This is a weekly update for anyone who wants to use science for personal, rather than professional reasons.

This week we show an example of using the new AI Chat large language models (“LLM”).

At this point, we’ve all heard of ChatGPT and its amazing — shocking — ability to give coherent and human-like answers to virtually any question. For many of my daily tasks, I’m finding Bing Chat to be even better — to the point where I now rarely use other web searches. Although ChatGPT seems more useful at generating long essay-style prose, Bing is better for Personal Science because it gives references to whatever it returns.

Unlike ChatGPT, which admits its answers are based on years-old data, Bing can answer current questions too. Here’s a sample interaction.

Testing Bing Chat on Blood Test Results

I recently received new blood tests from SiPhox. (You can read my extensive review here). It’s easy to download the results as a CSV file, which I then copy-pasted as-is into Bing Chat and began this conversation. (My prompts are in bold face, and Bing replies are in indented quotes):

Please look at my two most recent blood test results. Point out any anomalies and make actionable suggestions. Of course I understand that I should talk to a doctor for specific advice, but I’d like to hear your detailed overall assessment first.

Bing:

One value that stands out is your Homocysteine (HCY) level which was 4.65 (umol/L) on Nov 10, 2022 but increased to 16.15 (umol/L) on Mar 06, 2023. The normal range for HCY is between 5 - 15 (umol/L). Your Ferritin levels also appear to be low at 7.7 and 14.8 (ng/mL) on Nov 10, 2022 and Mar 06, 2023 respectively, with the normal range being between 20 -150 (ng/mL).
Your High-Sensitivity CRP levels are within the normal range of less than 3 mg/L . However, it’s important to note that reference ranges may vary between labs.
Most of your other test results appear to be within their respective normal ranges.

(Bing Chat gave me many more details, which I’ve omitted here)

Are there possible genetic causes for my high homocysteine levels? What specific variants should I look for in my 23andme results to see which genetic factors apply to me?

Yes, there are genetic factors that can contribute to high homocysteine levels. One such genetic factor is a mutation in the MTHFR gene The MTHFR gene provides instructions for making an enzyme called methylenetetrahydrofolate reductase. This enzyme plays a role in processing amino acids, which are the building blocks of proteins.
Inherited mutations in the MTHFR gene can lead to an enzyme that is not optimally active and may result in elevated homocysteine levels . The most researched MTHFR variants are MTHFR C677T (rs1801133) and MTHFR A1298C (rs1801131). You can look for these variants in your 23andMe raw data by going to the raw data search field and finding the relevant rs ID of the MTHFR gene you’d like to evaluate.

I have TT at rs1801131 and AG at rs1801133.

That means you have two copies of the A1298C mutation and moderately reduced MTHFR enzyme activity. This may affect how your body metabolizes folate and homocysteine. You may benefit from taking supplements of methylfolate or other active forms of folate that can bypass the MTHFR enzyme.
You have one copy of the C677T mutation and reduced MTHFR enzyme activity by about 35%. This may also affect how your body metabolizes folate and homocysteine. You may benefit from taking supplements of methylfolate or other active forms of folate that can bypass the MTHFR enzyme.

I asked about specific foods I should consider to improve my homocysteine levels and it gave the usual recommendations (more leafy greens, legumes, whole grains, etc.). But when I asked it what to avoid, it suggested an item I hadn’t considered:

According to a study, caffeine and coffee treatments increased fasting homocysteine by 5% and 11%, respectively, compared with placebo . Caffeine treatment had a much weaker acute effect on homocysteine (4%; P = 0.09). Effects of caffeine were stronger in women than in men .

The rest of the interaction continues like this, with Bing Chat giving useful answers that would have otherwise taken me an afternoon of searching through Google or Pubmed.

From simple news reading (I can ask it to find and summarize “interesting” discoveries from the past week) to analyzing my self-collected data, Bing is becoming an essential tool for everything I do on the web. It’s built into the excellent, cross-platform Microsoft Edge browser, which I highly recommend as the best way to experience the new power of large language models.

OpenAI: “albert einstein solving a rubic's cube in the style of diego rivera in the style of Bauhaus”

Useful LLM resources

We’ve seen an explosion of think-pieces about ChatGPT and the consequences of these AI models, but here are two specific ones I think are useful for Personal Scientists:

Jason Crawford asks “Can submarines swim?” with a good easy-to-understand explanation of GPT technology. Conclusion: recognize that it’s a tool, with advantages and disadvantages over “thinking” similar to the way submarines are different from swimming.
Stephen Wolfram explains how ChatGPT works with lengthy, step-by-step diagrams. It took me a couple of hours to read – it’s very dense and technical – but the time was well-spent. The conclusion, how LLMs point the way to a better understanding of human language and thinking, is especially provocative.

Some other tools and ideas I’m finding useful

I like the Copyleaks AI Content Detector, which quickly evaluates text to see if it was generated by an LLM. When I copy/pasted the text from an odd, but lengthy comment submitted to one of my ResearchHub papers, it gave an immediate red flag.
Wharton Professor Ethan Mollick’s Substack One Useful Thing is the best resource for up-to-date ideas for how to integrate the technology into your workflow. See his tips for how to use Bing Chat (I like the suggestion to ask it to return an answer in the style of a particular person).
We mentioned Elicit.org and other AI topics in Personal Science Week - 02 Mar 2023.

Finally, before you go too crazy with excitement, it’s a good idea to remember what the technology can’t do. Long-time AI pioneer Rodney Brooks AI Scorecard 2023 and the associated discussion on Hacker News are a good summary. “Nothing is ever as good as it seems at first, or as bad”, he concludes.

What other resources do you recommend? Let us know in the comments.

Leave a comment

Personal Science

Personal Science Week - 230316 AI Chat