Week of December 23, 2024
Having two kids home from college has elevated our household conversations. Comments about Michel Foucault, Umberto Eco's semiotics, and applied mathematics now pepper our dinner table discussions. While it is satisfying to watch our boys embrace college (and I am sure I only hear the parent-friendly parts), I now (covertly) use ChatGPT to comprehend their conversations. It feels like just yesterday, I could use Google (for words like 'rizz' and 'ratioed') to understand my children.
I will be traveling next weekend, so there will be no update until January 6. I wish you a peaceful holiday week and a Happy New Year.
(I added a few extra articles to tide you over through holiday travel or downtime.)
---
Listen to a Google Notebook LM A.I.-generated newsletter podcast.
Our AI hosts did well this week, though the last 6-8 minutes were extraneous - too much "big picture" extrapolation. The algorithm is improving, but I need more control over the amount of podcast-ending pontification.
https://drive.google.com/file/d/1oAE0Cg75Zn0deDKG5Ly43ayiEJhopd3s/view
About NotebookLM: https://blog.google/technology/ai/notebooklm-audio-overviews/
------
Science and Technology Trends
Mark Cuban has been engaging in a series of conversations on social media. While I'm sure he has an agenda (i.e., get more Cost Plus Drugs customers), he offers a provocative combination of insights on the U.S. Healthcare system, drug pricing, politics, crypto, and the potential of AI for improving medically high-risk patient populations. Also, listen to his discussion on the power of reverse engineering social media algorithms toward the end of the Wired interview.
(My favorite) 12/4/24 Wired Interview with A conversation with Mark Cuban - "How Much Fucking Money Do I Need?"
https://www.wired.com/video/watch/battling-big-pharma-a-conversation-with-mark-cuban
and
(Most in-depth on healthcare) 12/20/24 Mark Cuban: A Master Disrupter for American Healthcare
https://erictopol.substack.com/p/mark-cuban-a-master-disrupter-for
and
(Most entertaining) 12/8/24 Jon Stewart's Weekly Show - Mark Cuban Talks 2024 Politics, Healthcare, Crypto and AI
https://www.youtube.com/watch?v=bnhxbZlvMRQ
The Journal of the American Medical Association's most-read articles of 2024 is a biopsy of what U.S. clinicians are thinking about. The most common topics included GLP-1s (Ozempic), obesity, reproductive rights, treatments for hair loss, breast cancer, teen mental illness, and the management of gallstone pancreatitis.
https://jamanetwork.com/pages/2024-most-viewed-articles
Here is one last article to remind us to exercise (since resolutions are a thing this time of year). "In this population-based cross-sectional study of septuagenarians, physical inactivity, obesity, diabetes, and stroke/TIA were independently associated with [older-appearing brains on MRI]. [...] Regular physical activity moderated the relationship, yielding [younger appearing brains] in individuals [even] with obesity who were physically active." These data are part of a larger corpus of studies demonstrating how exercise improves quality of life, quantity of life, and cognitive functioning as we age.
https://alz-journals.onlinelibrary.wiley.com/doi/full/10.1002/alz.14435
Here is an AI-generated summary of the research:
https://claude.site/artifacts/9c7a7684-391a-42e6-b9eb-88b993e4345b
Anti- Anti-Science Articles of Note
A key study "that stoked enthusiasm for the now-disproven idea that a cheap malaria drug (hydroxychloroquine) can treat COVID-19 has been retracted — more than four-and-a-half years after it was published." This topic partly inspired the creation of this newsletter - how to think about and understand medical data. The problem of parsing any given study (much less an entire body of research) for flaws and then figuring out if, how, where, and when to act on the data is exceedingly challenging. The "I do my own research" crowd has amplified the noise. This study, its broader impact on consumer understanding of medical data, and the French physician who published it made my job as a physician harder.
https://www.nature.com/articles/d41586-024-04014-9
and
https://bioethics.com/archives/95165
and the retraction letter is at the bottom of the screen
https://www.sciencedirect.com/science/article/pii/S0924857920300996?via%3Dihub
Living with AI.
This story exemplifies the tensions AI may exacerbate. Harper's published a fascinating story about how Spotify works (in recent years). Spotify contracts with high-volume music producers to generate generic music for specific popular playlists, for which they pay lower royalties (as compared to the musicians releasing original tracks). While this "muzak" is currently made by sub-contracted humans, AI-generated genre-themed music is coming. Per the article, some of the more popular Spotify-endorsed (and preferentially displayed) playlists may contain a large percentage of paid-for music attributed to fake artists from contracted producers.
https://harpers.org/archive/2025/01/the-ghosts-in-the-machine-liz-pelly-spotify-musicians/
Harvard researchers recently tested ChatGPT o1-Preview in multi-step reasoning in medical tasks. "We performed a series of experiments to evaluate the o1-preview abilities across several medical reasoning domains: differential diagnosis generation, presentation of reasoning, probabilistic reasoning, and management reasoning. Physician experts assessed the quality of LLM outputs with validated psychometrics. Across this diverse set of tasks, we compared the performance of o1-preview to the responses of hundreds of physicians and prior LLMs." ChatGPT o1-preview did remarkably well across numerous metrics evaluated. The question is, when will AI be the front-line screening tool for healthcare? (See 1-800-ChatGPT below and ponder your health insurance company employing this tool for 1st-touch interactions.)
The paper (not yet peer-reviewed): https://arxiv.org/pdf/2412.10849
And here is a good summary from Twitter:
https://x.com/adamrodmanmd/status/1869023056917864640
Or you can read the Claude summary written in the "Adam's amused observational essayist" style:
https://claude.site/artifacts/de80b87c-238e-4dbe-91fd-3740edaffcdd
Or you can listen to Speechify AI reader read the Claude summary:
https://speechify.page.link/zKaWcZRVh2hQ2NcU8
You must choose some means to ingest this information. AI cannot yet insert the knowledge into your brain (Matrix-style,i.e., "I know Kung Fu" https://www.youtube.com/watch?v=6vMO3XmNXe4).
OpenAI announced new features or products for 12 days, ending last week. Some releases (like Sora - the text-to-video LLM) are worth reviewing. Day 10's release, my favorite (along with several loyal readers), introduced 1800-ChatGPT, allowing individuals to call the 800 number and connect with ChatGPT in voice mode over any telephone. Imagine having your techno-limited family members do this.
https://openai.com/12-days/
(P.S. AI or not, no 1-800 number will beat 'Callin' Oats' - the number to listen to Hall and Oats music:
https://www.npr.org/2011/12/21/144069758/callin-oates-the-hotline-you-dont-need-but-might-call-anyway )
Google also made a series of significant AI announcements last week (I wonder why?). Upgrades to their Gemini LLM, the GenCast weather prediction model, and the Veo2 video LLM were impressive.
https://blog.google/technology/ai/google-ai-updates-december-2024/
Infographics
My favorite "infographers," the Compound Interest team, offered a 2024 Advent chemistry of Christmas series looking at the chemistry of everything from holly to egg nog.
https://bsky.app/hashtag/ChemAdvent24
I did not know there was a tradition of giving oranges at Christmas, much less that people squeeze said oranges into lit candles, making a flamethrower of sorts (see day 15 https://bsky.app/profile/compoundchem.com/post/3ldeirjftgs2q). I am grateful to the internet for teaching me yet another way to injure myself and potentially lose my possessions.
https://tangofscience.blog/2017/11/29/orange-peel-flamethrower/
Things I learned this week
Thanks to a loyal reader, I learned that pickleball relates to tennis as futsal relates to soccer. The question is WTF if futsal?
https://felixonline.co.uk/articles/2018-05-24-wtf-is-futsal-felix-sport-s-guide-to-little-known-high-velocity-quick-paced-team-orientated-spanish-sport/
and
https://en.wikipedia.org/wiki/Futsal
I learned there is a sizeable Chinese market for (unironic) self-help books based on the perceived wisdom and life lessons of Vladimir Putin.
https://www.new-east-archive.org/articles/show/12775/putins-chinese-literary-cult-self-help-books-ideal-russian-man
English Author Robert Sears offers the ironic version of these books, including (but not limited to), "Vladimir Putin - Life Coach," "The Poetry of Donald Trump," and "Choose Your Own Apocalypse with Kim Jong-un & Friends."
https://www.goodreads.com/author/show/14180034.Rob_Sears
Knowing that humans can still be at the top of the food chain is comforting in a chaotic world. Five years after the initial invasion, Washington state officials believe they have murdered all the murder hornets. We are much better at eradicating multicellular organisms (as compared to viruses).
https://www.nytimes.com/2024/12/18/us/murder-hornet-washington.html?unlocked_article_code=1.jU4.1UX2.Xv3GUAxcDpK_&smid=url-share
AI art of the week
(A visual mashup of topics from the newsletter, now using ChatGPT to summarize the newsletter, suggest prompts, and make the images).
A few interesting notes on this week's image:
- The ChatGPT/DALLE content filters are now more sensitive - it didn't like "murder hornets," but it was OK with "hornet."
- The filters do not like using the name of any political or public figure (I was trying to use the book titles from the Things I learned section).
- When asked to generate an image of itself in the role of a medical professional, ChatGPT envisions a human in a white coat with a hornet head. Read into that what you will.
A wide-angle, comedic bookstore interior. In the center, ChatGPT is dressed like a friendly doctor with a stethoscope, juggling medical charts while holding a giant rotary phone labeled "1-800-ChatGPT." Around ChatGPT, a line of worried customers waits, each with bizarre symptoms (like an enormous red "hornet" sting on their arm). Behind them looms a massive bookshelf filled with parody self-help titles like "Evil Wisdom for New Year's Resolutions," "Fun in the Apocalypse," and "Poetry for Horses," propped up by oversized pickleball paddles as humorous bookmarks.
https://drive.google.com/file/d/1Ung0WjU18idCkio8q4Oz4TExjDobQjsl/view
---
COVID rates are rising - 1 in 64 people are infected as of 12-16-24, up from 1 in 134 on 12/9. Travel safely.
In case you want to read about the prolonged lull (and now abrupt rise) in COVID cases, the Atlantic covers it well, "The confusion about how the virus will behave over the holidays reflects a bigger COVID uncertainty: Even after four straight winter waves, experts [are uncertain if we should] expect them. But, [as one] epidemiologist at Johns Hopkins said, "It would be "very unusual" if a wave didn't happen, given that the virus has generally followed a reliable pattern of peaking in the summer and winter."
https://www.theatlantic.com/health/archive/2024/12/covid-christmas-winter-wave/681133/
The Pandemic Mitigation Collaborative (PMC) website uses wastewater levels to forecast 4-week predictions of COVID rates.
https://pmc19.com/data/
based upon https://biobot.io/data/
Check out the influenza A (possibly avian flu or H5N1) and other wastewater hotspots.
Wastewater Scan offers a multi-organism wastewater dashboard with an excellent visual display of individual treatment plant-level data.
https://data.wastewaterscan.org/
----
Clean hands and sharp minds,
Adam
Comments
Post a Comment