what if instead of saying "to Co-operate" the other agent will be a critic? what would their interaction be like?
@AllAboutAI
Жыл бұрын
Great idea! Will do that for sure:) thnx for tuning in!
@cauemarcelltomaz470
Жыл бұрын
Or u can order to him to act like a boss.
@hayatogok
Жыл бұрын
I agree. While I was watching the video that was the first thing that came to my mind. What if one is a professional photographer and the other is in critic mode? Would that lead to better results?
@magejoshplays
Жыл бұрын
Take that idea a level deeper. System as is with 2 agents collaborating to write, then those twos output is treated as one agent that speaks to it's partner agent whose job is to be the editor that critiques and provides feedback for reflection, which feeds back to the first two agents. Loop until the editor scores the first pairs output above a certain definable point, like maybe 3 loops before it's done or something like that until we can run an aesthetic scoring system against the output.
@JOZZYmusic
Жыл бұрын
This is fantastically cool. The only improvement I can see is with prompting. You told them to agree so they are basically just yes-anding one another. If you add something in the prompt that asks them to critique and improve on the others input x times before agreeing, you might get some interesting results, or be able to solve some more difficult problems. Inspiring work. Excited to see more.
@zippytechnologies
Жыл бұрын
anyone see if he release how he connected these or any of the sort in the community on github? Looking to extend to autoGPT and multiple models not just gpt3.5/4 but maybe also vicuna 13b and gpt4 on the other - see how they critique each other then...
@whizadree
Жыл бұрын
You always need a 3rd agent that is harder to convince that is unbiased and can say no
@peterkonrad4364
Жыл бұрын
you are right about collaboration of two humans. the synergy effect comes from their two different live experiences. what you have done is let two identical humans talk to each other, so basically again only one human talking and agreeing with itself. the prompt would have to describe two radically different personalities.
@SyntaxDomain
Жыл бұрын
Looking forward to an update video when openai allows you to send an image for gpt4 to analyze and you have the agents reflect on their images and suggest their own optimizations.
@AllAboutAI
Жыл бұрын
Yes!! That is a cool idea i will take a note of:) tnx for tuning in!
@demischi3018
Жыл бұрын
I love that they are so polite, saying Goodbye 15 Times 😂
@bautibunge737
Жыл бұрын
It's great! I think it would be interesting to give each agent its own personality, where one can be more creative while the other one could be more perfectionist and detail-driven for example. Just suscribed!
@valdasdr8432
Жыл бұрын
This basically feels like everyone racing to get the biggest advantage before the revolution begins manifesting :D I love it.
@sovereignberserk1346
Жыл бұрын
I really appreciate witnessing two agents falling in love with each other during work. So relatable 🤣 Gives a whole new perspective on the Matrix movies.
@JohnDlugosz
Жыл бұрын
You should include the instruction, "when you are done, just respond `over and out.`" and have your Python script stop relaying when it sees that.
@AllAboutAI
Жыл бұрын
Great idea:) will implement!
@shadeall
Жыл бұрын
great content thank you for this. Also how do i signup for the membership for the tutorials, thank you.
@fabio4674
Жыл бұрын
I did a similar thing, making other bots too, before you released the agent contents. The only point is that I never integrated it with anything out of ChatGPT. Your idea is good!
@magejoshplays
Жыл бұрын
Oh that architecture is a great solve for my story engine's SD prompt section needing a method for setting specific lore checking.
@cosmiclettuce
Жыл бұрын
Hi could you elaborate on why the use of multiple agents? Isn't this just the LLM talking to itself? Might be interesting to do the same thing with a single agent and compare the results. Cheers from southeastern arizona usa and thanks for the great videos!
@BlooFlame
Жыл бұрын
What do the colors of the arrows signify in the diagram? I got sorta confused looking it over.
@AllAboutAI
Жыл бұрын
haha, tbh no color code :P
@BlooFlame
Жыл бұрын
@@AllAboutAI lol fair enough.
@MosesHightower
Жыл бұрын
Is there a link where it can be downloaded, or is this within AutoGPT? Can it be run on a Mac?
@zippytechnologies
Жыл бұрын
omg man - show me how to do this on my local machine - and let me know if we can change it up to use vicuna 13b instead of gpt4 and local db to store everything
@zippytechnologies
Жыл бұрын
I want to have one agent using chatGPT 3.5 and one using 4 or vicuna13b to really mix it up and maximize critical thinking over larger sets of training from different models...
@m.ehtizan
Жыл бұрын
Hey Man, I just watched your video and I must say that it was really informative and well-made. I loved your videos. I was wondering if I could help you edit your videos and also make highly engaging shorts for you?
@davediamond
Жыл бұрын
Extremely entertaining. And that's all that matters!
@MultiMam12345
Жыл бұрын
A great portrait is 90% psychology. They should make photographs of each-other for practice.
@matsberggrund8403
Жыл бұрын
I think the idea is interesting indeed. Did you try to give the two agents slightly different rules/instructions how to behave? Like one prof photographer and one sculpture artist ... :) Nice work anyhow!!!
@bek6506
Жыл бұрын
is there a github repo for autosd?
@JohnDlugosz
Жыл бұрын
A couple of questions: First, you're just saying "generate the text prompt" but not giving it details on how to use SD and examples of prompts that give the kinds of results you like. How does it know what "generate the text prompt" is supposed to mean? Second, why does it need two agents? They don't hash things out at all -- they agree on everything. With both given the same prompt, they will think alike. I think you're just providing a way to *iterate* so it can work step-by-step and review its output. More interesting to see if one of the ideas had a goof in it and the other agent caught it on review. Your script could inject such errors to test that out. Finally, it says the agent can use Google. How? Is it a special keyword output, or is that part of the API that it knows about and you just need to enable?
@milothelight
Жыл бұрын
How do members access this repo?
@JohnDlugosz
Жыл бұрын
1:17 Wow, the Chatbot doesn't even care about your bad spelling! I'm always careful to correct spell-check indicated words, so ChatGPT is not confused or doesn't understand. I think the situation is a lot better with GPT-4, as it has an awareness of words beyond just being tokens. In one of my first uses of ChatGPT, it responded saying it doesn't know what that is. I said "it's a misspelling of ___" (whatever it was) and was amazed that this conversational response was understood in context, and it repaired and ran the original question.
@AllAboutAI
Жыл бұрын
Good points John:) thnx for tuning in!
@sherpya
Жыл бұрын
do the agents role play game really produces improvments?
@FabioAngela
Жыл бұрын
Have you tried to make these agents less accommodating? They always agree to what the other propose, I think would be interesting to add some plot twist
@zippytechnologies
Жыл бұрын
or use different randomness setting in the 'temp" setting
@ArunCx
Жыл бұрын
A captivating hyperrealistic photo portrait of a stunning, early 20s, mixed-race fema le with mesmerizing eyes that exude both my stery and allure. Her unique and enchanting features are skillfully captured in a high resolution image, showcasing her flawless complexion and radiant smile. The photograph is taken using a Sony Alpha a7R III mirrorless camera with a Sony 85mm f/1.4 lens, e mphasizing the depth and intensity of her g aze. The settings used are an aperture of f /2.2, ISO 200, and a shutter speed of 1/250 sec, allowing for soft, natural lighting that enhances her beauty. The composition is framed to highlight her captivating eyes, which draw the viewer in and leave a lasting impression.
@NeferLopez
Жыл бұрын
1:10 I don't see the python code in the Github repo for members. Where can I find it?
@ImmoBlaese
Жыл бұрын
I'd love to see them arguing as well : )
@AllAboutAI
Жыл бұрын
Will def set that up soon :) tnx for tuning in :)
@GothicDragonX
Жыл бұрын
Prompt them an argument, one of them will defend the argument, the other plays devil's advocate. They don't have to be rivals just curious if they will agree to compromise and agree on a result that is beneficial for both.
@anhtule7758
Жыл бұрын
How to get access to this?
@achiche1337
Жыл бұрын
How do I run this myself?
@goyashy
Жыл бұрын
Good stuff!
@duhai1836
Жыл бұрын
GPT Agents are cool BUT is this better than what you can come up with a simple prompt in GPT-4 ? ( Asking for the image quality alone )
@AllAboutAI
Жыл бұрын
No dont think so, but is Autonomus, so it can generate 1000s of images! Thnx for tuning in :)
@duhai1836
Жыл бұрын
@@AllAboutAI Yes, and you get a kick out of watching them communicate ;D ! THANK YOU for keeping us up to date on all the advances in AI, very hard to keep up.
@Ricolaaaaaaaaaaaaaaaaa
Жыл бұрын
In theory it will provide images with better accuracy when giving the initial prompt/prompts. When looped back to GPT-4 for instance it can run through multiple models before reaching the boss agent receiving final approval prior to output.
@user-on6uf6om7s
Жыл бұрын
I'd be interested to see a comparison between the prompts generated by this and a human with experience generating SD prompts. I imagine the human would do better as these prompts seem to contain somewhat verbose descriptions that use a lot of filler words that don't seem like they're likely to be as efficiently tokenized as what you'll see with a typical prompt which is a bunch of 1 to 2 word tags that succinctly capture a desired quality. Combine that with the fact that the human can use negative prompts which this doesn't seem to and the ability to tweak the results and I don't think AI is going to be replacing human prompt engineers anytime soon which is a relief for the AI-based economy.
@Ricolaaaaaaaaaaaaaaaaa
Жыл бұрын
@@user-on6uf6om7s It's not about the prompts. It's about image value verification within the generation loop. Say you write - "generate an image of a woman with her legs crossed sitting on a couch reading a book" the regular output might be a woman reading a book on the couch but the legs aren't crossed. The agents verify that the image relevancy doesn't meet the required criteria so it loops back until it does then it sends it to the boss agent for final verification that the image is accurate to the prompt before output.
@Sunnygoesbrr
Жыл бұрын
15 times saying goodbye is exactly me
@VaibhavShewale
Жыл бұрын
well many can make or try to make just by looking at the flowchart!
@skydou6680
Жыл бұрын
Where in Norway are u from my friend? Love ur videos😍 also do u have a degree in computer engineering or something similar?
@arnonil
Жыл бұрын
Thanks!
@AllAboutAI
Жыл бұрын
Thank you!!:))
@thefoxialbunny
Жыл бұрын
Guys, I'm lost. What are "agents" in the GPT context?
@kil0x6969
Жыл бұрын
can you teach how to use two agents please?
@RBlake-tu6xc
Жыл бұрын
Is there a way to make the agents less agreeable? I wonder if that would create a better end product?
@futurizerush
Жыл бұрын
We will have a Company running by AGI soon!
@lcfnyc2007
Жыл бұрын
Why are you not using Midjourney v. 5 -the most advanced graphics AI model to date?
@barulicksama3838
Жыл бұрын
Why all the time they prompt mixing races? Just WHO is running this AI? Really got you thinking.
@AllAboutAI
Жыл бұрын
Good pick up :) will check the log
@sorayadevries5314
Жыл бұрын
Funny to see the results, even when you wrote professional wrong
@horrorcoder
Жыл бұрын
funny how they greets each other a dozen of time!
@andreiburcuta4254
Жыл бұрын
Where do i find autosd to use?? You have no guides on how to use it??????????????????
@TheJustinside
Жыл бұрын
This is neat and all but I don't understand the point? Why not just ask 1 agent to generate the prompt and image? it seems superfluous to have multiple agents agree on something.
@LOFIBOT-bp6hm
Жыл бұрын
you are an expert in autogpt, why don't you ask autogpt to rewrite itself and make a better UI so you don't have to do the tasks from the console
@adamfilip
Жыл бұрын
each agent should have a different personality profile so that they dont always agree
@Raffa987
Жыл бұрын
Looks fun but the one agent should be more like a critique to the other agent. Like watching over what the other agent does or suggest. Because now it seems like he is always agreeing to everything so he has no real purpose i guess?
@AllAboutAI
Жыл бұрын
Yes that works great :) the conversations is much longer , so did not showcase that in this video! Thnx for tuning in!
@marklondon9004
Жыл бұрын
Make two more agents who are running an art class of agents, and are tweaking the instructions of their Agent pupils.
@brianj7204
Жыл бұрын
I'm just waiting for a ChatGPT model that can do or create anything i ask it to, at a fast pace, is easily accesible and free of charge, and is not politically charged or restricted.
@AllAboutAI
Жыл бұрын
Yes:) i dont know about free, but low cost yeah:)
@brianj7204
Жыл бұрын
@@AllAboutAI why not free? (btw i'm Dutch i love free stuff)
@maikonpagani1222
Жыл бұрын
@@brianj7204 because it has server costs, if it's "free" it would be full of ads and using your information.
@witness1013
Жыл бұрын
The last line in the instructions to the agents around @2 made no sense to me.
@DirtyRobot
Жыл бұрын
AI likes mixed-race women for some "unknown" reason. For some reason the races that are being mixed are very specific.
@toolteardown1
Жыл бұрын
I wonder what the result would be if you made the two bots compete and a 3rd bot to decide the winner. The loser would have to be eliminated from existence of course. :)
@AllAboutAI
Жыл бұрын
Great and fun idea:) thnx for tuning in!
@Elit3KD
Жыл бұрын
Thing is, GPT engine do not typically disagree with itself so maybe it's not so different as using one engine and feeding back the responses
@AllAboutAI
Жыл бұрын
I think i can change it up with system prompts, but the conversation will be much longer, so i skiped it for this one. Thnx for tuning in :)
@Elit3KD
Жыл бұрын
@@AllAboutAI Thank you for making I look forward to it
@takumithao1992XD
Жыл бұрын
I wanna see two AI bots in a roast fest, just going at it.
@barulicksama3838
Жыл бұрын
Can you force them to discuss on how to extinguish humanity? So we can be prepared ofcourse.
@AllAboutAI
Жыл бұрын
I guess haha😅 thnx for tuning in
@juhu3709
Жыл бұрын
Why do you need two agents for this? wouldn't one totally suffice for this use case? They kept agreeing on everything anyway
@PCAN411
Жыл бұрын
Of course, they're gonna give the same answers to getting information from the same place.
@justinnine4940
Жыл бұрын
those are just exactly the rookie user generated images who are new to SD...
@AllAboutAI
Жыл бұрын
Rookies for sure 😅 thnx for tuning in
@RomboDawg
Жыл бұрын
You really shouldn't put the github repo in the video if it shows private repos in that frame, anyone could easily brute force the url to the repo and not have to pay for your membership to download from your private github repo meant for members only
@RomboDawg
Жыл бұрын
I suggest at least blurring the repo names next time
@admortem8901
Жыл бұрын
but can it generate NSFW content? ;D. how about implementing alpaca instead of chatgpt ;D?
@RAMENAKASHICHANNEL
Жыл бұрын
How many Agent can colaborate? 100?
@giorgisopromadze3655
Жыл бұрын
Isn't prompting engineering only needed cuz midjourney and generally ai art generators are illiterate unlike chatGPT? and as soon as devs incorporate chat gpt into it the "prompt engineering" will become obsolete as fast as it was conjured up?
@AndrewQPower
Жыл бұрын
I've never seen agent 007 be such a simp 😅
@barulicksama3838
Жыл бұрын
Can you have 3 agents with controversial tastes?
@AllAboutAI
Жыл бұрын
Have not done 3 agents yet, will look into it tho:)
@rustyxof
Жыл бұрын
This conversation might be interesting to you: kzitem.info/news/bejne/rY6E3WWKo198amU
@theartificialrealms
Жыл бұрын
Then they said goodbye 15 times hahahahaha
@HasimFN
Жыл бұрын
Those has to be from the internet
@2o4II112II26o2
Жыл бұрын
spelling is : professional not proffesional
@jd_turner_ai
Жыл бұрын
This is absolutely unnecessary lol Just finetune an LLM model to create amazing prompts and then automate. So much more straightforward and will provide better prompts than this.
@malakiblunt
Жыл бұрын
BULL SHIT !! name even one 'captivating' famous image that was a colaboration - now name any famous work of visual art - more than one creator - de vinci - picaso rembrandt Michael angelo mattiese -
@davaguco
Жыл бұрын
Im sure you could contemplate different IA agents as different parts of an artist's brain.
Пікірлер: 113