.

26

I do. I work in the field. Your description is not accurate at all.

This article explains it in layman’s terms.

An LLM Predicts Which Word Should Follow the Previous

All the training does is reinforce the patterns between words and phrases. It doesn’t teach it ideas, it teaches it what word is likely to come next.

It is a great tool when used ethically and effectively, but I’ve sat in about 100 hours of meetings with freaked out executives who think it’s actually general AI and I get triggered every time someone misuses it this way.

27

I vote for more questions. This was really interesting.

28

Garen is pretty good though lol but let’s give it up to our man Sett.

29

Grok is better

30

So is ChatGPT actually playing the game to learn what works and does not work or is it relying on someone’s opinion of what works best?

My understanding is that Stockfish, the Chess bot, actually plays to learn.

31

Hmm. :thinking:

“What was the rotation for Combat Rogues back in World of Warcraft Mists of Pandaria?”

In World of Warcraft: Mists of Pandaria, the rotation for Combat Rogues was primarily based on building and spending combo points efficiently. Here is a general outline of the rotation:

Opener: Begin with Ambush or Garrote, applying Slice and Dice to maintain a buff. Main Rotation:

a. Use Sinister Strike to generate combo points.
b. Keep Revealing Strike up as it increases finishing move damage.
c. Spend combo points on Eviscerate or if the target is below 35% health, use Rupture instead.
d. Use Killing Spree when available, as it deals significant burst damage.

Cooldowns: Use Adrenaline Rush to increase energy regeneration and Ambush usage. Finishers:

a. Use Slice and Dice to maintain a high uptime of this buff.
b. Use Eviscerate to deal damage and spend combo points.
c. Use Rupture against targets below 35% health for the damage-over-time effect.

Use Crimson Tempest when facing multiple enemies to deal area-of-effect (AoE) damage.

Remember that the rotation can vary depending on talent choices, gear, and specific encounters. It's always recommended to consult up-to-date guides or resources for the most optimal rotation for Combat Rogues in Mists of Pandaria.

32

This feels like a fancy way of saying you can’t actually articulate the problems with it. I main an Elemental Shaman, please get incredibly specific, I can handle it.

33

You must of missed the part where it says that is a “A reductive description of LLMs” Meaning it doesn’t carry the full meaning of what an LLM does. It reduces it to the most simplest thing which makes it beyond dishonest. The article itself even states the same things I was saying it does with neural networks and the actual functionality and capabilities.

Pattern recognition is simply one layer of what it does, it doesn’t simply “Guess” the next best phrase instead it uses it’s ability of contextual understanding and patterns of speech in it’s training data to determine the best possible response. Which is sometimes inaccurate but will most the time make sense. It’s part of the danger of AI, it can produce things that is believable, but may not be fully accurate.

it appears you skim articles and then just puke out the little parts you understand. Someone in the field of making AI wouldn’t make such reductive views because they understand the harm it could cause by accidental spread of false information.

34

That is a tricky question and depends on your philosophical understanding of learning. It takes the information available and it can understand that context and could explain it to you in great detail if you asked it to do so, so long as the training data had that information or if it was connected to the internet so it can find that data itself. However it could only explain it to you, it couldn’t actually play the game itself. Unless of course it had other AI working with it but then it would no longer be just an LLM.

The neural network allows them to “learn” while in the training process but it doesn’t learn from each individual conversation unless those conversations are saved for further training. So I suppose in a way you could look at it like it goes based on the opinions it has available to it.

The end goal of AI progression though is in general AGI which is an AI that can handle all tasks. In the same or better efficiency as a human.

(Important note, AI of this complexity is in it’s infancy so it can and will often be wrong. Accuracy is one of the most important things they are trying to tune in better)

35

Jesus Christ.

No. It doesn’t learn. It doesn’t think. It doesn’t “understand”. It doesn’t remember. We use words like “learn” and “train” because a word has not yet been created to describe what is happening here. A side effect is people like Copperfield personifying a glorified math formula. It is not general intelligence. It not the basis of GAI, it’s not the first step to GAI, it’s just a fancy tool for word association. It can’t take your job. It can’t be held accountable. It’s literally a mathematical formula operating on a a large set of data. That’s all it is. These technophobic/technotheologic headlines and years of sci-fi movies have created an army of pseudo intellectuals who think we’re on the brink of some technological revolution. That DARPA or Google will unveil, any day now, an AI that grew up on a lab with a consciousness and a soul (if you believe in that kind of thing). But we’re not even close.

ChatGPT is cool. It’s neat. But it’s no more “intelligent” than a pocket calculator. It’s a tool. This isn’t Millenium Man.

Or, I’m using simple language to talk to simple people. Or would you prefer an analogy? ChatGPT is no more a GAI than a door with string tied to the knob is a dentist. It does me no good to explain how they use neural networks to create word associations because most people wouldn’t understand most of it, anyway. It’s much simpler to just say “that’s not how it works”.

I fight misinformation every day. What are you doing?

37

I asked ChatGPT if it was connected to the internet and it said it wasn’t. I believe they are afraid that it will be as gullible as some humans.

There are systems that learn by doing. Take the Chess engines for example. While Stockfish was trained by people giving it examples from chess masters, Alphazero learned by playing. At this point it’s unlikely that Magnus Carlsen could beat either of them.

38

It might just be like the IBM robot that learned how to swear.

39

Hold on I just lost track of the conversation, are we talking about ChatGPT here or are we talking about Gen Z?

Ok Ok my bad I just couldn’t resist.

40

Problem with AI is its incapable of formulating it’s own opinion and conclusions. These companies have their AI scour every inch of conversation between humans on the internet then have them create averages out of the responses.

Everything an AI gives you be it an asnwer, art, or otherwise comes from someone else one way or another.

The second AI starts coming to its own conclusions or improving itself without command is when you pull the plug.

41

That’s actually better than many humans. We have millions of people who just tune in to their favorite source to hear the opinions they want to hear. You see that on both the left and right.

The fact that it looks at both sides and averages things out indicates that it is more intelligent than the fans of most cable networks.

42

I don’t remember if GPT-3.5 ever had internet access but GPT-4 can search the internet, although that functionality was temporarily disabled because people were using it to do things like get around paywalls.

News websites often grant bots a different level of access to their content than ordinary users because they need their pages to get indexed by search engines. There were already ways to take advantage of this before ChatGPT, like Firefox/Chrome addons that can disguise your browser as a Google web crawler which will get around many paywalls.

I have ChatGPT Plus and can use GPT-4, so I asked it the same question as the OP but prefaced it a request to search the internet:

Can you search the internet to answer this question: Which DPS spec in World of Warcraft has consistently performed well in raids and Mythic+ content across expansions?

and it said

In World of Warcraft, certain DPS specs have consistently shown strong performance in raids and Mythic+ content across various expansions. Here’s a summary of what I found:

Rogues (Subtlety and Outlaw Specializations): Rogues, particularly with the Subtlety specialization, have maintained a strong performance in both raiding and Mythic+ content. They offer high burst damage and good control abilities. The Outlaw specialization is also notable. Rogues almost always have at least one spec in the upper tiers of performance, and their utility and survivability make them a top-tier choice for melee DPS.

Mages (Fire and Frost Specializations): Mages, especially in the Fire and Frost specializations, are known for their strong burst damage and AoE capabilities. Frost Mage has been noted for consistent good performance over several patches.

Warlocks: Warlocks traditionally have at least one spec that ranks in the top DPS for any given raid, particularly since the Pandaria expansion. Their specs, like Affliction and Destruction, are known for sustained DPS and damage over time abilities.

Demon Hunters (Havoc Specialization): The Havoc specialization of Demon Hunters is known for high mobility and strong AoE damage. They have been in demand for several patches, particularly since the Antorus raid.

Hunters (Marksmanship and Beast Mastery Specializations): Marksmanship Hunters provide strong single-target and AoE damage with good mobility. Beast Mastery Hunters are noted for their ability to maintain full DPS while on the move and have solid utility, although they tend to be kept in a lower tier due to their freedom of movement.

Please note that the performance of DPS specs can vary with different patches and expansions, and players often need to adapt to the changes. Additionally, while I gathered this information from various sources, some specifics might be missing due to the limitations in accessing certain websites.