Egoprompting

I read a news article recently about someone suing ChatGPT for defamation. Prior to finding out this news, I never considered asking a Large Language Model (LLM) about myself. This reminded me of egosurfing, which is the act of using your own name as a keyword on a search engine or search engines to see what comes up. (Fun fact: a long time ago, I had to manually submit my domain to Google to get it indexed. Nowadays everything gets indexed unless you explicitly specify no indexing.)

This got me curious as to what the GPT models would return if I asked about myself. I will hereby designate the term "egoprompting" as the act of using your own name in a prompt to see what results a LLM returns.

I will use the gpt-3.5-turbo model because I do not have access to GPT-4. I have a script (ask.py) that takes input from the command line and uses ChatCompletion to fetch a response.

I started with the following.

./ask.py "Can you tell me what you know about Dave Tang?"
# I am an AI language model, and I do not personally know Dave Tang. However,
# there are several people with that name, and I would need more specific
# information to provide accurate details.

Fair enough. I provided some additional information.

./ask.py "Dave Tang is a bioinformatics researcher. Can you tell me more about him?"
# As an AI language model, I do not have access to personal information about
# individuals unless it has been publicly disclosed. However, according to a
# LinkedIn profile under the name Dave Tang, he is a bioinformatics researcher and
# software developer based in Sydney, Australia. He has worked for various
# academic and research institutions, including the Garvan Institute of Medical
# Research, the University of New South Wales, and the University of Sydney. His
# research interests include genomics, transcriptomics, and sequencing
# technologies. Additionally, he is involved in developing computational tools and
# pipelines for analyzing high-throughput genomic data.

I have never worked at Garvan, UNSW, or USYD but I am connected to people working or have worked at those places on LinkedIn. The second part is accurate and was probably based on my LinkedIn profile.

I repeated the question again.

./ask.py "Dave Tang is a bioinformatics researcher. Can you tell me more about him?"
# As an AI language model, I do not have access to personal information like home
# addresses, phone numbers, or private life. However, I can tell you that Dave
# Tang is a bioinformatics researcher who specialises in developing computer-based
# tools for analyzing biological data. He has a Ph.D. in bioinformatics and has
# worked on projects ranging from next-generation sequencing to metagenomics to
# protein structure prediction. He has also published many peer-reviewed articles
# in scientific journals and is highly respected in the bioinformatics community.

Now that's just complete bollocks (apart from the fact that I do have a PhD).

One last time.

./ask.py "Dave Tang is a bioinformatics researcher. Can you tell me more about him?"
# Unfortunately, as an AI language model, I do not have access to personal
# information about individuals unless it is publicly available on the internet.
# However, based on the given information, Dave Tang is likely someone who focuses
# on using computational tools and techniques to analyze biological data and solve
# biological problems. This may involve using programming languages, statistical
# methods, and machine learning algorithms to analyze large datasets related to
# genetics, genomics, protein structure, and other areas of biology.

That's just completely weaseling out.

Summary

Short prompts like the ones I have used in this post are unreliable because of the lack of context. However, I did this on purpose to see what the model would conjure/hallucinate up. From the responses, it is clear that LinkedIn data was scrapped, trained on, and used to generate the response/s. Google Scholar data was also likely used. The rest of the response was just based on words associated with "bioinformatics".

In summary, it is not reliable to use LLMs to find out about yourself (if you are not famous). I would have been impressed if the model returned "Dave Tang is simply the dude with the bioinformatics blog."

2024 September 18 update

I asked ChatGPT "Who is Dave Tang?" and the hallucination is enough to make me blush:

Dave Tang is a bioinformatician and blogger known for his contributions to bioinformatics education and computational biology. He has made a significant impact on the field by providing accessible tutorials and resources for those interested in learning bioinformatics, especially in areas like next-generation sequencing (NGS) and data analysis using R and Python.

Here are some key highlights about Dave Tang:

Educational Focus:
Dave Tang is well-known for his blog, "davetang.org", where he shares detailed tutorials and guides on bioinformatics tools, scripting, and data analysis. His content is especially helpful for beginners in bioinformatics, as it covers a wide range of topics, from Unix basics to advanced topics in RNA-seq and genome analysis.
His blog is a popular resource for students, researchers, and anyone looking to understand how to analyze biological data with tools like R, Python, and various bioinformatics software.

Blog Content:
His blog covers topics such as RNA-seq analysis, variant calling, genome assembly, and visualization of genomic data. Many of his posts are detailed tutorials that include scripts and step-by-step instructions for solving common bioinformatics problems.
In addition to tutorials, Dave also discusses general bioinformatics topics and trends in the field, making his blog a well-rounded educational resource.

Background and Experience:
Dave Tang has a background in molecular biology and bioinformatics. He has used his expertise to build a community of learners around his blog, contributing to the open science movement by making educational resources freely available.

Contributions to the Bioinformatics Community:
Through his tutorials and blog posts, Tang has empowered many early-career bioinformaticians and life scientists to become more proficient with bioinformatics tools and techniques.
His clear, approachable writing style has earned him a reputation as a go-to resource for learning how to use bioinformatics tools effectively.

In summary, Dave Tang is a prominent figure in the bioinformatics community, particularly recognized for his educational outreach and high-quality tutorials aimed at making bioinformatics accessible to a broad audience. His blog remains a valuable resource for learning and applying bioinformatics methods.