AI will kill PoE and all related online games that have trade

"

tsunamikun#0433 wrote:

No its quite neccessary because for your speechpattern I can easily recognize that you don't actual know anything of AI.

Look in a mirror

Agents lmfao i'll give you a tip right now, field experts don't use marketing lingo they never have. That's used by the middlemen so they can toss off in a bucket with each other to raise funds.

"

Draegnarrr#2823 wrote:

Look in a mirror

Agents lmfao i'll give you a tip right now, field experts don't use marketing lingo they never have. That's used by the middlemen so they can toss off in a bucket with each other to raise funds.

Cute, if that is your whole answer to any points I brought up, I guess is was right. And of course i used your language, you yourself employed it, when talking with people you kinda need to use words they understand. Why should I use language you don't understand?

In all your posts you actually never talked about any technical term, technology or model(s) even after i asked twice about used models, systems, protocols etc. You didn't even try. Big talk, no substance. Either post something actually showing that you're right.

Btw in at least 3 seminars even last year i heard researchers talking about AI Agents and ofc the word agentic too. Weird how that works huh? You could actually believe speech is changing as time goes on. or maybe ML researchers are not field experts hmm....

ps: for further clarification, I'm not an expert by the way, I'm more into probabilistic learning, for me an agent is an entity with statuses and possible actions, not some weird mix of llms and tools with communication protocols between them. Hence I asked you and the OP whats what ;)

Claude Mythos.

Bye, bye account security / game integrity.

allready did

Am not an expert either. My girlfriend kind of is when it comes to AI and gigantic databases(Snowflake Cortex). Obviously not asked her about that but I think I have at least a tiny bit of understanding what AI is capable of doing.

I think what AI has over a regular bot is the ability to learn. Right now we talk about margins. Once in a while traders actively manipulating the market to skyrocket prices on certain items where they expect the demand to go up. As it's working you don't need to sell everything to generate profit assuming you can still sell the items at the previous price without any or a slightly loss later. The bots will even support your strategy and in best case buy stuff from you making it like a tiny snowballsystem that is manually triggered while bots just take over the reacting part.

What I expect to be the biggest influence of AI is that direct margins are not that important anymore because it can calculate ahead. F.e. No one would buy scarabs for 1c per because the margin of 3:1 them is very low if positive at all. But an AI would assume that the demand rises, if not for the scarabs you bought then for the expensive ones. And it also doesn't count profit per hour so that less interesting trades and flips can be considered now removing the bottom cap step by step. So in other words my thesis is: AI will manipulate the market in a more widespread range and as a last step create a gigantic snowball system with a lot of winners and loosers. Simply because I expect there will be way more tools triggering market manipulation than there are traders doing so right now.

Will it be a long term problem? That depends on the ability of the tools to learn how to limit themselves. Because once these tools start to only communicate with each others instead with actual players they will fail. Simply because PoE is not dependent on the worth of actual currencies but the relation of currency per hour and relevant gear that will be much harder to be valued by tools. It might affect other games way more though. So right now I am tending to say that from my perspective that will be a massive problem for 1-2 leagues only, because I don't expect anyone to put enough money in training the AI. But who knows what the future will bring.

well i just the following things to always remember:

1. AI does not do anything if you don't trigger it. This is important, there is no kind of analysis system that configures it own weights or something, you give it a clear goal. I you ask it to predict the market on scarabs lets say it will do it.
And maybe it will predict a rise of the price based on data from old leagues etc.
maybe it predicts a rise because some weight in its training data got more impact than it should. maybe you should invest real money and let an llm decide for you where to invest that money xD

See the point is, it will tell you bullshit, and sometimes this bullshit is right.
If the currency market were some kind of physical system were outcomes can be predicted through complex math or specific input, yeah it can predict. But currency markets are like real markets, you would need a lot more data to correctly predict rises and falls. And then your ultimate problem is how to get that information into the AI fast enough (its too late when the streamer is already playing the new strat)

2. The ability to learn, or better the illusion of it. Ok you do not rebuild your llm each time you knowledge get bigger, you just keep a history of your queries. This is the reality for consumers. And no the AI doesn't learn, If you want your history of older queries to matter in result of your current one the AI will expand your current query with all historical ones. Yeah, that gets slow pretty fast. You would need to build your custom system to actually get game knowledge even into an llm. At least if you want knowledge not found casually on reddit/forums/wiki. Especially with currency data we have only a few sites which list them, so data for it in llm should be quite weak (if someone wants to slam me, just test it please, but use historical data from 1 or 2 leagues ago or else it will just pull current values which it gets via websearch, and that is not what we want, now is it)

3. Stop talking about llms as if they are entities. There are like a hammer, inert as long is does not get used. And no, making a cronjob doesn't change it. They do not communicate with each other, they query each other. Communication is exchange of information with the information held and changed by both parties, they just command each other and a user (and a programmer designing the agent(s)) define the queries for the master llm. That's it, there is no intelligence found.

and mythos? is lit as fuck, we're currently applying for access but currently it seems more like its bad news for bad guys. All those juicy zero days will be found and closed, so sad.

Your assumption is simply wrong. You don't need too much data to predict something in PoE. The market is so small that you can dictate the price of smaller stuff like Mortal Ignorance relatively easy. If people saw they were 1:5d and fell to 1:10d there is always someone who tries to flip agressively and in the back there is someone that build up hoarding huge piles by already flipping div cards or simply taking them from the market before waiting for exact that moment turning the 4c investment into 1/10th of a div which is a gigantic undercut now while also a huge profit. And there will always be people that buy 300 fragments now instead of 20 because they think they made a good deal by seeing the price that comes up next. You saw this in every league where Uber Atziri was relevant for feared (chose an example that people cannot do anymore because I am actually annoyed by that). And usually the flipper also wins, otherwise no one would do it.
With more currency there is a larger scale of course. You can see that exact behaviour f.e. here: https://poe.ninja/poe1/economy/mirage/scarabs/scarab-of-the-sinistral
Is there a limited supply? No, that thing is common.
Do people have a monopol on them? Surely not. The price drops again within a day if not directly. The profit is made though and if you sell the rest for 1c no one can take it from you.

The only reason why that works is because people believe in logics of the market where they actually don't apply and people who know that and use it to their advantage. When I said that I believe in a gigantic snowballsystem with winners and loosers then I actually meant that everyone who will go for margins and treat the currency market like a stock exchange will just burn currency on the long run. There are laws and tit for tat strategies in the real world that are there for a reason and create an illusion of a liberal market at least to a certain point. But without that I expect pure agression to be smarter than those scholars. And for that your tool doesn't need to be educated in economics too much.

I would skip the discussion about the technical part. Firstly, because I expect you to be more into it than me (although I would have some points in my head that make me feel in a certain way. Just don't see myself in the position where I actually want to defend things that I have no clue about compared with professionals). Secondly, because I believe that, as I stated, there is not as much complexity to it as you assume (might be wrong on that as every other thesis that will stand in that thread in the end).

You don't know this, and obviously no one is allowed to show that they know this, but the game has been "solved" in terms of botting for years now. Every league start theres a lot of bots farming on day 1-2 already, which is the main driving factor pushing divine orb price (irl price) down at record speeds. This is not as bad as it sounds, since for most people the game still looks normal. Point being, only the top .1% and "handfarmed" RMT players suffer from this, for the rest of people this currency injection is dwarfed by the sheer size of the playerbase.


GGG does what it can to combat these issues but naturally its a cat and mouse race with heavy monetary incentive for the perpetrators.

"

cavexs#4102 wrote:

Your assumption is simply wrong. ...

Yeah thats all very nice and whatever but i'm not talking about a human predicting anything, i talk about an llm predicting anything. The first is, as you say, just some simple market manipulation. The second is a task not bound to strats you train a bot in, but datadriven learning. And hint, it won't learn it.

and yes, there are gamebots since forever, everyone knows this, but that bozu talks about some kind of agentic trade bot thats o so much better than enything nonAI (whatever that means)