Dota 2 But Balanced By AI
ฝัง
- เผยแพร่เมื่อ 25 ส.ค. 2024
- Visit squarespace.com... for 10% off your first purchase of a website or domain.
Play Sweet Dreams Alex on Steam! Thank you! store.steampow...
Check out Stories of Dota! / @storiesofdota
Subscribe!!😡 www.youtube.co...
I stream on Twitch every day: / baumiandcats
Join the Community Discord! / discord
Chroma: Bloom and Blight is a digital card game I created: store.steampow...
Disastles is a physical card game I created: www.disastles....
Social Media:
/ baumiandcats
/ baumiandcats
/ baumiandcats
Business Mail:
baumi4hire@gmail.com
#Baumi #dota2 #DotaBut
As a computer science graduate, I must say that your undrestanding of large language models is essentially accurate. They have no grasp at the semantics of the text they produce, only the form. It is not wise to rely on them for information.
That's not true, the whole point of language models is semantics. The main problem is that they're probabilistic models and hallucinate as a result, but that has little to do with semantics. They're actually really good at semantics. That's why ChatGPT is able to answer your questions. It just doesn't have a grasp of the truth.
Baumi's "word calculator" refers to current LLMs essentially being calculators that predict the most likely next word in the sequence based on semantics.
Gonna chip in my 2 cents too...
LLM models rely on the vectorization of the input data (turning the words and their context into a large collection of numbers, this collection is known as a vector) this process is somewhat of an art.
Beyond that, the LLM is a massive database of all these vectors, along with training. It is useful to think of the vectors in the database as coordinates.
The training is designed to build associations between these coordinates, allowing you to understand how you'd move between points in the database. Essentially training adds a current to the sea of vectors.
After that, running a query is a case of splitting out some keywords for actual query parameters, setting these parameters, and tossing your query into the ocean of vectors at the closest matching location based on similarity between your query and everything else, and seeing where the current takes it.
This process means that the model will hallucinate because it is not designed to find factually correct information, but rather to take your context and based on tons of historical data, build a similar answer to the training data.
It makes it great for story writing, great for applying new ideas to existing concepts. Terrible for ensuring that the flow it follows is valid.
An idea for a bad LoD build: the Unintelligible build. It consists of abilities that you're not quite sure about what they do exactly. Like stuff from Oracle, Chen, Visage, probably Enfeeble, etc. Probably unplayable.
Best 'leave a like' I've seen so far, the one where you roll the green screen down and go 'oh look at that' was the previous favourite
Miki's "fucking like the video" is best. And baumi the cat leave a like is chef's kiss
Just realized the heros available are the first ones by the alphabet order, which makes me think the creator was planning to add all of them and started with the first 10 and got bored after XD
ah my favorite alphabetically arranged heroes, Abaddon, Ancient Apparition, Anti-Mage, Arc Warden, and Aunderlord
(i know it's probably because of Abyssal Underlord being his old name but it's funnier this way)
I wasnt bored but it takes a lot of work to 1. give the AI the KV files of each individual ability, 2. write the prompt in a way that it produces an output that is even working and 3. refine/retry it until it gives something that is funny/cool to add to the game ^^ I have since added like 50 or 60 new heroes and plan to add almost all and change stuff like items and neutrals XD
@@DiamantenPony oh wow I am impressed you're still working on the mod, cheers to you my friend for making something fun for people to enjoy
I love that global call... you are just fighting some neutrals, then everybody hear some wierd yell and are like "what the hell, lets stop fighting and go check that" and you just forgot that you are enemies that just wanted to kill each other and just go see this angry red guy that is yelling like crazy :D.
And about the AI -- using it for geting synonyms is great... I use google translator for this, just type the word in diferent language and see all the synonyms that google throws at me :D
I like how axe stacked whole map
Chat gpt is most definitely AI. Its not AGI or ASI.
Artificial intelligence is just that, artificial learning/cognition/intelligence. Its a mimicry of some facets of human cognitive capacities.
the unfortunate thing is the general public associates AI with artificial consciousness, they think that the AI actually processes thought and information
it's so apparent when you talk to people that want to "add AI to [insert process]" and they think it will magically automate everything
Nice haircut Boomi
ChatGPT is definitely AI. Not AGI though, as someone in chat seemed to confuse AI with.
once the person said "my teacher said wikipedia is not a good source" i personally would have linked him the video of Fern/Simplicisimus, depending on if you want the english or german video, about "the biggest hoax of the Internet" or in german "die größte Lüge des Internets" and said this happens if you trust wikipedia~ xD
the side got better and better of the years, after that, sure. but i personally would double check the infos i get from there before using it as a source of anything with real importance
That last point is the best way to think about Wikipedia. It's a good way to find general information on what you are looking for, but always check the sources before accepting it as fact.
Me: future me, is ai dota great?
Future me: You're a loser.
33:35 Path of Exile is like that. over 1000 hours and still learn something every time I play. have tried a lot of content to see what it's all like and still have tons to learn.
berseker's phone call
I can definitely see why an AI would buff AA that much. You kinda need the context of the whole game and how it's played in order to reason why his damage should be significantly lower than everyone elses, and why his stun has to come with a condition that is pretty easy for an opponent to break. ChatGPT wouldn't know the significance of his ult being the only source of 100% healing reduction (which also pierces BKB), or how significant his W is with the vision and spell amp it provides.
But the other heroes I really can't explain lmao.
that aba has 1 sec respawn time seem fair
What would be interesting would be to train a model to play every heroes for millions of games and to adjust parameters until it reaches 50% win rate for every heroes
0:25 Smooth
how about "stories of magic" where you talk about the competative side of magic, what decks where played in pro tour xyz, why it was good etc.
Chat gpt is good for information as long as u can verify the information yourself. then again you have to verify most of the information available on the internet so it basically there isn't much of a difference in that case
It's very good to have fast answers to questions that you can verify by trying (coding) or in science, when you are unfamiliar for a osbcure concept, it does quite well for explaining it.
I have to read a lot of scientific litterature in my masters and sometimes some authors will use obscure algorithms with obscure concepts that are not well explained... And sometime I just want to have a general understanding of their works. In that case, I found that chat gpt is interesting. I ask it :" what is this concept and and what is it typically used for?".
It will then give me a general direction. From those direction I can more precisely do a google search and waste less time.
i try to help chatgpt with feedback . tell chatgpt which answer is true or i send true answer
I did it... I succumbed to the press the like button ad.....
Elden ring dlc when?
Calculator comparison is on point.
DOTAI 2
38 sec 0 views, baumi fell off
Fell off of what?
Reported for spam
that joke was never funny and you should feel bad for thinking it was
22 hours and only 9 likes, kujo18 fell off.
Or rather was never on it to begin with 😉
First
3rd
Fourth?