Using AI to generate strategies for MCOC

Knordy

With the development of AI in recent times, I wondered if we could ask the AI to generate strategies, team selections or provide informative guides in clearing content or defeating bosses. Can an AI makes sense of all the data if you have a database with all information of the nodes, champions, buffs, debuffs etc to help players make strategic decisions. Because there currently is no tool with that knowledge yet, I'm experimenting with a tool that uses search results as input. I'm experimenting with https://weblangchain.vercel.app/ (search engine Google, Model GPT-3.5-turbo) and asked it some question. Since haven't cleared that content yet, I'm curious if the answers of the AI make any sense or is generic non-sense. Or have you found some other tool that can handle this? ChatGPT for instance only has knowledge until 2022, so that can't help with recent released content.

Best champions against Nameless Grandmaster:
Answer
When facing the Nameless Grandmaster in the Necropolis of Marvel Contest of Champions, there are several champions that can be effective in the fight. Here are some of the best champions to consider:

• Ghost: Ghost is a popular choice due to her ability to phase and evade attacks, making her difficult to hit. Her high damage output and ability to bypass the Grandmaster's abilities can be advantageous in the fight.
• Doctor Doom: Doctor Doom's ability to nullify buffs and control power makes him a strong contender against the Nameless Grandmaster. His Special 3 attack can also deal significant damage.
• Human Torch: Human Torch's Nova Flame ability allows him to bypass the Grandmaster's abilities and deal damage over time. Additionally, his immunity to Incinerate and his ability to purify debuffs make him a solid choice.
• Quake: Quake's ability to evade attacks and deal massive damage through her aftershock ability can be effective against the Nameless Grandmaster. Her playstyle requires precise timing and skill, but she can be a powerful option.
• Nick Fury: Nick Fury's high damage output and ability to gain Fury buffs make him a strong choice against the Grandmaster. His ability to bypass the Grandmaster's abilities with his Special 2 attack can also be advantageous.

It's important to note that the best champion for the fight may vary depending on your playstyle and the specific nodes and buffs present in the fight. Experimenting with different champions and finding the one that suits your skills and strategy is key to success

Best champions against Nameless Collector
Answer
When facing the Nameless Collector in the Abyss of Legends in Marvel Contest of Champions, there are several champions that can be effective in the fight. Here are some of the best champions to consider:

• Aegon: Aegon is widely regarded as one of the best champions for the Abyss of Legends, including the fight against the Nameless Collector. With his ability to ramp up persistent charges and deal massive damage, Aegon can quickly take down the Collector 0.
• Blade: Blade is a popular choice for countering the Nameless Collector due to his ability to nullify abilities and deal high damage. His Danger Sense ability allows him to bypass the Collector's abilities and deal increased damage.
• Ghost: Ghost is another strong counter for the Nameless Collector. Her ability to phase and evade attacks makes her difficult to hit, and her high damage output can quickly take down the Collector.
• Corvus Glaive: Corvus Glaive is a powerful champion with the ability to deal massive damage. His True Strike ability allows him to bypass the Collector's abilities and deal direct damage.
• Omega Red: Omega Red's ability to inflict Bleed and Death Spores makes him an effective counter against the Nameless Collector. His Life Steal ability also helps him sustain during the fight.
• Hyperion: Hyperion's high damage output and ability to nullify abilities make him a viable counter for the Nameless Collector. His Power Gain ability allows him to launch special attacks frequently.
• Medusa: Medusa's ability to shatter armor and deal high damage with her special attacks makes her a strong counter for the Nameless Collector. Her petrify ability can also reduce the Collector's ability accuracy.

Remember, the best counter may vary depending on your roster and playstyle. It's important to practice and understand the mechanics of each champion to maximize their effectiveness against the Nameless Collector in Abyss of Legends.

How to defeat Superior Kang?
Answer
To defeat Superior Kang in Marvel Contest of Champions, it is important to understand his abilities and develop a strategy accordingly. Here are some tips to help you beat Superior Kang:

• Exploit Weaknesses: Superior Kang has weaknesses to Bleed, Poison, Incinerate, Shock, Coldsnap, Rupture, Degeneration, and other damaging debuffs. Champions who can apply these debuffs consistently can deal significant damage to him. Examples of champions who excel in this regard include Blade, Ghost Rider, and Void.
• Power Control: Superior Kang gains power quickly and can unleash devastating Special Attacks. Champions who can control his power gain, such as Magik or Doctor Strange, can prevent him from using his Special Attacks and reduce his overall damage output.
• Evade and Intercept: Superior Kang has certain attack patterns that can be evaded and countered with an intercept. Pay attention to his movements and timing to evade his attacks and counter with your own. Champions with good evasive abilities, like Spider-Man (Stark Enhanced) or Nightcrawler, can excel in this aspect.
• Defensive Abilities: Superior Kang has high damage potential, so champions with defensive abilities can help mitigate his attacks. Champions like Hyperion, Medusa, or Captain America (Infinity War) have abilities that can reduce incoming damage or provide additional survivability.
• Buff Nullification: Superior Kang gains buffs throughout the fight, which can make him more difficult to defeat. Champions who can nullify buffs, such as Vision (Aarkus) or Scarlet Witch, can remove his buffs and weaken his overall effectiveness.

Remember to study Superior Kang's abilities and adapt your strategy accordingly. It may also be helpful to practice the fight in order to learn his attack patterns and find the best openings to strike.


Counters for Cerastes
Answer
Cerastes is a champion in the game Marvel Contest of Champions (MCOC). Based on the search results, here are some champions that are considered effective counters against Cerastes:

• Blade: Blade is a popular choice for countering Cerastes due to his ability to nullify regeneration and deal increased damage against villains. His Danger Sense ability allows him to bypass Cerastes' abilities and deal significant damage.
• Ghost: Ghost is another strong counter against Cerastes. Her ability to phase and evade attacks makes her difficult to hit, and her high damage output can quickly take down Cerastes.
• Doctor Strange: Doctor Strange's ability to nullify buffs and regenerate health makes him a good counter against Cerastes. His power control abilities can also prevent Cerastes from using special attacks.
• Hyperion: Hyperion's high damage output and power gain abilities make him a strong counter against Cerastes. His Cosmic Supremacy ability allows him to deal increased damage against villains.
• Medusa: Medusa's ability to inflict armor shatter and reduce ability accuracy makes her an effective counter against Cerastes. Her petrify ability can also prevent Cerastes from regenerating health.
• Archangel: Archangel's ability to inflict poison and reduce ability accuracy makes him a good counter against Cerastes. His neurotoxin ability can prevent Cerastes from regenerating health.
• Void: Void's ability to inflict debuffs and reduce ability accuracy makes him a strong counter against Cerastes. His fear of the void ability can also prevent Cerastes from regenerating health.

Please note that the effectiveness of these champions may vary depending on their level, rank, and the specific abilities of Cerastes. It is always recommended to study Cerastes' abilities and adjust your strategy accordingly.


Best team for quest 6.1.6
Answer
To determine the best team to use in quest 6.1.6 of MCOC, it is important to consider the boss and defenders you will encounter. One recommended champion is Magneto (Mags) at Rank 4 (R4) because he not only counters the boss but also many defenders on the path. Magneto is particularly effective in this quest because most of the defenders and bosses are metal. Another champion to consider is Hyperion (Hype), especially if he is duped, as he is meant to counter Act 6. If Hyperion is not duped, then Cosmic Ghost Rider (CGR) may be a good choice.
In addition to selecting the right champions, it is also helpful to follow the easiest paths in quest 6.1.6. For Act 5.4, you can refer to this link:.... For Act 6.1, you can refer to this link: .... These guides provide information on the map, paths, and nodes, which can assist you in planning your team and strategy for a successful quest.
Overall, the best team for quest 6.1.6 of MCOC may include Magneto (Mags), Hyperion (Hype), or Cosmic Ghost Rider (CGR) based on their abilities and the specific challenges you will face. It is also important to consider the easiest paths provided in the guides to optimize your chances of success.

AlphaNeymar10

• Hercules: Brute force your way through every piece of content.

EdisonLaw

Not sure if these are correct , these tips look outdated

Pikolu

Doesn't Nameless collector counter miss? Feels like all the suggestions are only from 2019 and sooner too.

Emilia90

EdisonLaw said:

Not sure if these are correct , these tips look outdated

You can do a lot of cool stuff with AI, but yeah, they aren’t up to date. GPT itself says it doesn’t have fully up to date information on Mcoc and the cutoff is like 2020

EdisonLaw

@Pikolu

danielmath

Sounds like pandamanpete wrote it

Greekhit

There is an ancient Greek quote that suits great at this post:
"Η ημιμάθεια χείρων εστί και αυτής ταύτης της αμάθειας"
Translation:
"Half knowledge is worse than ignorance"
🤔

Crcrcrc

You should try copy pasting the nameless grandmaster’s kit into the AI so it can actually strategize based on the info given and see what it says

Knordy

EdisonLaw said:

Not sure if these are correct , these tips look outdated

That's unfortunate, I was hoping because it could retrieve relevant up-to-date search results (guides etc) it could generate correct strategies.

Crcrcrc said:

You should try copy pasting the nameless grandmaster’s kit into the AI so it can actually strategize based on the info given and see what it says

In this tool that's not quite how it can be properly used, I think. Based on the query it retrieves search results, and uses that to try to answer the question. Providing the entire kit, could (in this tool) perhaps pollute the query. However, ideally that's what you want. Simply provide a query about the grandmaster, the AI receives the GM's abilities (and perhaps quest nodes etc), analyzes it and provides (for example) the counters for these information. If you could connect it to "auntm.ai" in this way, it would really give meaning to the ".ai".

Chuck_Finley

Geez, I guess I need to R5, max sig, and ascend my 6 star Blade. Thanks AI!

Xguard77

The Bing Chat version of the GPT models are kept up to date (fed the current internet, instead of being locked in 2022). You can toggle from GPT 3.5 to 4, which I did. Here's a response to "in MCOC, what is a good rotation for Onslaught"...

According to my video search results, the best rotation to use with Onslaught is:

Start the fight with a Heavy Attack to gain a Neuroshock charge.
Use basic attacks to build up Neuroshocks on the opponent. Neuroshocks have a higher chance to trigger on crits, so use your Precision buff wisely.
Use your Special 1 to refresh your Precision buff and inflict a Neuroshock that lasts longer than normal.
Use your Heavy Attack again to pause all Neuroshocks on the opponent and gain a Crush charge. Crush increases your attack rating and reduces the opponent’s armor rating.
Use your Special 2 to convert all Neuroshocks into Degenerations that deal more damage and last shorter. Your Special 2 also inflicts Reverse Controls, which makes it hard for the opponent to evade or block.
Repeat the cycle until the opponent is defeated.
You can watch some gameplay videos of Onslaught using this rotation here: Top 10 Best options for Onslaught boss, Onslaught | First Look, or FULL BEGINNER GUIDE MCOC. I hope this helps you master Onslaught. Have fun! 😊

Knordy

Xguard77 said:

The Bing Chat version of the GPT models are kept up to date (fed the current internet, instead of being locked in 2022)....

According to Bing Chat these are the best champions to use against Nameless Grandmaster, and is referring to these forum posts:
https://forums.playcontestofchampions.com/en/discussion/251297/best-team-for-grandmaster-6-4-6
https://forums.playcontestofchampions.com/en/discussion/188330/best-champions-against-grandmaster

Answer
The Nameless Grandmaster in the Necropolis is a formidable opponent. According to the Marvel Contest of Champions community, some of the best champions to use against the Nameless Grandmaster are:

Doctor Doom: He can help take out Annihilus and Angela and is a tank against the Grandmaster 1.
Mr. Sinister: He is immune to inverse controls, can block unblockable attacks, and regenerates 70% from crits. He has DOT for sp1 and sp3, which can take off 7% in wounded state 2.
Stealthy: He can evade the Grandmaster’s specials and has a high critical rate 1.
Please note that these are just some of the recommended champions and there may be other viable options as well. Good luck!

Darkraw346

If you can't beat them, join them

@Knordy rn:

Wicket329

This is a prime example of how current model AI isn’t really artificial intelligence at all, it’s a pattern recognition software. It “knows” that certain words are used together in certain contexts, i.e., Hyperion and nullify (because nullify is a frequent tactic used *against* Hyperion), but it doesn’t understand that context and so it says that Hyperion has a nullify ability in the above write up.

This is all fun and games here, but I’m begging everybody who reads this not to rely on AI in a professional or academic setting. Or really anywhere with consequences. AI is not smart, it just knows when certain words get said together a lot.

Knordy

Wicket329 said:

This is a prime example of how current model AI isn’t really artificial intelligence at all, it’s a pattern recognition software. It “knows” that certain words are used together in certain contexts, i.e., Hyperion and nullify (because nullify is a frequent tactic used *against* Hyperion), but it doesn’t understand that context and so it says that Hyperion has a nullify ability in the above write up.

This is all fun and games here, but I’m begging everybody who reads this not to rely on AI in a professional or academic setting. Or really anywhere with consequences. AI is not smart, it just knows when certain words get said together a lot.

That's somewhat true indeed. The key factor here is, can you provide the 'AI' with the correct context so it can generate an answer based on that context? In the above results, if you can provide the AI with several (proven and correct) guides on how to defeat the Nameless Grandmaster, the AI can then interpret the context using language models, and generate an answer to the provided question, that's currently what the AI can do. The hard part is to retrieve the correct context for the AI to interpret. If the context clearly states the that nullify is a counter measure to be used against Hyperion the current state of AI models should be able to interpret that correctly.

The nice thing about using AI as we try to do in the above examples is that we do not have to read several guides on how to clear certain content, but that the AI is processing these guides and generates a 'guide' of it's own, tailored to the provided question. The next step would be that the AI understands nodes and effective champions against those nodes, understands abilities and effective champions against those abilities etc and tries to 'reason' how you can clear that content. But again you have to provide node descriptions and effective champions etc as a context for the AI to interpret.

Currently you also cannot provide unlimited data to the AI to use as context, there is a limit (which is currently getting bigger, GPT4 has a token limit of 8k, while the newest model gpt4-turbo has a limit of 128k). Therefor perhaps the AI is not receiving a complete guide on how to clear certain content, because it does not fit the context window, and that will affect the generated answer.

If the context is not correct, the answer will also not be correct.

Marvelfan30

The ai stuck in 2018

DNA3000

Wicket329 said:

This is a prime example of how current model AI isn’t really artificial intelligence at all, it’s a pattern recognition software. It “knows” that certain words are used together in certain contexts, i.e., Hyperion and nullify (because nullify is a frequent tactic used *against* Hyperion), but it doesn’t understand that context and so it says that Hyperion has a nullify ability in the above write up.

This is all fun and games here, but I’m begging everybody who reads this not to rely on AI in a professional or academic setting. Or really anywhere with consequences. AI is not smart, it just knows when certain words get said together a lot.

I agree with the admonition at the bottom, but I think the way LLMs work is much more complex than you're stating here. It is unclear to me how context extraction in the attention mechanisms translates to what we would consider comprehension (nor do I think this is a settled issue).

The issues with GPT3.5 are as much a product of the less sophisticated model and outdated training. For example:



It is extremely difficult to explain this in terms of context-free association.

In fact, I would argue that self-attention LLMs are not pattern recognition engines in the same way most people think about pattern recognition. Yes, they are design to recognize patterns, but the patterns they are recognizing are not exactly text patterns. They are patterns in a tokenized language space that contains the representations of the structure of the data they are exposed to, not just sets of words in specific orders.

If you Google search "In Marvel Contest of Champions does Hyperion have a nullify ability?" you do not get a direct answer, because this is not a common question that anyone asks:



So how does GPT-4 know that Hyperion does not have a nullify ability? It cannot be that it simply recognizes the question and is finding the answer.

GPT-3.5 did have problems basically "knowing" wrong things that were common misconceptions. For example, when I tested an early version of Chat GPT3.5, I asked it "what is the equation for the perimeter of an ellipse?" When prompted, it gave me three different wrong answers that are common incorrect answers people give to this question. GPT-4 presumably also has issues with this, but it is much better. For example, it gives the correct answer to this question:



Part of the problem with presuming that LLMs are just simple pattern recognition engines is the fact that a lot of the data they were trained on gives very certain precise and wrong answers to certain questions. The correct answer is often the least common answer in the training data. How these models correct for this kind of thing is one of those complexities of how they work, or sometimes don't work.

I wouldn't call generative AI general artificial intelligence at the moment. But they aren't just doing the simple things a lot of their critics claim they are doing. Or rather, if they are, so am I.

Emilia90

Wicket329 said:

This is a prime example of how current model AI isn’t really artificial intelligence at all, it’s a pattern recognition software. It “knows” that certain words are used together in certain contexts, i.e., Hyperion and nullify (because nullify is a frequent tactic used *against* Hyperion), but it doesn’t understand that context and so it says that Hyperion has a nullify ability in the above write up.

This is all fun and games here, but I’m begging everybody who reads this not to rely on AI in a professional or academic setting. Or really anywhere with consequences. AI is not smart, it just knows when certain words get said together a lot.

I’ve fed some questions to GPT and it’s spot on a lot of the time. It’s been able to explain things to me in simple terms and it’s useful for a lot of general knowledge. Relying on it entirely is stupid, but it has capabilities. It does make mistakes too as DNA pointed out but it is getting better

EdisonLaw

Well the problem with using AI is that they aren’t always correct

Zeke_the_Xbot

AI is 1s and 0s all depends on the data it can aggregate and draw from. If it lacks data it will give a best guess which for an ever evolving and changing meta/game state does not make for wise council.

Also not a strong believer in AI as a tool yes I’m one of those dinosaur types that likes the old ways.

SummonerNR

DNA3000 said:

GPT-3.5 did have problems basically "knowing" wrong things that were common misconceptions. For example, when I tested an early version of Chat GPT3.5, I asked it "what is the equation for the perimeter of an ellipse?" When prompted, it gave me three different wrong answers that are common incorrect answers people give to this question. GPT-4 presumably also has issues with this, but it is much better.

And that's basically why all the various posts up here about “reworked champs” or “here's what I think the abilities of xxx champ should be” long descriptions, will manage to find their way into AI answers in the future.
Even though they would totally be wrong as to how the champs actually/currently work in MCOC.
Especially when the titles of some of those posts would lead you to believe that *IS* how the champs work. (and is rather confusing for new people just starting to read these forums)

jdschw

DNA3000 said:

I wouldn't call generative AI general artificial intelligence at the moment. But they aren't just doing the simple things a lot of their critics claim they are doing. Or rather, if they are, so am I.

OMG GUYS, DNA3000 IS A CHATBOT

jdschw

Emilia90 said:

I’ve fed some questions to GPT and it’s spot on a lot of the time. It’s been able to explain things to me in simple terms and it’s useful for a lot of general knowledge. Relying on it entirely is stupid, but it has capabilities. It does make mistakes too as DNA pointed out but it is getting better

As somebody who works in a related industry, I can say that the scary part about the current generation of LLM AI systems is that they can very confidently make completely incorrect/counterfactual statements. There's even a wonderful industry term for this phenomenon: we call it "hallucination".

Generally speaking, you should absolutely make use of LLM systems as very powerful, sophisticated tools. You should also cross-reference all new facts you learn from an LLM system with other references and sources of truth.

DNA3000

jdschw said:

Emilia90 said:

I’ve fed some questions to GPT and it’s spot on a lot of the time. It’s been able to explain things to me in simple terms and it’s useful for a lot of general knowledge. Relying on it entirely is stupid, but it has capabilities. It does make mistakes too as DNA pointed out but it is getting better

As somebody who works in a related industry, I can say that the scary part about the current generation of LLM AI systems is that they can very confidently make completely incorrect/counterfactual statements. There's even a wonderful industry term for this phenomenon: we call it "hallucination".

I'll do you one better. In my opinion, LLMs are not sometimes hallucinating, they are *always* hallucinating.

The fascinating thing about LLMs is that they are the exact opposite of what people accuse them of being. They are generally accused of just spitting back out what they were trained it. But actually, they almost never do that, except under certain very specific circumstances. Rather, they are *always* creating text they were not explicitly trained on. Its all creation, and in a sense all hallucination (there's even a statistical random component to their output generation). The LLM doesn't "know" anything in the normal semantic sense, so everything it says is something it doesn't actually "know." It has no experience except the data it was trained on.

When GPT-4 says that Hyperion doesn't have a nullify ability, is it just parroting what someone else said? Probably not. But has GPT ever actually played the game? Not yet. It has no experience with Hyperion. So when it says Hyperion has no nullify, it is, in my opinion hallucinating that fact. It just happens to be correct. The amazing thing about LLMs is not that they sometimes hallucinate, rather it is that in my opinion they are always hallucinating and almost always hallucinating a world that happens to be consistent with the real world.

I should point out that as I understand it transformer networks were originally designed to handle the problem of language translation, thus transformer network. The fact that they just happen to apparently *know* things is not entirely expected or intuitive. The idea was to learn how to extract the contextual importance of words and phrases in one language's grammar, so that they could be translated to other languages with proper structure and grammar. But if you think about it, translation and explanation are very related processes. And text generation is really just the problem of contextual prediction. Given a set of words, what is most likely to be the next few words that properly fit the context and structure. Translators turn into story-telling encyclopedias.