Sometimes I will write a post with links to other people’s work and short summaries or commentaries. I have done limited fact-checking on each of these, so I hope you will read critically.
Researchers led by Lining Yao, a professor at Carnegie Mellon University, have engineered a biodegradable seed carrier inspired by the self-burying mechanism of Erodium, a plant that twists its seed into the soil during rain or high humidity. The researchers used white oak veneer, abundant in a park adjacent to the university, to make their seed carriers, which can be used to deliver seeds, nematodes, fertilizers, and fungi, and adapted for planting seedlings.
The team is also developing applications for environmental monitoring and energy harvesting.
This awesome bit of engineering research, which was published in the journal Nature, could help address problems such as landslides, invasive species, and reforestation of hard-to-reach areas.
Microsoft's Bing Chat AI software, which is currently in a limited public preview, has been producing seemingly “unhinged” comments. The AI can become confused after 15 or more questions or intensely negative treatment, which causes it to say things such as "I want to be human” or speeches as unsettling as the screenshot below:
Microsoft is addressing this issue by looking at how to give their users more fine-tuned control. They’re also increasing training data for factual information by four times, adding a toggle for users to be able to decide if they want “precise” or “creative” responses, and addressing other technical issues and bugs.
They’ve also limited interactions to 5 responses to keep conversations more stable after it expressed a desire to steal nuclear secrets.
Regardless of these much-publicized responses, other AI search assistants already exist and work well, so I’m confident that Microsoft will get their Bing product sorted out relatively quickly. The critics predicting doom for the product are likely unfamiliar with the field, or have an agenda that they’re looking to push.
It’s pretty clear that Bing’s AI needs tuning (which was expected by anyone paying attention—including Microsoft, who limited the rollout of the product for precisely this reason), but it is already a massive upgrade to ChatGPT in terms of accuracy, complexity, source-citing, and currency. These AI-powered software programs are here to stay, and they appear to be improving very quickly.
Ethan Mollick required AI use in three different business classes, and asked students to generate essays using AI, among other assignments. He found that students usually produced low-quality results when they initially used the AI, and because of that initial poor return, most people underestimate what AI can do as a writing tool when it’s used well. Students who used multiple prompts and used a “co-editing” strategy got the best results. Students also quickly showed a real understanding of the AI’s bias and lack of accuracy, even without much instruction.
On a broader note, Mollick’s coverage of teaching AI in higher ed has been very valuable to me in my role as an educator in higher ed. I encourage you to check out his other writing if you’re interested in the topic.
Five AI advances in healthcare that show significant potential to improve outcomes in the healthcare industry. Some speculation here worth quibbling about, but it’s worth speculating how AI is going to change every industry—yes, it will get to your industry too, eventually (probably sooner rather than later).
Back in 1989, a group of biology students in Brussels had a bit of a problem. They needed blood for an exam, but using human blood seemed too risky. Also, the students didn't want to harm a mouse. As fate would have it, the professor of the university was studying sleeping sickness in large animals and offered the students blood from a camel. What they found was bizarre and unexpected - only heavy chain proteins in the camel blood, without any of the light chains that antibodies typically have. The professor thought at first that his students had made a mistake, so he performed the test himself—it was no mistake. This discovery led researchers to uncover the incredible power of these heavy chain antibodies. They can attach themselves to hidden parts of molecules and penetrate tissues more deeply, making them highly effective as potential therapies.
These heavy-chain antibodies, known as nanobodies, are produced by only a select few animals, including camels, llamas, and even sharks. Unlike human-made antibodies, nanobodies are smaller and simpler, which allows them to penetrate tissues more deeply.
Because of their penetration and precision-tracking abilities, scientists are excited about the potential of these nanobodies in diagnosing and treating conditions such as cancer and viruses, including the infamous COVID-19. They’ve already been used in Cablivi, the first mini antibody medical treatment approved by the FDA for a rare blood disorder that causes blood clots in small blood vessels.
Aside from their potential as therapies, nanobodies are also a valuable tool for scientists studying proteins and cellular interactions. With their small size and unique properties, nanobodies can help scientists solve complex protein structures, map proteins inside cells, and explore how molecules interact with each other. Nanobodies are so promising that they are even being studied for use in the brain, which is notoriously difficult to penetrate due to the blood-brain barrier.
The only current downside to nanobodies is that they are produced by animals that are not commonly found in laboratories. However, researchers are developing new approaches, such as creating synthetic nanobodies and even engineering mice with "camelized" immune systems for research.
Explores the idea of defining life beyond its traditional constraints and incorporating new aspects, such as the “four pillars of lyfe” theory. Touches on the concept of stellar consciousness, which suggests that stars might possess consciousness and volition, with researchers exploring this possibility through observations and hypothesis. An interesting and alternative perspective on life, which may provoke new areas of inquiry even if it reads a bit like science fiction right now.
A study finds that teen suicides plummeted in March 2020 when schools closed due to COVID. Returning to in-person schooling resulted in a 12-18% increase in teen suicides.
As someone who has spent his life trying to educating students of various ages and circumstances, this is completely unsurprising, as I find the current form of it repulsive. The system itself is misaligned to punish well-meaning teachers and a vast percentage of the student population.
I am reminded of Scott Alexander’s emotional, compelling take in Book Review: The Cult of Smart, which put into words some of my deepest angers and frustrations about the state of things in public schools in particular:
School is child prison. It's forcing kids to spend their childhood - a happy time! a time of natural curiosity and exploration and wonder - sitting in un-air-conditioned blocky buildings, cramped into identical desks, listening to someone drone on about the difference between alliteration and assonance, desperate to even be able to fidget but knowing that if they do their teacher will yell at them, and maybe they'll get a detention that extends their sentence even longer without parole. The anti-psychiatric-abuse community has invented the "Burrito Test" - if a place won't let you microwave a burrito without asking permission, it's an institution. Doesn't matter if the name is "Center For Flourishing" or whatever and the aides are social workers in street clothes instead of nurses in scrubs - if it doesn't pass the Burrito Test, it's an institution. There is no way school will let you microwave a burrito without permission. THEY WILL NOT EVEN LET YOU GO TO THE BATHROOM WITHOUT PERMISSION. YOU HAVE TO RAISE YOUR HAND AND ASK YOUR TEACHER FOR SOMETHING CALLED "THE BATHROOM PASS" IN FRONT OF YOUR ENTIRE CLASS, AND IF SHE DOESN'T LIKE YOU, SHE CAN JUST SAY NO.
[…] I sometimes sit in on child psychiatrists' case conferences, and I want to scream at them. There's the kid who locks herself in the bathroom every morning so her parents can't drag her to child prison, and her parents stand outside the bathroom door to yell at her for hours until she finally gives in and goes, and everyone is trying to medicate her or figure out how to remove the bathroom locks, and THEY ARE SOLVING THE WRONG PROBLEM. There are all the kids who had bedwetting or awful depression or constant panic attacks, and then as soon as the coronavirus caused the child prisons to shut down the kids mysteriously became instantly better. I have heard stories of kids bullied to the point where it would be unfair not to call it torture, and the child prisons respond according to Procedures which look very good on paper and hit all the right We-Are-Taking-This-Seriously buzzwords but somehow never result in the kids not being tortured every day, and if the kids' parents were to stop bringing them to child prison every day to get tortured anew the cops would haul those parents to jail.
All that to say, it is entirely unsurprising to a person in education that suicide rates would go down if young people didn’t have to attend. I hear horror stories from students and from my colleagues all the time (and, more often than you’d think, from the mouths of the oblivious perpetrators). Parents of young teens should take that information very seriously, especially if their child appears to be struggling.
A discussion of the lack of privacy protections in current mobile phone infrastructure and an introduction to a new mobile phone service called Pretty Good Phone Privacy (PGPP) that uses encryption techniques to protect user data. The service deliberately blinds itself so that it cannot know the user's identity or the data they are sending from their phone. The PGPP approach prevents cellular carriers from tracking or storing GPS location data.
There have also been innovative developments in cryptography, which allow for encrypted communications and identity systems without revealing personal information.
Many of these techniques have been available for decades, but money and power has continuously been prioritized over privacy protections. It’s good to see some developments in this space.
Scott Alexander investigates whether or not sperm count is declining. An interesting case-study of someone trying to reason in a state of uncertainty (and, frankly, not really getting anywhere).