8 Comments

I don't think AI getting a higher score on a medical test tells us much, if anything. Answering some questions sets a minimum for familiarity. It is not a measure of how well one can diagnose and treat.

While AI as an alternative may be an obvious possibility, we also need to think about how it can improve outcomes. After the first computer beat a chess master, it was still the case that the combination of chess master and computer was significantly better than the computer. I want a doctor who is willing to use AI to make the best possible decisions.

Expand full comment

Yup - this seems to me to be similar to a database application managing to pull up the correct database record when given an identifier. It's a more complicated for these as the LLM has to figure out the question's mapping to its internal training data, but it's still basically a glorified google search + reformatting data into an "answer".

It's still going to put lots of people out of work as a variant of this is what a lot of people do in their jobs, but the AI isn't a "doctor" just because it managed to pass - or even ace - a written exam.

Expand full comment

We should ask how AI can make a bad doctor better. Under PCP care, patients are not getting the standard of care for basic things like high blood pressure and diabetes. These drive large costs. Using AI to augment the care delivery system could help patients ask better questions and help weak doctors improve outcomes.

Expand full comment

https://www.aaronrenn.com/p/betrayal

Aaron used a voice clone to create a podcast from newsletter.

AI for all translations remains a big useful set of not yet user friendly ai tools.

Expand full comment

Before we get AGI, we’ll have talking aiBot agents willing and able to do anything digitally that humans can do with a computer.

Including meetings, reporting, and report summaries.

The big productivity will come thru even more reduction of middle management jobs, along with the top frontline workers augmenting their work with an aiBot that accepts from the human and coordinates many other specialized bots. Like for each MS Office product, Word, Excel, PowerPoint, there will be a special expert aiBot one can tell what is wanted, or some other bot can tell what is wanted.

Expand full comment

Does anyone know a good way of achieving diarisation (identifying speakers) for YouTube transcripts? I keep running into stumbling blocks using Whisper AI. An alternative or a really, really good tutorial on WAI would be appreciated, thank you.

Expand full comment

"I think they are under-estimating the ability of LLMs to extract information by conversing with patients, not just by reading images."

You might be right but I think I can say with equal certainty you might be underestimating how much better AI is than humans at interpreting images.

Expand full comment

Regarding “today’s ai is the worst you’ll ever use”: while pithy, it only holds if there are continued improvements in ai tech. Mollick’s thought experiment asserts, though, that if we get no further improvement in ai tech, what we have today will still revolutionize the way we work and entertain ourselves, once these technologies are fully integrated into our workflows.

Expand full comment