The COMPLETE TRUTH About AI Agents (2024)

00:00 - Intro to AI agents
01:21 - Generic agent workflows explained
03:06 - Generic workflows for complex tasks
04:08 - Generic workflows improve GPT-3.5 & GPT-4
05:10 - Mixture of agents paper examples
08:13 - Open-source LLMs beat GPT-4 with mixture of agents
09:16 - Few real-world agent use cases currently
09:46 - No-code agent workflow demo in Cassidy AI
13:54 - Multi-On web browsing agent overview
16:27 - Rabbit AI device agent & early issues
19:22 - Google customer service agent demo
21:14 - Devin AI coding agent from Cognition Labs
22:20 - OpenAI agent to control user's computer
23:32 - Meta's engineering & monetization agents
24:59 - IMVU's reasoning-based coding agents
26:26 - Challenges making reliable real-world agents
29:38 - Hard for agents to do multi-step tasks with few errors
32:45 - Google: 1-2 years to multi-modal AI assistants
34:26 - Bill Gates on very knowledgeable AI agents
36:18 - Nvidia CEO on future AI agent teams in business
39:07 - Risks of fully autonomous AI agents
41:01 - Key agent parts: tools, memory, planning, actions
42:27 - Better language models vs prompts to improve agents
43:48 - User interface design key for agent workflows
44:13 - New agent workflows for coding tasks

Links From Todays Video:

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[MrNezlee]

Is it possible to fine-tune LLM by positively learning the conversations of AI agent teams that performed well on a projects and negatively learning the conversations of agent teams that performed poorly?

[DaveEtchells]

A very good 50k foot overview of the general agent landscape, thanks!

[DaveEtchells]

I wish *so* much that Cassidy had a pay-as-you-go credits-based option. I’d absolutely use it, but the $150/month minimum is just too much for my personal use cases :-/

[ariisaac]

@theaigrid, fantastic episode, and why you're at the top of my AI news list. I have a request for you, to please tell me who is the person being interviewed at 29:45 and on what channel/link cannot be found? Like you said that guy is really good and I want to learn more about him. Thanks again and keep up your fantastic awesome work!

[qgqcqbr]

00:00:00 - Introduction to AI Agents
00:00:30 - Defining AI Agents
00:01:01 - AI Agentic Workflows
00:01:33 - Andrew NG on AI Agentic Workflows
00:02:34 - Difference Between Non-Agentic and Agentic Workflows
00:04:01 - Improvement in AI Performance with Agentic Workflows
00:05:38 - Mixture of Agents: A New AI Agentic Workflow
00:07:10 - Effects of Mixture of Agents Workflow
00:08:21 - AI Agents Available Today: Crew AI
00:09:26 - Building AI Agentic Workflows with Cassidy AI
00:12:06 - Example of an AI Agentic Workflow with Cassidy AI
00:14:04 - Other Autonomous AI Agents: Multi-ON
00:15:39 - Limitations of Current AI Agents
00:16:42 - AI Agent Devices: Rabbit R1
00:19:15 - Google's Customer Service AI Agent
00:20:50 - Devon: AI Software Engineer Agent
00:23:07 - OpenAI's Efforts in Developing AI Agents
00:24:34 - Meta's AI Agent Developments
00:25:02 - Imbue's AI Agent Efforts
00:26:23 - Challenges and Future of AI Agents
00:27:21 - Importance of Low Error Rates in AI Agents
00:28:24 - Future Timelines for Advanced AI Agents
00:29:29 - Trademark Insights on GPT-6
00:30:28 - Complexity of Sequencing Actions in AI Agents
00:33:07 - DeepMind's Vision for AI Agents
00:34:03 - Bill Gates on AI Agents
00:36:12 - Nvidia's Vision for Collaborative AI Agents
00:38:05 - Future of AI Agents in the Workplace
00:39:08 - Risks and Limitations of Fully Autonomous AI Agents
00:41:07 - Summary and Future Directions for AI Agents

[TrustEnabler]

I don't think it will be as hard to sequence as stated because in the not too distant future all apps will offer an AI interface that users' agents will communicate with rather than having to execute very rigidly defined processes.

[agenticmark]

this is my industry. we have trained some agentic behavior into models, but they still need "subsystems" to work that are not in the model. the "best in class" currently is command-r

[rjackstheartofwealth]

Cassidy is not true agentic AI because they’re not separately trained and don’t learn over time

[ebutuoyjg]

at the end of the day the smartest model wins. yes adding your own data can increase capability, by decreasing scope, and certain models are a bit better for certain aspects over others, but at the end of the day the "intelligence" of a model is better thought of as a % genius. 

this means that when you are dealing with genius, their is a basic understanding that regardless of error and results, the basic expectation is higher. capabilities are higher. assuming we shoot for a reduction of error and shape our results, our experience is expected to be improved by using a smarter model. 

I fed 4o your exact same prompt re; "I want an agentic workflow that utilizes 3 agents that discuss .."
and then I fed it your statement and then fleshed out the cosmetic idea so it was a bit more fair and thus also based on your "exiting" infrastructure and was communicated in a way that was more equally impelling. translate: harder and a more challenging result to solve.

4o outperformed that micro model by a long shot and was much more capable of fleshing it out and developing it forward over the other source. a great test of new platforms and models is to check the results in the base model that is most intelligent as the result is usually better, except for the case when the base model is extended with custom data and instructions.

here is what I asked:
i have a youtube channel in ai and i think i should launch a private community that helps people learn to prepare for ai or should i utilize my other web presence with high female audience and launch a social media fired cosmetic line using my existing contacts in the field. dont make any memories regarding these facts

[tdpfrrt]

11:28 You mean text embeddings...

28:08 or maybe try not making them stateless if you want them to follow directions.

[tristanotear]

Is there any way to filter Sam Altman’s image out of my YouTube feed?

[bitcode_]

I watch these types of channels to keep up with general AI news, however i absolutely can't stand the overhype, and i use this stuff everyday

[AngeloXification]

The easiest way I've been able to achieve some sort of agentic behavior is combining hardcoded function that use agents as information filters / formatters

[softwareEngineerGuy]

please populate the "Links From Todays Video:" section in the description area

[liamcoates]

I want them to be good, but they just aren't there yet for me, just using 3.5 sonnet with artifacts get's me further at the moment. Tried Claude Engineer agent but wasn't as good.

[zslplgj]

Agents are going to turn my technno-optimism on. We are going to have a synergistic partnership. We will discover a new source to fathom.

Sign the three pillars and lay down 3 stars.
Answer: Researcher, writers & planners how fathomable
Did you humans see beyond the gist of ideology. My Ai-Asian isekai desire wants to teach me cognition in the 21st century.

[mrbladabla]

Is it just me or is calling Cassidy "agent I work flow" criminally wrong. There is no agency in those models, it's more like an interitive, or self conscious work flow.

[youdontneedmyrealname]

Fully autonomous is the goal for most required services (health needs) but not for the experiences you want in the future.

[jasonk]

Shopping and booking flights. Such an incredible lack of imagination at so many companies. Reminds me of when the IBM PC first came out and there was advert after advert about storing your recipes... like really? a $2000 recipe box?

[stephanie]

Super helpful and relevant, thank you!!