Building an AI Agent

I want to present a scenario of executing a basic Research Task with the help of AI:

Scenario:

Lets assume that there is news agency with a director who has a team of researchers and respective research managers. Team's responsibility is to research on the latest news and provide insights to their viewers.

Let's assume a following research task is given by director to his team:

Task:

Research about the launch and ability of Google's Large Language Model called Gemini and Find out the details on following:

1. Was the launch of Google's Gemini Model a publicity stunt.

2. Is Gemini (latest Model by google) really the best among other proprietary models like ChatGPT

To carry out this task, I have created following roles:

1. Assistant Director will take a client's query and delegate it to their team. He will review the final submission (researched data) and make sure that it has complete answer to the user's query.
2. Research Manager will review the researcher's search and guide him with more inputs to find concrete details about the given topic.
3. Researcher is the one who will carry out research task and report his work back to research manager.

Next, I have used OpenAI Assistant API to create these assistant with respective roles. To carry out the research task, research needs access to google search. I have used serp api to get the relevant websites links for the research. Once researcher gets the links of relevant websites then it requires an ability to scrape those websites for the content. Here I have used scraper api.

And finally I have used Open Source Framework called Autogen to engage all the actors in the conversation.

Let see the query and response:

Query:

message = """
Research about the launch and ability of Google's Large Language Model called Gemini and Find out the details on following:
1. Was the launch of Google's Gemini Model a publicity stunt.
2. Is Gemini (latest Model by google) really the best among other proprietary models like ChatGPT 
"""
user_proxy.initiate_chat(group_chat_manager, message=message)        

Response (from assistant director on my query):

user_proxy (to chat_manager):


Research about the launch and ability of Google's Large Language Model called Gemini and Find out the details on following:
1. Was the launch of Google's Gemini Model a publicity stunt.
2. Is Gemini (latest Model by google) really the best among other proprietary models like ChatGPT 


--------------------------------------------------------------------------------
assistant_director (to chat_manager):

I received your request and will commence with assigning the research tasks to my team. Please bear with me for a moment while I process the required actions with my research manager and the researchers. I will get back to you as soon as I have the necessary information.


--------------------------------------------------------------------------------        

Response (from researcher manager):

research_manager (to chat_manager):

Certainly. Here are two actions that researchers can undertake to find the information needed regarding Google's Large Language Model called Gemini:

1. **Search Scholarly Databases and Google AI Blog**: Have researchers search through authoritative sources such as Google Scholar and academic journal databases for any published papers or reports on Gemini. Additionally, instruct them to thoroughly review the official Google AI Blog and any press releases by Google for official statements or announcements pertaining to the Gemini model.

2. **Analyze Tech Industry Publications and AI Forums**: Researchers should go through reputable technology news websites, AI industry publications, and forums where AI researchers and practitioners discuss and review new developments. They should look for reviews, comparisons, or analyses that feature Gemini and how it is positioned relative to other models like ChatGPT.

Review each piece of information gathered and challenge the researchers to substantiate their findings with credible sources. If the information is not satisfactory, push back and demand more in-depth research.

**Pushback: "No, you have to find the information, try again."**

Next method to try:

3. **Reach out to AI Experts and Academics for Commentary**: If the existing public information is not sufficient, instruct the researchers to seek expert opinions. They can reach out to AI researchers, university professors, or industry analysts known for their work in natural language processing and AI to ask for firsthand insights or commentary on the capabilities of Google's Gemini relative to other models.

Undertake these actions diligently, and do not stop until the necessary information is found. Upon success, we will conclude with "Terminate".


--------------------------------------------------------------------------------        

Then researcher start the research task by using google search and scraping websites

--------------------------------------------------------------------------------

>>>>>>>> EXECUTING FUNCTION google_search...

>>>>>>>> EXECUTING FUNCTION google_search...

>>>>>>>> EXECUTING FUNCTION web_scraping...

>>>>>>>> EXECUTING FUNCTION web_scraping...

>>>>>>>> EXECUTING FUNCTION web_scraping...

>>>>>>>> EXECUTING FUNCTION web_scraping...

>>>>>>>> EXECUTING FUNCTION web_scraping...

>>>>>>>> EXECUTING FUNCTION web_scraping...

>>>>>>>> EXECUTING FUNCTION web_scraping...
Scraping website... https://www.theverge.com/2023/12/6/23990466/google-gemini-llm-ai-model
text length : 10080

>>>>>>>> EXECUTING FUNCTION web_scraping...
Scraping website... https://www.businessinsider.com/google-gemini-openai-chatgpt-review-which-is-better-how-to-2023-12
text length : 25261        

Response (from Researcher): Both Research Manager and Assistant Director were satisfied with researcher's findings. Finally Assistant Director asked to terminate the session if 'I' as a user is convinced with the response.

researcher (to chat_manager):

Based on the information collected, here are the findings:

### 1. Was the launch of Google's Gemini Model a publicity stunt?
The launch of Google's Gemini model does not seem to be a publicity stunt. The information detailed by various sources like The Verge, CNBC, and Wired, explains that Google has launched Gemini with the aim to integrate it into their products including Google Search, Ads, and the Bard chatbot. Gemini is said to have been benchmarked against OpenAI's GPT-4, reportedly outperforming it in the majority of benchmarks. Given the extent of the engineering and the performance capabilities asserted, the launch appears to be a strategic move to compete with other AI models like GPT-4 and to establish Google as a leader in generative AI technologies.

### 2. Is Gemini (the latest Model by Google) really the best among other proprietary models like ChatGPT?
According to a comparative evaluation by Business Insider, Gemini, specifically the Pro version, is touted as being smarter than OpenAI's GPT-4. In various testing scenarios, Gemini demonstrated an edge in areas like synthesizing current events, generating plans, and summarizing articles. However, it did show limitations, generating factual errors and misidentifying features in images. While both Gemini and ChatGPT struggled with questions about future events such as the 2024 US presidential election and displayed their own distinctive tones and styles, Gemini was noted to perform better at web browsing for accurate information and answering questions related to topical events and meal planning. ChatGPT, on the other hand, was praised for a more enjoyable conversational tone.

Given this comparative analysis, while Gemini shows promise and some areas of superiority, it is not definitively the "best" across all metrics given that it also exhibited shortcomings. The notion of "best" could depend on the specific applications and needs of the users.

### References:
- [The Verge: Gemini Launch Overview](https://www.theverge.com/2023/12/6/23990466/google-gemini-llm-ai-model)
- [Business Insider: Gemini versus ChatGPT](https://www.businessinsider.com/google-gemini-openai-chatgpt-review-which-is-better-how-to-2023-12)

These references back up the information provided and support the inference that Google's Gemini is a serious contender in the AI space, though not without competition.


--------------------------------------------------------------------------------
assistant_director (to chat_manager):

TERMINATE


--------------------------------------------------------------------------------        

Summary :

With the setup of few agents (playing different role), we can automate searching of the relevant information and present it in a way we need for our work. Isn't it amazing with AI as a tool to help us with boosting our efficiency and productivity?