We have an even mix this months between science news about the biological sciences and LLMs. Progress!
The most shocking of the bunch is the discovery that so-called sex reversal seems more common in birds than previously thought. The team dissected the bodies of nearby 500 birds of five common Australian species and found sex-revered individuals in all five species at rates of 3% to 6%. Nearly all of these samples were genetically female but had male reproductive organs. However they also found a few genetic males with ovaries and one had evidence of recently having laid an age. This is yet another demonstration that sex is more fluid and mutable than many think but it should be cautioned as this finding applies only to birds.
News on the mRNA vaccine development front has been depressing recently as the Trump administration keeps cracking down on it and cutting funding. This paper shows some of the potential for major medical applications that might be lost. It proposes that a broad-spectrum mRNA-based antiviral could be formulated. It works by activating interferon-stimulated genes that protect cells against viral infection at the cost of possibly causing mild and often self-resolving inflammation. What’s exciting here is that it seems to help make the treated cells resistant to a wide range of different viruses including including the Zika virus, vesicular stomatitis virus, and SARS-CoV-2. As usual this is an extremely early finding and there are many practical issues, but it demonstrates the viability of a completely fresh approach to developing antiviral prophylactics.
Both of the LLM papers are about the impact the deployment of the technology is having on society, possibly in subtle ways. The first one discusses whether ChatGPT in particular is being used by investors to inform their trading decisions. The authors hit upon a novel way to measure this by matching outages of the ChatGPT service with stock trading activity. They show that trading volume does indeed decline when the service is not available and that this effect is stronger for companies which have released corporate news immediately before or during the outage. There’s not much else that can be concluded so far but it does suggest that AI use is widespread and is indeed being used to make investment decisions.
A much more subtle and yet important influence of increasing LLM use among the general public can be seen in this other paper. They attempt to quantity the impact of LLMs on human culture by analyzing human discourse from sources such as YouTube videos and podcast episodes. They claim that they were able to measure an increase in the usage of words commonly used by ChatGPT. As they note, this marks the beginning of a cultural feedback loop in which LLMs are trained on human data yet the widespread usage of LLMs itself reshape how humans use language.
Once again, a good number of biology articles this month but there is one about the Earth’s magnetic field which may be related to the planet bearing life, or it might not.
Starting with that first, the study is about the realization that the strength of Earth’s magnetic field seems correlated with the abundance of oxygen in the atmosphere, roughly rising and falling together over the past billion years or so. The strange thing is that there is no obvious reason why they might be correlated with the best guess being merely that the magnetic field helps to deflect solar wind, thus slowing the loss of oxygen and other gases into space. Another explanation might be that tectonic motions can both affect the geomagnetic field and release nutrients that fuel blooms of oxygen-producing algae.
Next is a development combines both biology and information technology. A private institute has just announced the creation of the first virtual model of a cell. This specific version is designed to predict how stem cells, cancer cells and immune cells respond to drugs, cytokines or genetic perturbations. The obvious benefit is that using a virtual model of the cell is far easier than having to run the experiment using real cells. Everyone will be testing to see how accurate the responses of this particular model will be but this is only the first attempt and future versions are only going to be better.
Then we have this news about an experiment that has dramatically extended the lifespan of mice by giving them monthly injections of an antibody. The antibody in question, X203, works by inhibiting the natural protein interleukin 11. This is a cytokine whose concentration in the body increases with age and is associated with inflammation and cancer. The team claims that interleukin 11 has been observed to work similarly in humans as in mice but of course we still need to what effects this injection has on more subjects beyond mice and what the long-term effects other than extending life would be.
Finally here’s an article that I included because it feels like something right out of the videogame Bioshock to me. It talks about how a specific species of sea slugs are able to gain new abilities by consuming algae with those abilities. For example they are able to gain photosynthesis from the algae when consuming. Instead of digesting the chloroplast cells, they build structures of their own around them, and keep them working. Similarly they are able to store the stinging cells from the sea anemones they consume and deploy them later against their own enemies.
A bunch of cool science stuff this month, all of which stem from biology but touch upon different topics.
The simplest and most directly beneficial of these announcements is the US approving a twice-a-year shot that will completely prevent HIV. This doesn’t quite count as a vaccine as it does need to be taken twice a year but it is the longest lasting protection against HIV yet. In its original form, it’s probably too expensive for widespread use in developing countries but generic copies will eventually become available and that could well mean the end of AIDS.
Ever since it was first identified in the 1980s, the Flynn effect has invited endless speculation on the cause of the increase in intelligence and whether the trend will continue. This new paper uses data from the Norwegian Armed Forces which has administered a general mental ability test on recruits since 1954. It argues that while some types of skills improved over time, notably non-verbal reasoning ability, word reasoning and numerical reasoning abilities decreased over the same period. This suggests that it might not be accurate to say that intelligence in general has increased over time.
Next we have a paper that claims to advance our understanding on how sperm whales communicate. It argues that their codas, the series of clicks that they use to communicate, not only resemble human vowels acoustically but also appears in patterns similar to human language. These findings are used to construct a case that these codas are intentionally controlled and constitute a type of language that we could one day unlock.
Finally here’s a paper that attempts to do the seemingly impossible, study the phenomenon of qualia itself. The team purports to answer the age-old question of whether your ‘red’ is the same as my ‘red’ by doing the following: collect detailed reports about the relations between sensory experiences of various participants; construct what they call qualia structures from the data, meaning embeddings of qualia that represent the similarity or lack thereof of the participants’ judgment of what they have experienced; compares two individuals’ qualia structures in what they call an unsupervised alignment method which doesn’t assume that there are particular correspondences between the structures. The upshot is that the team could align the structures of color-neurotypical group, meaning those who are not color blind and self-report seeing colors normally. They could also align the structures of the color-atypical group despite differences in the type of color blindness within that group. But they could not align the structures between the color-neurotypical and color-atypical group meaning that those structures are too different. It’s hard to say what we can conclude from these results, perhaps that it might be possible to prove that those who experience colors in the same way do actually share the same qualia. But the ambition behind the project is something I can certainly get behind.
A few interesting announcements for this round and once again, it feels that the really cool developments are going on in the life sciences.
By far the most significant bit of news, and one that I’d hoped would be shouted from the rooftops, is the successful use of the CRISPR gene-editing technique to treat a child’s unique mutation. The patient in question was born with a rare disease known as carbamoyl-phosphate synthetase 1 (CPS1) deficiency which usually kills in infancy. In this case, doctors were able to devise a targeted fix to edit DNA in the liver cells, test the treatment in mice, get approval from the FDA and administering the treatment all in a matter of months. They still don’t know if the effect is permanent but for now the patient is well enough to be able to leave the hospital since being born. The key here is the incredible speed of this achievement and the fact that this is a treatment personalized for that specific patient. We should all hope that this is only the first of many such achievements.
Next up is a medical development that feels like something out of science-fiction. An American man, Tim Friede, had for personal reasons exposed him to the venom of a large variety of snakes in escalating doses over the course of 18 years. As a result, he has generated antibodies that are effective against multiple types of neurotoxins. By isolating the antibodies in his blood, researchers have devised a broad spectrum antivenom that should now be almost universally effective against snakebites. If the project comes to fruition, this would be a radical improvement from the current practice of needing to manufacture and stock multiple types of antivenoms in case of emergencies.
The last paper was released a couple of months earlier but I hadn’t noticed it then. It’s about a study conducted on the Hadza, a hunter-gatherer tribe in Tanzania whose nomadic lifestyle is thought to have been unchanged for thousands of years. This makes them an ideal subject to interrogate about human nature before the advent of civilization. In this instance, researchers were interested in whether such peoples are inherently disposed to be egalitarian as some anthropologists contend. They gave the Hadza participants food endowments to be shared with others and studied how fairly they carried out the redistribution. They found that like just about everyone else, the Hadza mostly did not seek perfectly equitable distributions. They tolerated inequality when it benefited but complained about it when they thought it was unfair. As usual, it’s unwise to generalize too broadly from such studies but I’d always thought that stories of pre-civilization peoples being inherently more noble to be too fanciful to be true.
Not that much in the way of science news this month. Even the scientists are talking about economics and politics.
The really big news this month has been in cosmology from analysis of three years’ worth of data from the Dark Energy Spectroscopic Instrument. The consensus is that, whatever it is, the effects of dark energy seems to be weakening over time, rather than being constant as previously believed. This finding is also supported by measurements of the cosmic microwave background. We don’t know what this means yet, only that it throws into question our current model of the universe.
In more uplifting news, Japanese scientists have announced some success in giving paralyzed people the ability to walk again through stem cell therapy. The team had already succeeded earlier by transplanting neural stem cells into patients with spinal cord injuries, but back then they had used cells from fetuses. This time they repeated the feat with reprogrammed stem cells, which takes mature cells and induces them to become pluripotent. The success so far is only partial but the benefits to humanity of making this work consistently are obvious.
Finally here’s a report about how DNA analysis of two naturally mummified remains found in the Sahara Desert a different and long lost lineage of humanity. Around 7,000 the area was not a desert and instead a lush, green savannah. These two ancients humans however came from a branch of humanity that split off much earlier than that, perhaps as much as 50,000 years and so were something like ancient fossils even while they were alive. It’s simply wild how much we don’t know about our past and now will probably never know.
This month we have both a good number of science announcements interesting enough to be included here and they are also so important that everyone should know about them.
Let’s start with the one that will raise the most hackles. It’s a study of studies aimed at measuring the accuracy of gender stereotypes. The impolitic yet unsurprising finding is that they are mostly true in that people are generally accurate at assessing whether men or women were higher on some given trait or characteristic. That said, they underestimate gender differences in cognitive abilities and academic performance and overestimate them in personality traits and behaviors. Also worth noting that individuals were less accurate at these judgments than the group as a whole, a finding that is in line with similar studies.
Next is a paper that seems especially relevant to us in Malaysia given recent news about health insurance costs. It discusses the rise of genetic testing and how the data thus obtained can be used to predict health outcomes. Yet many jurisdictions around the world have banned insurers from pricing their products using genetic data. This creates a mismatch between individuals who have undergone genetic testing and insurers that will only grow worse as genetic testing becomes more pervasive and more accurate. It is difficult to see how health insurance markets would continue to operate under such conditions.
The major medical news this month has to be the so-called ‘tumor to pork’announcement. It’s about a Chinese team inserting a pig gene into a virus and using it to infect cancerous cells. This then causes the body’s own immune system to treat the cancer as a foreign entity and attack it. The virus used, the Newcastle disease virus, is thought to be relatively harmless to humans though it is deadly to birds. The team used the technique to treat a variety of untreatable cancers including liver, ovarian, cervical etc. with excellent results.
Finally a major discovery in the field of cosmology. Data from the James Webb Space Telescope reveals that around two-thirds of observable galaxies spin in one direction while remainder spin in the other. This contradicts the existing model of a homogenous universe in which more or less equal numbers of galaxies should spin in the opposite directions. One explanation that has caused excitement is that our own universe exists within a black hole and therefore was itself born rotating. This carries the implication that every black hole in our universe is a doorway to another baby universe. The less exciting explanation is that our own perception is skewed by the rotation of the Milky Way galaxy. I like the exciting explanation but in cases like this it’s usually the mundane one that holds true over the long term.
More science stuff this month and there’s been a lot of buzz about humanity getting close to the Singularity. I’m going to maintain a more skeptical attitude and curate only the more plausible news announcements.
Let’s start with an announcement that is very exciting and has made headlines around the world as Microsoft no doubt intended, but has also left many physicists being skeptical. They claim that they now have a path towards viable quantum computers that work based on topological qubits. The qubits are built out of superconducting nanowire with each end being able to host a different topological state called a Majorana quasiparticle. They’ve also shown off a chip that they claim has eight such qubits. Other experts however say that there’s no proof in their paper that it works as claimed especially since the company has made similar claims before that were retracted later.
Developments in the LLM space are too numerous to cover in detail so I’ll only focus on the big picture stuff. If you understand how the technology works, it’s always been a little strange why LLMs can do math at all. Through experiments on mid-sized LLMs, the writers of this paper claim that they do so using algorithms that no human would naturally use. Specifically they argue that numbers are stored in the LLMs as a generalized helix that is then manipulated to produce an answer. They show that even for simple arithmetic operations, the LLMs perform trigonometry to arrive at the answer. As LLMs grow increasingly big and sophisticated in capability, how they actually reason internally becomes ever more obscure as well, making this type of research critical.
Next we have two papers on genetics. The first one suggests that severe depression and suicide ideation can be detected from blood markers. The tests are not identical for men and women as different metabolites are involved but the common thread is that they test for mitochondrial dysfunction. The researchers hypothesize that stress at the cellular level could overwhelm the body and trigger suicidal thoughts. The utility of using a blood test to detect severe depression are massive and so too would be the implication that treatments aimed at repairing metabolic function, including the use of supplements like folate and carnitine, could reduce the risk of suicide.
Finally the last paper covers the increasingly powerful predictive power of genetic data when applied to health outcomes. It notes that in many countries, insurers are banned from using genetic information to price their insurance products. Yet nothing prevents individuals themselves from making insurance-purchasing decisions based on their own genetic profiles. If such practices were scaled up enough, it would effectively break the entire health insurance market.