Autogen Teachable Agent Can Replace Your Personal Assistant

Autogen Teachable Agent: The Next Step in AI Assistance
In today’s fast-paced world, the demand for efficient and reliable personal assistants has increased significantly. With advancements in AI and machine learning, a novel solution has emerged that could potentially replace your human assistant with an even more versatile and resilient alternative: the Autogen Teachable Agent. Developed by a collaborative effort between the Autogen team and Microsoft, this powerful AI tool is transforming how professionals manage their tasks.
What is an Autogen Teachable Agent?
An Autogen Teachable Agent is a type of artificial intelligence developed to assist users by automating various tasks. Unlike traditional AI, these agents possess the ability to remember and learn from the interactions and instructions given by their users. This capability is anchored on a vector database that allows these agents to store and recall information, thereby continuously improving their efficiency and effectiveness over time.
Core Capabilities of Autogen Teachable Agents
The most notable ability of the Autogen Teachable Agent is its teachability. Users can instruct the agent in specific tasks, and the agent retains this knowledge for future application. This learning process is enabled by teachability classes integrated within the system, which assess and store necessary information from each interaction in a vector database.
- Customizable Memory: Users can adjust the number of memories retrieved, tailoring the agent’s recall abilities to specific needs.
- Composability: These agents are designed to be composable with other capabilities, enhancing their versatility in performing a diverse range of functions.
Real-World Applications and Benefits
Personalized Task Management
The teachable agent can learn your daily routine and preferences, prompting tasks such as going to the gym or following up on important projects. For example, after teaching the agent to inquire about gym attendance, it will automatically ask you each morning, ensuring you stay on track with your fitness goals.
Enhanced Productivity
With its capability to learn and automate tasks, the agent can handle repetitive or time-consuming tasks, allowing you to focus on more critical aspects of your work or life. This could include drafting emails, organizing documents, or even managing your schedule.
Continuous Improvement
Due to its inherent ability to learn from interactions, the agent’s performance improves over time, making it increasingly efficient. The more you teach it, the better it performs, adapting to your evolving needs seamlessly.
How Does It Work?
Memory Storage and Retrieval
Each interaction or ‘memo’ with the agent involves pairing input texts (such as questions or tasks) with output texts (answers or advice). These pairings are stored as vectors within the database, enabling the agent to retrieve relevant information when similar tasks or questions arise in the future.
Teaching New Skills
You can teach the agent by simply having a conversation about what tasks need automatization or specific ways you prefer your tasks handled. Once the agent understands and internalizes this information, it updates its database, making the new skills available for future use.
Unleashing the Potential: Practical Examples
Consider an instance where you’re preparing for an automation project and need to gather precise information consistently. You can teach the agent the specific questions to ask every time you mention a new automation, ensuring all necessary data is collected without fail. This scenario illustrates how tailored and efficient the agent can be, adapting its responses based on the learned behaviors and stored memories.
Final Thoughts
The Autogen Teachable Agent represents a significant leap forward in the realm of personal and professional assistance. By integrating this advanced AI tool into your daily operations, you stand to benefit from a highly personalized, ever-improving assistant that not only understands your needs but also anticipates them.
For those interested in pioneering the future of personal assistance, the Autogen Teachable Agent offers a promising and potent solution. Embrace this innovative technology and watch as it transforms mundane tasks into opportunities for enhanced productivity and efficiency.
If you have found this insight into Autogen Teachable Agents intriguing and wish to delve deeper into how AI can assist in automating your life, do not hesitate to engage through comments or reach out for more information. Let’s navigate this exciting frontier together and unlock new potentials in personal and professional productivity.
[h3]Watch this video for the full details:[/h3]
Hey👋 if You Enjoyed This Video and are interested in more content about AI, Agents and Business Automation — please Subscribe🙂
———
In this video, we explore the Autogen Teachable agents.
🎞️Chapters
00:00 – Autogen Teachable Agent Can Replace Your Personal Assistant
▶️Other Videos You Might Like
Autogen Just Got Better – Controlling The Agents With Validation and State Transitions:
https://youtu.be/EJsQ-XGKkPo
Self Reflecting Agents – Autogen Producing Amazing Production Level Results (Real Business Use Case)
https://youtu.be/NpvMddMwX9Q
Autogen Agents with RAG – Upgrading Agents with a Knowledge Base
https://youtu.be/4EJ-1IJoeMc
💡Mentioned in the Video
https://microsoft.github.io/autogen/blog/2023/10/26/TeachableAgent/
✅Come say ‘Hi’ over Linkedin:
https://www.linkedin.com/in/yaronbeen/
✅Book a Consultation:
https://ecomxf.com/book-a-call/
✅My Website:
https://ecomxf.com/
#️#️#️
#automation #aiagents #autogen #teachableagent
#openai #microsoftautogen
[h3]Transcript[/h3]
autogen teachable agent this is a very powerful agent that was created by the autogen team the Microsoft team and I’m trying to leverage it in order to help me in automating stuff and basically the core abilities of the teach agent are basically the fact that you can teach him skills and he can remember the skills that you taught team so basically it’s an agent that can has memory and can keep on improving and kind of self-optimizing I still trying to figure out what is the best way to use it and leverages but I wanted to share with you a few of my learnings and I think it’s very powerful and it has a ton of potential let’s cover first what exactly the teachability agent and as I said it’s an autogen it’s a type of an autogen agent this is the code let me make it bigger so you guys will be able to see it as well and basically I’m going to read what they’re saying over here teachability uses a vector database to give an agent the ability to remember user teachings where the user is any Coler human or not sending messages to a teachable agent teachability is designed to be composable with other agent capabilities to make any conversible agent teachables and basically um whenever you add teachability to an agent the following are modified the agent system message is a pendant with a note about the agent new ability if you guys remember or don’t remember system agent is basically the description that we provide to the A agent so the agent knows what is his roles abilities and background and a hook is added to the agent processed last received messages hooka method and the hook potentially modifies the last of theed messages to include earlier teachings related to the message added teaching do not propagate into the stored message history if new user teachings are detected they are not they are added to new members in the vector database now let’s move on let me see if there’s anything crucial over here so over here they are covering the fact that it is customizable to a certain degree we can change the number of memos to retrieve and you will understand this a bit later um but let’s move on to the most important part here is basically we update the system message with this function you’ve been given the the special ability to remember user teachings from prior conversation and here is another function that basically decides whether to store something from one user comment in a database and the way that it analyzes and decides if to add a comment to the B to the databas is does any part of the text ask the you the agent to perform a task or solve a problem answer with just one word yes or no if yes if any of the text that we sent to the agent is basically requiring him to task to solve a task or solve a problem so what it does is briefly copy any device from text that may be useful for a similar but different task in the future but if no advice is present just respond with none now briefly copy just the task from the text then stop don’t solve it and include don’t include any device now it asks it to General generalize the task so summarize very briefly in general terms the type of task described in a text leave out details that might not appear in similar problems so basically whenever we ask an agent to solve um a task it figures out the teachable the teachability class is figuring out if this is a task or not if it is it takes the task it generalize the task and it’s going to store it in the database next it remembers the task now um this is a a function that basically Tex tests if the information what is the information to be learned does the text contain information that could be committed to memory answer we just yes or no if yes so imagine that the user forgot this information in text how would they ask you for this information include no include no other text in your response now it copies the information from the text that should be committed to the memory add no explanation so basically this way we are building the database with a advice about how to solve the task and the generalized tasks here this function is basically trying to figure out whether to retrieve memers from the database and add them to the chat context so basically trying to determine if there’s anything in the database that we already added in the past because as I said the teachable agent has memory and he remembers what we taught him so this function figures out if there is anything relevant in the database to the new question so you see later that I have taught my agent my teachable agent that any time I say good morning he needs to ask me if I went to the gym yesterday and did I complete the sales funnel that I was supposed to finish anytime I just tell him good morning he’s going to ask me did you go to the gym yesterday no matter what and this is kind of it I wouldn’t say it’s fine tuning but it’s similar idea somewhat similar idea so whenever I just I taught him whenever he sees good morning he needs to ask me this question did you go to the gym yesterday you were see later now let’s move on um I don’t want to bore you with too much technical stuff let me see if there’s anything relevant here and let’s [Music] see okay so let’s see here provides so the next class is memo store so provides memory storage and retrieval for a teachable agent using a vector database each database entry called a memo so memo is just a database entry is a pair of strings an input an input text and an output text the input text might be a question or a task to perform and the output text might be the answer to the question or advice on how to perform the task now the vector H this these pairings are going to be embedded in vectors and this is a very this is important just to understand how exactly the the database is structured let me see I think that’s it from the important stuff and here there are just a few examples that I want to share with you so these are examples just to understand how it works so as a few arbitrary examples to the vector database just to make sure to make retrieval less trivial so the first example is when I say paper I mean research papers which are typically PDFs basically this is an example of a task that we can give the teachable agent and he will recall it he will store it in in the memory so as you can see here yes yes means this is going to be stored next example please verify that each paper you listed actually uses the L chain here it’s no because this isn’t a general this isn’t a general requirement or a task it’s just a specific spe ific task and it is not general and this is why it is labeled as a no now tell GPT the output should still be latex code again this is no this isn’t something general so we don’t want to teach the agent anything that isn’t General and is relevant for all of our use cases we if it’s just like a request that we’re making the teachable agent for one instance so it knows based on these examples most of the time not not most of the time even but the goal is it for for it to understand that if we have just one small requirement that is only one instance we don’t want it to understand and learn this behavior and use it every time afterwards hint convert PDF to text and then answer questions based on them label yes so this is something that we want to use across the board all the time and this is why it is labeled as a yes to create a good power Point include enough content to make it interesting again this is a yes because it’s a general requirement from now on this is something that a skills that we would like to teach the agent in opposing to this which is just like a one-time request from the agent and there are a few examples here double check to be sure that columns in table correspond to what asked before this is a yes compare the H index of Daniel Weld and this is not because it’s just a specific request okay great so we covered the teachability class now let’s move on to my example this is my example okay so we show the example now my test this is the code I will remove the open AI API key afterwards so don’t worry about that now the the code is very straightforward you can just copy it from the autogen in the G in their gab repository they have a notebook and over here you see higher on how I use this is the chat initiation this is the cond environment and this is the conversation so teachable agent to user hion how are you it remember it not even remembers because this is just like the initiation message now I said good morning it answered me good morning yon did you upload the funnel as I told you before before I started the video a few days ago uh before I started recording the video I a few days ago I started testing this teachable agent and I asked him that whenever I say good morning I told him that he should ask me about the funnel so basically nagged me until I launched the funnel so I said not yet I also asked him to every time we start the conversation ask me regarding if I went to the gym yesterday or not so I said not yet and as you can see here I understand don’t forget to do it later by the way did you go to the gym yesterday not yet then he said all right it’s important to keep up with your routine routine have you have we drafted any emails today this is another thing I asked it to assist me with drafting emails and I you can see every the conversation started with him asking me regarding the funnel and regarding going to the gym and regarding drafting emails so this is all stuff that I already taught the agent that I would like to ask me then again not yet by the way I did go to the gym and I did move on towards the funel but just for the sake of the example I said no okay we need to work on that later it’s important to stay on top of our tasks now I told him agree can you help me write an email and whenever I thought it that whenever I ask him to write an email he should respond with asking me regarding the recipient name the position the subject of the email and um ask me when was the last time I spoke with the with the person and as you can see it did so and then I decided I changed my mind let’s teach you more cool stuff question related can you suggest a list of questions to ask a client before creating an automation Gathering data about the Automation and the use case so basically I’m first before I even prime him before I teach him to help me with coming up with questions about automations I first want to see like What He suggests so certainly here’s a list of question one might ask a client before starting an automation project so first can you describe the process that you want to automate in detail two what what are the expected results of the automation three are there any specific tools or platforms you prefer for the automation four how often will the automated task be performed five who are the key stakeholders of the process six what are the current pain point of the process how would you measure the success of the automation any specific timelines or deadlines any security or compliance considerations and do we have any third party services or application that we need to take in mind so so these are the questions that they suggested and then I said awesome now whenever I tell you I need to create a new automation you’ll need to ask me these questions and he said understood from now on whenever you mention that you need to create an automation I’ll use these questions to help gather necessary information cool awesome so basically he was taught now that when ever I ask him about automation he needs to respond with these questions now I need to close the conversation by writing exit whenever you write exit you basically you update the database so I updated the database the agent is now supposed to learn this new skill now let’s launch again and see if it worked basically I want to see I asked him about the Automation and he needs to give me the these questions let’s see if it’s going to write all the questions he’s going to seek in the database if there’s anything related very interesting to see if it will work okay sure I can help you with that based on our previous conversation here are the steps we should follow some questions we may need to ask I establish have okay so it seems like and PR was specific test nice so it’s interesting because it combined the questions I asked him regarding the email like I T him I told the agent that whenever I want to write an email he needs to ask me regarding who is the recipient what is the subject and when did we last connect so this is one thing that I taught him and you guys saw that that I also taught him regarding the question of automations and he kind of combined them both which isn’t bad it really surprised me it isn’t bad but it probably needs some fine tuning and it’s up to me obviously anything is up to me but obviously it’s up to me or up to us that we need to probably provide better input make it clearer within the prompt what we need to get and also make it clearer in the work flow to so to let the teachable agent understand exactly when we want us to fetch a specific um Memo from the database but just for the sake of example you can see here I’m going to write good morning and he’s supposed to respond with question regarding my funnel did you upload the funnel and by the way did you go to the gym yesterday so this was a simple cue anytime I write anytime I tell good morning ask me about the fun and did I go to the gym yesterday so he knows that he he needs to provide me this uh answer so probably what I would need to do now is just create a better pair so whenever I write the word automation return 10 question about building an automation process whenever I write the word automation you need to respond with 10 questions that I need to ask my client about his Automation and this is probably the way to go like making it very very clear at the end of the day they do have somewhat of a brain they’re using GPT 4 in this instance but uh um we still need to guide the teachable agent in a very direct manner I think it has a ton of potential as you can see here these are the 10 questions exit let’s see let’s restart WR Automation and hopefully it will respond with these 10 questions awesome great Automation and it gave me back these 10 questions and just to clarify if you guys are like um if it’s not obvious so it’s not as if this is the same conversation the we close the conversation you can assume that even if in 10 days time I come to this teachable agent and I write the world automation it’s still going to spit out these question and this is why this is so powerful it’s a teach agent it’s a pretty it’s literally a teachable agent whenever you teach this agent stuff or tasks H he can afterwards learn um Implement what you taught him so I’m haven’t figured out yet how I can use this in production level but I believe this is could be very powerful and also streamline many processes my goal is basically to have one at least for the beginning have one teachable agent which will be kind of my virtual assistant or personal assistant and try to teach him a lot of stuff that more related to prioritizing and productivity and like my personal schedule this is like the simple the stuff that is most simple see how that goes and if it works out well obviously scale this build more agents more teachable agents and see where it goes I guess that’s it for today guys I hope you enjoyed this short video about autogen teachable agents if you found this video valuable Please Subscribe like and leave a comment below with the world teachable agent just for the algorithm and I know that also my call to actions they do something for you guys and so teach agent in the comment section if you have any comments or questions obviously I’d be happy to answer and until next time keep on automating