Topic: How AIs are looking at Widelands
WorldSavior Topic Opener |
Posted at: 2023-06-17, 12:24
I asked an alternative to ChatGPT called "Teach Anything" the following question: 'Which tribe in the game "Widelands" is the strongest?' Answer: 'Principle: As a language model, my understanding of the strength of tribes in the game "Widelands" is based on the statistical analysis of the game's mechanics and the different attributes of each tribe. The strength of a tribe can be determined by its ability to efficiently manage its resources, build a strong economy, and effectively defend itself against enemy attacks. Additionally, the strength of a tribe can also be influenced by its unique characteristics and special abilities. Application: Based on the analysis of the game mechanics, the strongest tribe in "Widelands" is the Empire tribe. The Empire tribe has a unique ability to generate a large amount of gold, which allows them to quickly build a strong economy and expand their territory. They also have access to powerful military units that can effectively defend against enemy attacks. Additionally, the Empire tribe has a variety of unique buildings and technologies that allow them to produce advanced units and improve their economic performance. Overall, the Empire tribe is the most well-rounded and powerful tribe in "Widelands", making it an excellent choice for players looking for a strong and versatile tribe to play with.' LOL Wanted to save the world, then I got widetracked Top Quote |
tothxa |
Posted at: 2023-06-18, 00:15
I guess this is another example that a language model is not really an AI, only a bullsh*t generator. And I almost got impressed by the principle part of the answer, until I started reading the application part… But this is even more puzzling: Q:
A:
It clearly has some data on Widelands, so how can it completely miss the economy part? Every introductory text about WL stresses it. Top Quote |
tothxa |
Posted at: 2023-06-18, 00:33
Wow, the plot thickens: Q:
A:
Q:
A:
It's getting ever more disturbing that many people seem to think these language models are on the brink of a productivity breakthrough… OK, I stop now. Top Quote |
Ron_of_Nord |
Posted at: Today 06:32
I volunteer to take on these A.I.'s on a large map with my Barbarians to teach them a thing or two about Widelands ! They wouldn't last a minute based on their responses above ... The aussies are coming, ya ho! ya ho! Top Quote |
BoeseKaiser |
Posted at: Today 16:44
Actually I think this is an example that language models are a first step to eventually reach general AI (don't make me say what I didn't, we're not there yet by a long shot !). Giving a BS answer on a subject about which we have really limited infos is basically how humans go about answering questions ^^ lack of accuracy seems to be a hard to avoid tradeoff for generality (there again, look at humans). I used to consider GPTs/LLMs as BS autocomplete, The 3Blue1Brown video series about their functioning (highly recommend those for anyone remotely interested in LLMs/GPTs, but warning, Math intensive) kind of changed that. I don't think it's worth the hype, but I am convinced it's something more than just an auto complete. I am actually impressed that it gives answer in the right ballpark for a niche theme like widelands. It just needs (Like most humans) to be allowed/learn to say "I don't know" but I'm pretty sure that's forbidden from most pre-prompts, because no one wants to read "IDK" from a helper program It's also probably not present on a lot of training data because humans also don't like to say they don't know ^^ Anyways, I actually never use them, because usually Wikipedia/a calculator/my friends give a better answer, But I do think they are a crazy feat of engineering and are not just autocomplete as we understand it. here a blog post (there is also a publication based on the post, I link the post for readability) about it's representation of chess game : https://adamkarvonen.github.io/machine_learning/2024/03/20/chess-gpt-interventions.html One way or the other, the discussion is much more nuanced than "it's an AI revolution" vs " it's a shitty autocomplete". But I might be biased, I like linear algebra.
Yeah, me too nice to post again on the forum, long time no write ^^ Top Quote |