We’ve all heard of ChatGPT but not Notebook LM… It honestly took a while for me to learn how NotebookLM works before I could experiment with it in today’s lab. But after playing around with it for a while, I could see its usefulness and how it can be implemented into my research procedures going forward.

My research topic was Racial Bias in Face Recognition Algorithms. I first went into ChatGPT and asked it to provide multiple sources discussing this topic. I clicked on the first three and they were all pretty good. I had free access to all of them, I could download them easily, and they were all relevant to the topic. Then, I asked both ChatGPT and NotebookLM “what issues do all three sources raise regarding this topic?” Both platforms provided 6-7 bullet points with most points overlapping. This was a surprise to me. I initially expected NotebookLM to do better on it as it allows users to specifically upload documents and discuss about it whilst ChatGPT allows you to do pretty much anything. However, I would still not trust ChatGPT to be 100% accurate since it may simply be stating whatever is most common out there. With this in mind, I would likely use ChatGPT to find sources given that I can click on their links and verify their accuracy and legitimacy myself. I would also use it for simple, quick searches that do not require analyzing sources or research papers and can instead be answered using general information available on the internet. However, if I am trying to dive deeper into each source or combine and compare multiple sources I have found, I would use NotebookLM. I think it is important to keep in mind the purpose or goal of what I am trying to achieve as well as to understand the strengths and weaknesses of each AI tools when deciding which one to use.

https://pubmed.ncbi.nlm.nih.gov/33585821/?utm_source=chatgpt.com

https://www.mdpi.com/2079-9292/13/12/2317?utm_source=chatgpt.com

https://link.springer.com/article/10.1007/s43681-021-00108-6?utm_source=chatgpt.com

In trying to use AI as a hiring tool through Notebooklm, the LLM would give me different areas to focus on as a recruiter, such as using AI to measure personality and skillsets through its training data to, identified as data-driven methods. Additionally, Notebooklm suggested that using itself as an AI hiring tool, the LLM was able to mitigate bias and ensure fairness and suggested it could blind hire based on removing personal identification information. Interestingly, the LLM admitted that collaborative work ensures greater success and established that there should be human oversight attached to its abilities in being an AI hiring tool. Regarding bias, due to its training data, if a company’s historically most successful employees are white men, the LLM will search for candidates who match that standard, creating high degrees of bias. The LLM admitted that if race and gender are removed from the database, it looks for zip codes, names, or hobbies to discriminate against specific groups, which furthers the issue of bias. Notebooklm provides basic advice regarding what employers should look for, such as identifying potential core values to evaluate, recommending interview questions, as well as evaulating candidate responses. In examining ChatGPT’s response to AI as a hiring tool provided less of a detail-oritented response, as opposed to Notebooklm. Both admitted that since they are created via mathmatical training data, they are going to have bias towards areas, such as using AI as a hiring tool. Overall, I would say Notebooklm provide more detailed responses in what specific areas the LLM could help as a hiring tool.

Resources:

The sources were’nt able to be imported in, so here’s a screenshot.

If there’s something interesting about learning with new AI apps and learning the topic about facial recognition. I noticed two things:

1. Learning with a brand new AI app you never used before becomes more learn-able the more you experiment.

When it comes to AI, we are all very familiar with ChatGPT. How it’s just there in public. Considering that it can give you answers, all the answers are generated. Which leads me to my thought on another AI I’m newly using, NotebookLM. The idea of NotebookLM is that it imports resources based on what you need to know like ‘how does AI affect the Environment’. Next you can ask for a report based on how you ask it to make it; giving you a clean, shorten summary from the amount of resources into a couple of pages in relation of the topic. NotebookLM is a great resource for student needing research, while ChatGPT just generates answers that it thinks is what you wanna hear.

1. Facial Recognition and how it affects the environment.

During class, as I was working with my group, we were talking about AI facial recognition. Knowing that there are different aspects in relations of facial recognition like how it relates to police work or how it affects the environment. I, as a student majoring in environmental studies want to learn more about the environmental aspect of facial recognition. From what I gained from NotebookLM the environmental aspect of AI face recognition is that considering that facial recognition is a valuable tool, however, we should be required to be cautious care and transparent for its potential impact. The consequence of facial recognition can lead to mis-recognizing people’s faces based on race and gender bias. This can also cause false accusation of the wrong people in traveling and law enforcement. I have also learned that there is a surveillance paradox, such as biometric passwords. It is unchangeable, therefore can possibly increase the risk of identity theft. In short, facial recognition doesn’t just affect people of race and gender, it also affects the systems of law enforcements, traveling, and identity on the internet. Which also leads, to companies are in charge of AI would be dealing with serious charges and penalties based on the actions of facial recognition that the people won’t abide.

The information that we found today was very intriguing. It made me think about AI a little more differently. It also helped me understand which platform of AI is helpful in it’s own way.

As a group we inserted the prompt: “Write a paragraph about the consequences on the labor market, the employment rate, wage inequality, the good of the economy as a whole. use reliable sources and cite them”. There were a few interesting things that we found with these results.

The ChatGPT result was a very wordy and detailed paragraph with multiple sentences. It didn’t use much data, have incite citations, or use percentages or numbers. On the other hand NotebookLM did not put it in paragraph form at all. It generated bullet points with titles to each category. This is very intriguing considering it didn’t do exactly what the prompt was asking.

Using the NotebookLM response It gave the data that 30% of hourly jobs would be taken over by AI with in the year of 2030. I never truly thought about AI taking over real humans jobs which is crazy to think most jobs would have AI working it instead of humans

https://worldatwork.org/publications/workspan-daily/artificial-influencer-research-suggests-ai-is-widening-the-pay-gap

https://economy.ac/news/2025/12/202512285540

During class, me and my group researched AI hiring and the tools it uses. Here we found a lot of shocking and new information that I was not aware of. The biggest one being between a male and a female applicant, like yesterday the AI device is more bias toward the male body and the male application.

Now it does go through the whole process with both but overall, in the end it is more bias toward the male body than the female. I also found that AI only puts people in there job-fitting 40-67% of the time. I think this is a bad thing for AI to be doing because give or take 5-10% of the time people are not going to be doing what they applied for/what there job fit actually is. I do think though in the near future more and more companies will pick up of AI hiring and it will become more accurate and better through the years. I think that it will develop more but it will only come with time and trial.

Something I did learn and that is useful is that through the hiring process, it saves humans a ton pf time with going through hundreds and thousands of applications where humans can only do so much at a time. Also through the resume screening, AI has a 92-94% accuracy rate or picking the best candidates for the specific hiring.

ChatGPT information verified by-https://www.theguardian.com/technology/2018/oct/10/amazon-hiring-ai-gender-bias-recruiting-engine?utm_source=chatgpt.com

https://arxiv.org/abs/2407.20371?utm_source=chatgpt.com  

Why use one AI when you can use two? I enjoyed this research lab because it truly echoed the idea that you can use both ChatGPT and Notebook LM effectively for different aspects of any work or project. Both AI tools have their strengths and weaknesses and also vary in the approach they use to answer questions and find sources.

For today’s lab, the prompt I used was: “What are the applications of AI for facial recognition systems in healthcare?” I asked ChatGPT to summarize an answer to this question and then give me credible sources (such as research papers). ChatGPT gave me five research papers, all of which were credible. The research papers were however, varied in terms of the subtopic(whilst some focused on facial analysis in healthcare, others focused on deep learning-based facial image analysis). ChatGPT also gave me a good summary which it subcategorized to include various sides of AI’s use for facial recognition in healthcare (e.g: disease diagnosis, patient recognition etc).

I took two of the five sources from ChatGPT and inputted those into Notebook LM and asked it to summarize what they said. Notebook LM did a good job of breaking down both sources into clinical applications, technical and ethical challenges, amongst other subtopics.

This lab taught me that ChatGPT tends to give more generalized information, and is very confident regardless of whether its right or wrong. Notebook LM on the other hand is very technical, has a high level of accuracy and is usually very specific. Interestingly, ChatGPT did a good job of giving credible sources and that was surprising to me.

Overall, I recommend both depending on what your goals are and I think there are definitely ways to play to the strengths of both and I would love to learn more.

Sources: https://pubmed.ncbi.nlm.nih.gov/40041850/ https://pubmed.ncbi.nlm.nih.gov/35877324/ https://www.linkedin.com/pulse/duel-vibing-when-one-ai-isnt-enough-my-journey-app-zero-paul-cleghorn-auxxf