How I Made AI Assistants Do My Work For Me: CrewAI
How I Made AI Assistants Do My Work For Me: CrewAI
In this video about crewai I’ve decided to show you:
1. how to build your own team of ai agents that debate and think about your business idea from multiple angles
2. how to give your agents access to real world data like google searches and reddit
3. performance of 15 models in total: 13 local models + 2 through api
🤖 Join my Discord…
source
Reviews
0 %
UPDATE: Thanks to the viewer @tryingET great suggestion, I managed to improve the prompts and make the output consistent. You can check out the improved version on my github.
Thanks for watching and I'm curious, what were your experiences with crewAI like?
This is so over my head it is not even funny!!! But oh so exciting!!! Need to find some time to look into all this!!! Where does one even start? Thanks so much for the video!!!
Now make PissAi.
Automate your pissing.
You are an idiot if you believe Ai is capable of HIGHER LEVEL THOUGHT.
It is NOT.
You are being sold a "pack of lies". AI can only do what it's programmers tell it to do….PERIOD. DOT. END.
If it's programmers CLAIM AI can think, think again. It cannot.
AGAIN, AI can ONLY do what it is told (meaning programmed to do)….period, dot, end.
Anyone telling you different….IS A LIAR.
Your brain is the greatest thing in the universe, billions of years in the making. Why are you wasting it trying to make a Ferrari out of a bicycle?
Build an automatic object tracking and shooting the object system
aweasome!
This is awesome! Building a team of AI agents that can access real-world data sounds incredibly powerful.
"You're missing out on one of the best, and that's Claude, who often outperforms OpenAI. I see a lot of programmers using it a lot."
3:40 😄
Can you do an update video for llama3 and phi3?
Thank you! This content is so good.
Hey guys! I'm immersed in the study of AI agents and I'm curious: would it be viable to build an agent that prospects customers for freelance professionals? I envision a system capable of exploring Instagram in an automated way, identifying potential customers and even starting conversations to schedule sales meetings. Is it possible to develop such AI agents? If so, do you know of any videos on YouTube or any mentors that explain how to create AI agents to automatically prospect customers through Instagram? Is there something like this in development or is this an idea for the future of AI?"
Great presentation. Congrats
very informative thank you. something like a carpet or room dampening could help the sound quality of the stream. thanks again.
Did not find openai_api_key, please add an environment variable `OPENAI_API_KEY` which contains it, or pass `openai_api_key` as a named parameter. (type=value_error)
# To Load GPT-4
api = os.environ.get("sk-proj-fNk……")
I completely understand how important it is for you to convey your content. May I suggest adding some upbeat or classical background music to enhance the overall experience? It can really make a difference and help create a more learning atmosphere. Good luck & God bless!
Even as a front end developer I would rather have an interface, this is very technical and intimidating to get into,
Great content, thanks!
The thumbnail is funny 😂
I don't trust wearers of Crocs.
Great video and LLM review.
great, very useful & interesting. Thanks a lot for your great video
this is so in-depth and really appreciate your hard-work and dedication Maya.
I couldn't get passed the missing R. It bothered me very much.
Tremendous! incredible research and findings ! THx
🎯 Key Takeaways for quick navigation:
00:00 🤝 Div Gar visited the LLaMA team at Meta and sees potential for collaboration, with Meta becoming a major player in open-source AI models
– Mark Zuckerberg is committed to building the best open-source models
– Potential future partnership to bring Mulon's agents to improve services across Meta's offerings
00:41 🤖 Zuck has a robotic public persona, but perception is turning around due to better marketing
– Div is bullish on open source catching up to closed source models like GPT-4
– Open source only needs data, compute, and talent to build the best models at scale
02:21 💼 Closed source models currently have the advantage in data and talent
– As open source becomes more common, there will be pressure for researchers to open source from different labs
– With equal access to resources, open source has the long-term advantage over closed source
03:17 🔒 AI research went from open to closed source due to profitability and perceived risks
– Companies thought AI models could become as dangerous as nuclear weapons
– Training billion-dollar models requires recouping costs through closed source offerings
04:39 *💸 Making the AI model the core business is a risky strategy *
– Companies are moving into new spaces like video modeling to stay ahead
– Long-term, useful applications that people are willing to pay for will be the most sustainable
05:35 🏛️ Google faces bureaucratic red tape in open-sourcing state-of-the-art models
– They are confused about strategy and whether to copy OpenAI or do something different
– As a non-core business, Meta can afford to open source without impacting metrics
07:28 🌐 Open-source pressure from Meta, Stability AI, and others may force Google and OpenAI to follow suit
– As OpenAI becomes a for-profit business, open-sourcing makes less financial sense
– Meta's strategy with LLaMA is to become the default framework for developers
09:16 🔓 Current AI models are not capable or risky enough to warrant closed source approach
– No major incidents in the past year with open models instills confidence in limited capabilities
– Agent AI solutions pose greater risks for malware, spam, and bad actors amplifying negative impact
11:21 👮 Open-source AI could enable development of robust anti-virus and anti-spam systems
– Race between viruses and anti-viruses as AI agents become more capable
– Moderation and alignment are critical, especially for agent AI, to prevent illegal activities
13:27 🚀 Mulon's vision is to build an AI assistant to automate boring everyday tasks
– Goal is to free up time spent on repetitive actions like emailing, booking, and paperwork
– Empower people to focus on high-level creative tasks by having a personal AI helper
15:21 📧 Most common use cases for Mulon are emailing, reservations, and ticket booking
– Potential for AI agents to handle complex tasks like planning an entire trip
– AI agents could be like giving everyone a personal intern to delegate tasks to
16:30 🤝 The next agent revolution will be unlocked by building user trust and capable, reliable systems
– Agents need to be controllable, listen to users, and consistently complete tasks without fail
– Seamless interaction patterns and an engaging user experience are also important factors
17:38 🤨 Average users are skeptical of trusting AI agents with critical tasks like email
– Lack of technical expertise makes AI seem like a "black box" that could be untrustworthy
– Visual interfaces showing the agent's actions can help build trust compared to invisible processes
18:46 👀 Visual interfaces build trust in AI agents by allowing users to see the agent's actions
– No way to fact-check invisible processes, but seeing an agent successfully complete a task multiple times instills confidence
– Making AI agents accessible to non-technical users is the next billion-dollar opportunity in UI design
21:06 🎨 Technical people often underestimate the power of UI because they focus on the underlying technology
– Non-technical users care more about the user experience and visual design than what's happening behind the scenes
– Great UI should change based on the user's goals and the underlying technology to provide a seamless, intuitive experience
23:10 🙋♂️ Mulon started as a solution to the founder's personal pain point of wanting to focus on research instead of logistics
– People are looking for AI assistants that can actually do useful things for them, not just engage in conversation
– Partnering with companies building smart hardware can enable the creation of highly useful AI agents that people already want
25:18 🚀 Mulon transitioned from a personal project to a company after a viral Twitter video and strong user demand
– Surveys showed people were willing to pay $100/month for an AI agent that could handle tasks for them
– Seeing people adopt and use something you've built as an inventor is highly fulfilling and motivating
26:42 🤝 The goal of AI agents is to empower humans, not replace them, by handling boring tasks and enabling more to get done
– Agents can reduce multi-step interactions to just a few approvals, saving time and mental bandwidth
– Parallelization allows humans to accomplish 10-100 tasks at once through their agents rather than being limited to one at a time
29:02 🕸️ A probable future involves agents interacting with each other to save humans time on coordination and logistics
– Automating small 2-minute tasks compounds to save significant time and reduce the "death by a thousand cuts" of minor annoyances
– Offloading these microtasks to AI could reduce mental burden and stress, leading to happier people
30:38 🍽️ The founder optimizes his own life using Mulon's agents for tasks like ordering lunch and managing documents
– Delegating to an agent removes the mental burden of figuring out how to complete the task yourself
– Mechanisms are built in for the agent to ask for help or login credentials when stuck to keep the user in the loop
32:29 📈 Crowdsourcing and network effects enable rapid improvement in agent reliability as more people use them
– The first user may have a 50% failure rate, but this quickly trends toward zero as the agent learns from more interactions
– Small improvements compound significantly but are often underestimated when evaluating the potential of AI agents
33:52 *🗣️ Striking the right balance in how often an agent asks the user for input is key to avoid annoyance while maintaining transparency *
– The agent should be friendly and helpful without becoming a nuisance that asks for guidance too frequently
– In the near future, most people will likely be managing teams of AI agents to optimize their lives and productivity
35:19 ⏳ Widespread adoption of AI agents as the new norm is expected within the next 2-3 years
– By the end of 2023, many people may already be using multiple agents for various tasks
– Not having an AI agent could become as unusual as not having a phone or internet access today
36:57 🚀 Widespread adoption of AI agents will happen faster than previous technologies due to the rapid pace of the digital world
– Even non-tech savvy people are becoming aware of AI through experiences like ChatGPT
– Social media enables new trends and information to spread virally, accelerating the adoption curve
38:20 🏗️ Mulon is using NVIDIA H100 GPUs to train proprietary action models focused on alignment and reliability
– Goal is to reach 99% reliability for key use cases within 3 months to build user trust
– Scaling inference cost-effectively to support millions of users without bankrupting the company is a major challenge
39:46 🤖 Robotics is one of the most promising areas in AI right now, with potential for rapid progress
– Robots could start appearing on sidewalks, in restaurants, hotels, and even homes within the next 5 years
– Robots are relatively inexpensive to manufacture compared to the cost of developing the underlying AI technology
41:37 🧠 AGI could manifest as either digital or physical intelligence, depending on the definition used
– A digital AGI might have comprehensive knowledge of everything on the internet, similar to Iron Man's Jarvis
– Physical AGI would likely involve robots capable of performing a wide range of specialized tasks better than humans
43:10 🚘 Tesla's Optimus robot has shown impressive progress, with rapid improvements between versions
– Elon Musk's track record suggests betting against him is unwise, especially in areas like robotics
– If Tesla maintains an exponential improvement trajectory, human-level robots could emerge within 3-5 years
44:19 📈 Exponential progress in AI is often underestimated, with new areas constantly emerging as others level off
– Language models may be nearing a saturation point in terms of scaling and improvement
– Agents, robotics, and AI hardware are just beginning their exponential ascent and have immense room for growth
Made with HARPA AI
Great content, thank you. Subscribed!
I've tried 4 different agent frameworks, including CrewAI. Absolutely none of them work out of the box with the default install and official examples. Not one. To be clear, I am not talking about simple issues like not setting an API key – I mean flat out, they do not perform.
How do you build multi-agents in CrewAI with Google_API_Key? I tried but failed. Can you help me, please?
Can I use/create something like to help with YouTube content and SEO?
Good stuff. Do you have any videos on how to make these agents executable from an app or a program ?
Love this. Really transparent and talking about limitations of models..
thanks maya. it's sad to hear open llm's were not performing well when you made this video but now llama3 is released as well as other improvements in that era. maybe its worth giving another chance again.
lol comment section full of bots
How I Made a Youtube Video Do My Work For Me
Actually a great video to start with AI agents, thanks
Could you cover running the crewai tools with ollama. Having a heck of a time understanding how to get the website searchtools to work