My friend and colleague Manuel Ruiz-Aravena has started a blog on AI. I thought I would share my thoughts as well for my dozens of readers!
I created this website with Claude Code. I mostly ripped off samlearner.com because I appreciate Sam's layout. I adjusted the colors and chose celeste. In the AI optimist's world, this is called "taste" and it's all humans have left to offer. But from what I can tell, most humans (myself included) have pretty basic taste. We all want Scandinavian-style coffee shops with polished concrete floors and big exposed air duct systems in the ceiling.
My point is: when it comes to this website, I am bringing nothing to the table. No HTML skills, no taste, no design. But here it is! Clearly, AI can generate a decent facsimile of a great website.
This is what I've been struggling with. Sometimes I'm convinced there's no need for me at all. Last year, I applied to be an electrician's apprentice.
But lately, I've noticed that I'm not really getting any more productive. If AI can replace me, then why can't I get it to replace me? I try using Claude agentically all the time, but it is not really working for me.
It's not even the "edge cases" it fails on. It's not even "ooh, all the busy work is done, all that's left for me are the really tough cognitive problems." No, I can't get it to do any of my work for me. A website, sure, I don't know anything about that. But when it comes to my day-to-day, it's not speeding me up. Here are the two main categories it seems to break:
Writing from scratch. Having Claude write something from scratch doesn't really work for me. It has this odd tendency to want to make big, bold statements. In science, we usually don't have big, bold conclusions. If I let it control this aspect of my work, it would probably apply for the Nobel Prize based on the amazing revelations we've found in wastewater. If I slightly push back, it goes the opposite direction — suddenly we've made incremental progress on a project nobody cares about. It's a real Jekyll and Hyde situation.
Code. The larger the code request, the worse the code gets. I find it generates these massive "mega-functions," as I'll call them. I like functions. I'm not a trained software engineer, but my preference is for lots of small 'mono-tasker' functions that live in a separate file I usually call useful_functions.R or project_tools.py. When I let Claude cook, it likes to make one mega-function that cleans my data and runs a Stan model all at once. If it breaks, I can never figure out where or why, and it's just kind of useless. Given lots of minor tasks, though, it does a decent job.
Now, Claude as an interactive tool does speed me up.
The most impressed I've been by any AI tool was ChatGPT-4 last year. We were applying for a complex, multidisciplinary grant. NSF wanted everything neatly woven together — everyone's science feeding in everywhere. It's the kind of work that's really hard to organize, especially over Zoom. I fed the grant into ChatGPT-4 and it was able to see and make the connections really quickly. It's sort of the perfect use case: consume a massive amount of poorly written draft text, draw on a general knowledge base, and identify the specific sections where we can link the disciplines together.
Here's how I use AI today. I write an outline by hand. I take a photo of that outline or transcribe it. I say, "Claude, turn this into a draft." I print out the draft, edit it by hand, make the edits in the document, and do a final Claude pass for typos. I think it works great. I get physical printouts, which I like — it lessens screen time, makes it easier for me to focus, and something about paper just feels good. And I get Claude to turn outlines into paragraphs, which I've always struggled with.
I think my work is improving due to AI. The quality of drafts, the code annotation, the figures. It does help with the mundane tasks I struggle to motivate myself to do. So, while I'm deeply paranoid it will replace me and I'll end up homeless, I don't see that happening this year at least. Maybe I'll be wrong!
I took this from the Journal of Animal Ecology Blog, check the original there: animalecologyinfocus.com
This blog post is provided by Dan Crowley and tells the #StoryBehindThePaper for the article "Cohorts of immature Pteropus bats show interannual variation in Hendra virus serology", which was recently published in Journal of Animal Ecology. Every winter, Hendra virus spills over from Australian flying foxes to horses and humans. The authors of this study spent four years tracking juvenile bats to test whether they drive these seasonal outbreaks.
Every November, hundreds of thousands of flying fox bats are born on the east coast of Australia. Like all mammals, we assume these bats are born with their mother's antibodies—proteins that provide critical protection against pathogens. While this has long been an assumption, we really understand very little about how these bats protect themselves against pathogens.
Most mammals receive antibodies from their mother and these wane over the first year of life. As these antibodies wane, the vulnerability to pathogens increases. Until the immune system matures and generates endogenous antibodies, young mammals exist in this window of susceptibility.
For many populations, the sudden pulse of susceptible individuals after a synchronized birth pulse has dramatic impacts on pathogen dynamics. For flying foxes, we were particularly interested in how this influx of susceptible individuals would impact Hendra virus dynamics. Hendra virus is a harmless pathogen for these bats, but it often "spills over" to horses and humans (cross species transmission), where it can be deadly.
We suspected Hendra virus dynamics might be driven, in part, by this influx of susceptible juveniles. Hendra virus spillover events tend to occur in winter. This is also when we predicted these juvenile bats would lose their maternal antibodies. If there was a synchronized waning event, this could introduce sufficient susceptible individuals into the population to cause a spike in transmission.
However, more happens in winter than just waning maternal antibodies. While Hendra virus spillover events occur in winter, they are especially common in winters with insufficient food. It was previously suggested that adult bats are starving and immunocompromised, unable to control Hendra virus replication, and shedding the virus in their urine—which then infects horses.
So, which matters more: the juveniles, or the food shortages? This question had not yet been investigated.
New serological tools and statistical methods let us tackle this question directly. By sampling Australian flying foxes over several birth cohorts, we tracked how antibodies developed in juveniles as they aged, data that's normally out of reach in wild populations.
So, what did we find? We didn't see the clean signal we expected: no consistent maternal antibody waning, and more strikingly, no clear wave of juveniles generating their own antibodies as they became exposed. These results were confusing and unexpected.
Fortunately, we also had data on Bartonella, a bacterial infection transmitted by bat flies. Bartonella became our control, a way to measure what "normal" transmission looks like in these bats. Like Hendra, it requires close contact to spread.
Unlike Hendra, Bartonella was remarkably consistent across cohorts. Every year, juveniles were born uninfected, and by about nine months of age, nearly all had acquired it—at the same rate, year after year. This told us something important: the inconsistency in Hendra wasn't because bats were behaving differently. Contact opportunities were stable. Something else was driving the variation.
So, what does this mean for Hendra virus spillovers? We didn't find a clear window of susceptibility in winter that could be driving spillover events. The lack of synchronized seroconversion suggests juveniles aren't playing a major role in Hendra transmission. Together, these results point us back to the food shortage hypothesis. It appears that adults, not juveniles, are the drivers of Hendra virus shedding. While additional work is needed to establish food shortage as the causal trigger, this study represents one more step towards understanding the mechanisms driving viral shedding in this complex system.
Read the paper: besjournals.onlinelibrary.wiley.com
...