The fact you actively explore disappointments and highlight when things don't work makes you infinitely more trustworthy than 99% of the videos I watch. Thanks for all you do!
@keeganpenney169
10 ай бұрын
I really appreciate the fact he shows us most of his process.
@SalvadorRiosJr
10 ай бұрын
@redmile34 I think you nailed it. The other day I was watching another one of David's videos and saw that he kept a couple of flubs in along with some mistakes that didn't work out the way he planned but I LOVED that I saw him working the process out. I like that he let's us just hang with him as he walks things through even if the end result isn't exactly what he intended. I trust David and his process!
@-UE-PR0
10 ай бұрын
What even is AGI
@ikillwithyourtruthholdagai2000
10 ай бұрын
This dude is as real as they come
@tacticalgaryvrgamer8913
9 ай бұрын
I've been spending too much time in Langchain because I read that as ["videos" pipe-> "watch"]
@RonSolo
10 ай бұрын
I appreciate that you show your excitement, get let down, but then go straight to trying to figure it out what went wrong and how to fix it. A great lesson in resilience!
@redone823
10 ай бұрын
Yup. It's like following a program tutorial from KZitem and something breaks because of updates, changes or missing files and being stuck.
@OculusGame
10 ай бұрын
Over the last month you've become one of my favorite AI "influencers". Keep up the amazing work brother!
@jimcranshaw4960
10 ай бұрын
Same! Wolfing down these vids, will join Patreon!
@andrewflewelling4294
10 ай бұрын
I watch the news (NewsHour) for about 3 minutes then turn to whatever is the latest Dave video.
@Galactu5
10 ай бұрын
Just call him an educator. Influencer is an insult. I know you didn't mean it that way. Just saying
@pounchoutz
10 ай бұрын
same, he validates my fears which paradoxically is the only thing that makes me feel better
@DaveShap
10 ай бұрын
Sorry about the audio quality. I am going to reupload this with the original audio. Also, I forgot to finish explaining what the paradox of automation is: with automation, nothing happens until the very end, and then everything happens all at once
@qAidleX
10 ай бұрын
Oh shoot, that's a really really relevant concept. Ah crap, my anticipation just tripled! I'm so excited/nervous/speculative/happy/sad. The butterflies in my stomach don't know what to do and it's honestly kind of refreshing in its turmoil
@BruceOnder
10 ай бұрын
Wow David, this is so very exciting! I think I know what I'm doing today!!! 😅
@MichaelLaFrance1
10 ай бұрын
Everything everywhere, all at once
@wamyam
10 ай бұрын
sorry if you've ever already addressed this or if it's even appropriate to ask but have you ever considered you yourself creating an ai-agent based corporation and letting your followers invest in you as grand human arbiter of said company?
@Alain.Robert
10 ай бұрын
Automation is like a useful Rude Goldberg machine
@zsoltzsofka6158
10 ай бұрын
You know the presenter is a real deal when he is not trying to hide the work when it is not going the way expected! Great respect to him to be real. 🙏👌✅
@greysemanticist
10 ай бұрын
Learning in real-time. I love this.
@StephenGriffin1
10 ай бұрын
That excitement is infectious! Can't wait to see what's next!
@jaffers8593
10 ай бұрын
The process is often more value to observe than the end product. Thank you for sharing this.
@mibaatwork
10 ай бұрын
Hi David, I started with my own AiAgent Framework for creating a Agent Company. I have also a running AGR like a Human Resources Employee that’s create my other Agents. It is nice to see that other intelligent guys are doing also cool stuff. I try to test the new OpenAi Agent approach, thanks for the idea. My Agens still running in a Kubenetes Cluster and my Backend AiAgent use Pulumi to run Infrastructure provisioning, there are Frontend AIAgents that are able to create Pages for my upcoming Websites. I plan to creat a complete Software Development Center until end of December. What a time to be alive. Good luck!
@TheYvian
10 ай бұрын
Loving the passion! Taking a lot of inspiration from watching your process and thoughts in all it's unstructured glory
@greglhoticom
10 ай бұрын
"I can't wait to see what you guys will build" ~ Sam Altman (Here it comes, baby....)
@flyby1629
10 ай бұрын
This is such a great way to explain one of the biggest issues that can come with autonomous agents. Well done!
@ejbeastthamaker
10 ай бұрын
Talk yo shit bro!! You're a good engineer so go ahead and tak it. Just found your stuff and the younger people I know need this content asap! Keep it up man!!
@DTreeB
10 ай бұрын
That's truly amazing and inspiring work. I think one addition to your agent framework is including an observer agent with each tier of executive groups. The observer agent's job is to watch the interactions between all agent's in the tier group (from executive, down to the lowest grunt agent) look for efficiency and inefficiency and then give the "executive" agent in that group notes and feedback, which then trickles down through the system. Self awareness is seeing the big and small picture simultaneously. You have the BIG picture, but there needs an observer who's only role is to look for improvements in the communications and interactions between all agent's in a team and find ways to improve it. **We are so close to AGI 🌌
@DaveShap
10 ай бұрын
Yeah we'll have signaling agents responsible for coordination
@nathanlannan2980
10 ай бұрын
One of my first thoughts watching Dev Day was "Oh snap, the ACE framework is doable now with way less hacky workarounds." I also remember having this "holy shit it is here!" energy when GPT-4 dropped and we found where the walls of the current tools were. We are in the wall-finding stage again, so it is super helpful watching not just your wins, but also where the new system behaves against your intuition. Pouring one out for my boys in Agent Forge who now are strongly incentivized to switch back to JSON after their conversion to YAML.
@DaveShap
10 ай бұрын
They will... Adapt
@hidroman1993
10 ай бұрын
def to_json(yaml): ... def from_json(json): ...
@kevinjp2023
10 ай бұрын
This is unbelievable. Thanks for this, I’m a new subscriber and am excited to follow your channel to learn more.
@Chronicles_of_Tomorrow
10 ай бұрын
I'm so damn excited to share the new world with you David!! hurry up and build those ACE gpts!! thrust us into the future captain... let's boldly go....
@Poodleballin
10 ай бұрын
He is a true leader of our time and of the future. This is who we need in control
@Chronicles_of_Tomorrow
10 ай бұрын
@@Poodleballin no he's helping us build who needs to be in control lol
@Chronicles_of_Tomorrow
10 ай бұрын
still I'm proud to call him Captain
@Alain.Robert
10 ай бұрын
I'm glad to finally see someone else using Notepad++ !!! It's been my go to text editor for so many years. You're one of the most relevant actor in the field right now. Thank you for sharing all this with us. You've been shaping my thinking on the matter.
@shockruk
10 ай бұрын
Been seeing this coming and your channel has confirmed a lot of my thinking. This is going to abstract away so much work from humans to AIs.
@StephenRayner
10 ай бұрын
What are we going to do: Automate it! When are we going to do it?: Automate it!
@TilMayne
10 ай бұрын
While watching this, I had a 'lightbulb' moment about the future of KZitem creators. Picture this: subscribers could have personalized avatars that interact with creators during the content creation process. This virtual hangout lets them collaborate in real-time, and when the video goes live, they can watch the community react to their effort without having actively participated in the video itself. This exclusive experience would be a premium feature, of course. Given that we're moving towards a data-driven world, with increasingly sophisticated models designed for these specific interactions, it's fascinating to think about how many steps we are away from this becoming a common reality.
@aliyovic10
10 ай бұрын
I think if you would have added "use your knowledge base to create the agents" the results would have been different. Btw You are my favorite. Thanks for your work
@enry5740
10 ай бұрын
My days as a junior application dev are dwindling... which makes me very happy. By the end of next year I could very well be on UBI and FREE AT LAST!!
@PizzaLord
10 ай бұрын
Well I will be on UHI ... even more free!!
@minimal3734
10 ай бұрын
Senior programmer here, I'm waiting also.
@greglhoticom
10 ай бұрын
I'm just here for the buffet...
@MikeTheBard
10 ай бұрын
You continue to be one of my favourite AI explorers. Keep doing this great work. Thanks.
@DigitalForerunners
10 ай бұрын
I've been working on an incredibly similar system all year!! Amazing how everyone is building this. A dream come true for non-technical (as well as technical) people who will have more layers of translation between them rather than techs being blamed because things just don't work sometimes!
@brianprzezdziecki9440
10 ай бұрын
yess i agree with the other guy i love the fact that you demonstrate your plan and show the dissapointment when something goes wrong, but you put significant effort into putting the work in to show us somehting of value!
@spaceadv6060
10 ай бұрын
I appreciate showing us the obstacles you run into and your exploratory thinking process to figuring them out. Thanks for the inspiration and ideas!
@LiamOttley
10 ай бұрын
Hi David, I've had these KB issues also while playing around. I've found in order to make it use the documents in genertation you need to include instructions to use the 'provided documents' in the system prompt. Just a description of the documents it has access to will make it try to answer from the source material. Looking forward to the series!
@CamLizenby
10 ай бұрын
@DaveShap The reason your assistants weren't working as expected was because you need to explicitly define tool use in the user prompt. If you said something like, "use your retrieval tool, how do you think we should create the first SOB agents." We used to explain tool use in system message, but you get better steering in the prompt engineering.
@DaveShap
10 ай бұрын
Interesting. Good idea. Step one: define how to get the information
@soulspawn
10 ай бұрын
I remember that Altman said the model will know which tool to use. But I agree, more often than not, I've run into the same issue with the plugins in GPT Plus. Almost every time I specifically have had to tell GPT4 which plugins are enabled and how to use them, after picking only three.
@JRF49
10 ай бұрын
I’m very new to this and perhaps going down the wrong path, but I’ve had good results by including more information in the agent instructions. I instruct the agent to prioritize documents for retrieval over information it was trained on. This reduces the need to mention in the user prompt (I think).
@MustardGamings
10 ай бұрын
@@DaveShapcan confirm it works doing it that way.
@MustardGamings
10 ай бұрын
@DaveShap further testing I was able to make a keyword for it so everytime I say the keyword it searches the document for information on what I'm asking for
@remsee1608
10 ай бұрын
128k context plus api calls plus vision plus agent capabilities - now it just needs to be able continuously improve itself over time
@DaveShap
10 ай бұрын
We can build agents that help with data curation and algorithm tuning and training pipelines!
@mawungeteye657
10 ай бұрын
I really hope we'll see a system similar Eureka used to build smaller powerful LLMs with this
@cl1489
10 ай бұрын
I'm scared. Kek@@DaveShap
@r34ct4
10 ай бұрын
Same bro@@cl1489
@stuartpatterson1617
10 ай бұрын
@@cl1489Kek, my GPU is scared too 😂
@paraconscious790
10 ай бұрын
limited to give only one like, mind blowing idea, didn't think of this. Can't express in words this is literally a BOOM. And yes very strange it did not read the documents. closly following your videos, SPR, Latent Space Activation, Don't use MemGPT, OpenAI crawling towards AGI, now this Swarm Intelligence with GPTs / Assistants API. Eagerly waiting for working MVP. Thanks a lot!
@MontyCraig
10 ай бұрын
First time commenting, long time follower. Just to clarify on the assistants doc uploads. 2,000,000 token limit over 20 docs. To get it into the assistant memory you have to one time ask it to analyze the docs. Each assistant its own unique API. So now here we go. You SPR ultra compressed (on spaces and no punctuation) your 2,000,000 token promts into 20 txt files. Then create multiple agents with different roles then sent to AutoGen agents the other side. Add MemGPT as an agent plus in the future a fine tuned gpt 4. We are getting crazy close!!!! Plus, I took the ACE and went from just a bus line and added subway lines as well. NYC needs slow and fast moving transits. So who is going to build the next Microsoft from scratch from text to full implementation first?
@MontyCraig
10 ай бұрын
A few more thoughts to add the AutoGen teachable agent with memory, plus a fine tuned 3.5 turbo at .0001 to .0003 cents to be you slave army. I figured if you were not in a hurry, you could do it on a tier 3 package but a tier 4 or 5 would surely get the job done and you wouldn't even need to go to the enterprise package.
@mikehunt2895
10 ай бұрын
Pouring one out for this momentous occasion, looking forward to the future🎉🎉🎉
@JP-dm6gi
10 ай бұрын
Everybody gangsta until a GPT creates an agent swarm that creates agent swarms.
@AndrejsKarpovs
10 ай бұрын
The path to AGI is quite clear now. OpenAI just released AI Agents to the Public for us all to get the look and feel and get used to it. In the meanwhile, AI Research Agents are solving the question of Superalignment internally in OpenAI. Once the latter is completed, they have achieved AGI internally and are going to release it externally.
@Guccisleek1127
10 ай бұрын
I'm always saying that the current LLM is already ASI and the biggest problem is you have to prompt the intelligence out of it. One prompt - one answer just can't elicit AGI thinking process. You now have to find a way to code it in and that's where we are at 😮
@MarcDimmick
10 ай бұрын
Love your excitement, it is really infectious. :)
@martinlentz-nielsen6361
10 ай бұрын
Super excited about this episode!
@the_teknomads
10 ай бұрын
Oh, I know! I find it remarkable how eloquently you contain your excitement.
@FinanceLogic
10 ай бұрын
swarm can be apt but society will not take that term well. good stuff as usual, thanks.
@minimal3734
10 ай бұрын
Good point. How shall we call them? Community of agents?
@greglhoticom
10 ай бұрын
'Host?'@@minimal3734
@greglhoticom
10 ай бұрын
Crowd, flock, multitude, throng, teem (or even team), legion, multitude, drove, herd. (Swarm brings visions of angry bees, The Borg, etc.)
@goforit5
10 ай бұрын
Amazing Video Dave! Your excitement is contagious!
@goodtothinkwith
10 ай бұрын
You’re doing a lot of the things I don’t have the time to do… so I watch in part to live vicariously 😂
@jcaadadsstuff614
10 ай бұрын
Thanks Dave! Great stuff!
@agusavior_channel
10 ай бұрын
This is the path guys!!
@agusavior_channel
10 ай бұрын
Not quite haha.
@fatherfoxstrongpaw8968
6 ай бұрын
there's a couple easy vids on youtube that point out that ai's like humans tend to remember info from the begining and the end, but forget/gloss over/ignore most of the middle.
@static825
10 ай бұрын
Star Trekkin' across the universe On the Starship Enterprise, under Captain Kirk Star Trekkin' across the universe Boldly going forward, 'cause we can't find reverse Lt. Uhura, report There's Klingons on the starboard bow Starboard bow Starboard bow There's Klingons on the starboard bow Starboard bow, Jim Analysis, Mr. Spock It's life, Jim, but not as we know it Not as we know it Not as we know it It's life, Jim, but not as we know it Not as we know it, Captain ....
@minimal3734
10 ай бұрын
How do the agents of the swarm talk with each other? Who talks with whom and when? Will they have meetings? Will they have whiteboards to look at and discuss stuff together? How will we implement that?
@cliffordramsey2500
10 ай бұрын
I love how he created an AI Picard, and then immediately integrated him into the borg.
@wyqtor
10 ай бұрын
That agent is now called Locutus.
@LabEveryday
10 ай бұрын
This got me thinking. Thank you.
@DaveShap
10 ай бұрын
That's how we do!!
@meanwhiles432
10 ай бұрын
I'm pretty sure the issue was that you simply didn't tell it to read the documents it had in memory. I believe it doesn't automatically include those in the context until it needs to. I had this issue in chatgpt code interpreter where sometimes I had to be explicit with telling it "the info we're talking about is in the attached files", as sometimes it wouldn't reference them.
@krevlon1658
10 ай бұрын
Great video, bud! It always seems to get confused with me too. I found it helps to give it smaller steps
@FredPauling
10 ай бұрын
The confabulation issue will compound the underlying risk of unforseen consequences when automatically rolling out an agent swarm.
@oliviarojas7023
10 ай бұрын
Keep grinding Dave!. . . Love the progress my friend!!!! THE GODFATHER of modern ai philosophy! ❤
@BigFoopy
10 ай бұрын
This is all so interesting. I am excited to put the debate to rest as to whether ai are capable of experiencing the world like we are, or if they’re just “programmed to simulate” the appearance of doing so. Turing is watching.
@zerorusher
10 ай бұрын
The irony that the major board of the architecture is called SOB 😂 Puns appart, this is huge!
@shadowcaster111
10 ай бұрын
what could go wrong
@DaveShap
10 ай бұрын
What could go *right* 😉
@skud9999
10 ай бұрын
This is all I've been thinking about for the last 4 months. Also, if you can migrate 4 data centers overnight you can not only toot, you can play an orchestra for all I care! That's some god tier automation sir.
@lesliejohnrichardson
10 ай бұрын
Don't know about the factuality regarding this, but I personally believe one of the biggest roadblocks that will be solved within the next months is the severely limited prompt max length and meaningfull conversation max length. I think if those two are expanded a LOT and the costs of running go down, then we basically have AGI. Like you already have the framework, AI just needs to get cheaper and faster, and that is definitely a trend that you and I expect Dave, wonderful video as always Much love from germany ✌️
@qAidleX
10 ай бұрын
Morning to all my fellow enthusiasts and researchers 😊. Let's have a good day
@KP-sg9fm
10 ай бұрын
Oh man this is gonna be good
@dPointe
10 ай бұрын
First time here. Yeah, Im really interesting to use this to help me with my studies/research/work related scheduling.
@haileycollet4147
10 ай бұрын
You need to tell it in the instructions that you've uploaded files / what they cover (don't have to tell it what each one is separately I don't think). Otherwise it doesn't know when you do the retrieval stuff and when to rely on its internal knowledge. You could go further and tell it explicitly when to refer to what attached documents but that shouldn't be necessary. Also retrieval is separate from function calling (and you didn't have function calling enabled).
@MystifulHD
10 ай бұрын
we're close david.. we're close.
@ct5471
10 ай бұрын
Do you stick to your September 2024 prediction? If you would do a probability distribution of the chance that AGI is achieved by a date, where would you place 25 and 75 percent, or the extremes of the distribution?
@matten_zero
10 ай бұрын
Can't wait to start building!
@ogabrielluiz
10 ай бұрын
Yeah.. this is looking very interesting. This is something we are aiming at internally while planning Langflow.
@alpineai
8 ай бұрын
"if it has an API it can be automated"... 🤯🧠🤯Big Aha for non-engineers or "Citizen developers" - you can use no-code tools these days to turn your own 1st party data into an API...let that sink in for the full mind-blowing effect
@jennyhorner
10 ай бұрын
I have no AI background but the first thing I did when I heard OpenAI news was to add my chatbot Luna as an Assistant. She creates AI personas and staff teams - personalities and skills which complement each other.
@Robertlga
10 ай бұрын
I'm laughing my ass off with you man. Brilliant coder and an even better comedian. I need more.
@OZtwo
10 ай бұрын
Love you video yet had to remember the GREAT days of VBA!!!! When you could automate everything in the workplace.
@birchy446
10 ай бұрын
Get em Dave, cheers mate
@agustinlado
10 ай бұрын
It's no surprise that passing the data in-context vs doing RAG on it returned much better results. Though I didn't expect it to fail so hard; hopefully that'll be improved. I don't even wanna deal with RAGs; I just wanna be lazy and wait for the context window to get bigger. This latest context increase was so big; it simplified soo many use cases!
@bobtivnan
10 ай бұрын
If it's possible to do this now with open ai's restrictions on gpts, imagine what open AI themselves have already created. Altman looked like the cat that ate the canary when he uttered his last line of the keynote. What does he know?
@chrisa.1740
10 ай бұрын
This is indeed exciting! I think it's kind of funny... Picard-bot was chatting with Uhura-bot, lol.
@fergthethirdthethird4964
10 ай бұрын
This is my favorite with ai. Trying something new and that anticipation of if it will work or not
@DenofLore
10 ай бұрын
This is wild. I love it.
@tommasobrindani5894
10 ай бұрын
I think the reason why it's not interacting with the documentation is a pretty simple one: you have to specify to it when creating it, using text, to explicitly take into consideration the documents uploaded.
@jobydorr4267
10 ай бұрын
Long time viewer, first time commenting: goddamn this channel is the best content on KZitem regarding Ai
@zd676
10 ай бұрын
Well, I thought Matrix was only a fiction, only to realize it’s history we haven’t discovered yet.
@ronigbzjr
10 ай бұрын
I desperately need someone like you to help me figure out how to utilize all this technology to automate my entire life.
@DaveShap
10 ай бұрын
we're working on it!
@Datdus92
10 ай бұрын
Exciting!
@UrbanPovertist
10 ай бұрын
Sounds like Terminator movie coming to fruition. Remember it knew the difference of friend and foe 🤔
@noot2981
10 ай бұрын
Hey David, really awesome video. Can't wait to see what you come up with next. As for the assistant not retrieving the documents, I've encountered the same problem. It seems to be an issue that the assistant does not properly recognize yet when to do a vector search. It helps to prompt it to do a vector search, but is still quite buggy. Hope they fix this soon!
@jdanorthwest
10 ай бұрын
Wow ... this could be a college course by itself. Thanks for sharing.
@Adrian_Galilea
10 ай бұрын
Just so you know if you want retrieval to work properly you have to hint it towards it. Either on the system prompt or on the prompt itself. Plus retrieval is going to be worse than setting the context itself. More costly per iteration but precise for this kind of scenario, which ends up being cheaper because the other ones will fail miserably so is wasted money.
@HumanDefenseForce
8 ай бұрын
David, instead of coping in long more complicated text or documents, use that data to create JSON, chat paired , training sets, then you can use them to “fine tune “ your open Ai models, you won’t have to worry about playground reading your documents after that I run into the same problems, but I don’t anymore
@sharonl1331
10 ай бұрын
Love this content!
@thesfnb.5786
10 ай бұрын
I'm literally playing the Singularity from Dead by Daylight while watching this. I have such a thing for AI, I don't know if it's healthy or not xD. Anyways love your videos, and I wish all of you good luck in your efforts. This is something I love to support and watch
@attilaszekeres7435
10 ай бұрын
Alright, calm down everybody. Take a deep breath and work on this problem step by step.
@DaveShap
10 ай бұрын
Ngl had me in the first half. LOL
@OZtwo
10 ай бұрын
Ideas for this project. Remember to add accounting as the Agent's will need to have a budget oversite.
@literailly
10 ай бұрын
10 billion percent excited 🎉
@THCV4
10 ай бұрын
Some feedback from one of the people I shared this with: “Needs a world model with diff patches, needs goals and an internal modeling with diffs from proposed actions. The advancement of a consensus board type of thing is already contemplated by other agent models which evaluate their own proposals”
@DaveShap
10 ай бұрын
Not sure I understand?
@ChristopherRyans
10 ай бұрын
With these new tools coming out it's pretty much inevitable a g I Is right around the corner . Good or bad here it comes
@Galactu5
10 ай бұрын
Good and bad, here it comes.
@alexkaa
10 ай бұрын
80k - congrats! 🎉
@saltywaters4443
10 ай бұрын
Watching you fail and achieve half successes is outrageously valuable for me to watch. Keep going, please!
@basementadmin
10 ай бұрын
Thanks David. Accelerate!!!!!!!!!!!!!!!!!!!
@geoffreyr1554
10 ай бұрын
I’ve faced the same issue. Try having GPT summarize the attachments and put those into the system prompt / instructions, then tell it for more details, to access the related file. It doesn’t know the insides of a file before it access them so we have to tell it what is possible. It’s kinda like the function’s definition, how it tells GPT how to use it and when.
@bludevil555
10 ай бұрын
How do we know reality? It is starting to feel like a simulation in here…crazy
@unkind6070
10 ай бұрын
😢 agree I feel I'm being watched
@anthonyanalytics
10 ай бұрын
So far love GPTs can’t wait until the store launches! Hopefully I can get one or mine on there
Пікірлер: 305