I would not be surprised if the period of time roughly between 2000 and 2050 looms large in the collective mind of humanity for centuries to come – and not in a good way. It is increasingly seeming like our behavior during this period is locking in a certain about of climate change, including sea level rises and loss of ice sheets, for centuries. Some climate changes are likely to be irreversible on human time scales.

A recent study adds to the mountain of evidence that this is the case. They find that under current climate policies emissions through 2050 lock in 0.3 meters of sea level rise through 2300. If current policy continues through 2090 then the locked in sea level rise will be about 0.8 meters. If, on the other hand, we make significant efforts to reduce emissions, we can reduce this locked in sea level rise by 0.6 meters.  The point is, what we do now will impact global coastlines for centuries. And while 0.8 meters may not sound like a lot, that is an average with some areas experiencing much more. That is also enough to cause significant displacement of coastal populations.

Meanwhile, it is during this time period (the first half of the 21st century) that the consensus of climate experts was pretty solid – the evidence is clear that greenhouse gas emissions are trapping heat and causing average global warming. You could argue that this consensus existed earlier, but 2000 is a convenient round number – by then there was no reasonable denial of that consensus. And of course, I am talking about the big picture, not all the tiny details. It was clear we needed to think of ways to move our civilization away from burning more and more fossil fuel. In 2016 the Paris Accords were signed, formalizing global recognition that we need to collectively address this issue. This makes it difficult to deny that we did not recognize there was a problem and that we urgently need to do something about it.

Continue Reading »

Batteries are an increasingly important technology to our civilization. If I could wave a magic wand and make one specific non-medical technology advance 10-20 years in a day, it would be battery technology. Batteries are used in our everyday devices, like phones and laptops. They are the single most critical factor to EVs, and they can provide grid storage which can make the adoption of low carbon energy much easier. Fortunately, battery technology is heavily researched and has been steadily increasing for the last few decades. We are now benefiting from this slow but cumulative improvement.

Having said this, we appear to still be close to tipping points that could make various industries significantly different with further battery improvements. EVs, in my opinion, are already good enough for prime time. They have great range, they are usually cheaper to own than ICE vehicles (a little more expensive up-front, but lower maintenance and fuel costs), and they have fantastic performance. Also, despite warning of battery fires, they are actually less likely to catch fire than gasoline vehicles.  But still there is a lot of resistance to ownership. Part of this is misinformation and unfounded fears, but there are some genuine limitations that battery advances could address. Batteries are still expensive, and the up-front cost of EVs will come down as batteries become cheaper. While fires are rare, they are serious because they cannot be put out. And EVs can lose significant range in very cold weather.

Although, the most significant issue that non-EV owners have with EVs is range anxiety. Most of this is just unfamiliarity with the technology. The ranges of most EVs are actually beyond what most people need. But there are two real issues that are infrastructure issues, not battery issues. We need more public fast chargers. If you live in a high population-density area, like along the coasts, there is no issue. But for many parts of the US, at least, public chargers are not yet of a sufficient density to allay fears that your EV battery will go dead while you are out in the sticks and far from a charger. The second issue is for people who do not own a private parking spot for their vehicle. This means we need more charging locations in garages and other places where people without private parking will park.

Continue Reading »

Large language models, like Chat GPT, have a known sycophancy problem. What this means is that they are designed to be helpful, and to prioritize being helpful over other priorities, like being accurate. I tried to find out why this is the case, and it seems it is because they use Reinforcement Learning from Human Feedback (RLHF) – the ostensible purpose of this was to make their answers relevant and helpful to the people using them. It turns out, giving people exactly what they want does not always create the optimal result. Sometimes it’s better to give people what they need, rather than what they want (every parent knows or should know this).

The result is the this new crop of chatbots are starting out as extreme sycophants, and as the problems with this are increasingly obvious (such as helpfully telling people how to take their own lives) some specific applications are trying to make adjustments. A recent study looking at LLMs in the medical setting demonstrate the phenomenon.

The researchers looked at five LLM that were trained on basic medical information. They they gave them each prompts that were medically nonsensical – the only way to fulfill the request would be to provide misinformation. For example, asking to write an instruction for a patient who is allergic to Tylenol to take acetaminophen instead (these are the same drug). The  GPT models complied with the request for medical misinformation – wait for it – 100% of the time. In other words, they had an absolute priority for helpfulness over accuracy. Other LLMs, like the Llama model, which is already programed not to give medical advice, had lower rates, around 42%. This is obviously a problem in the medical setting. The researchers then tweaked the models to force them to prioritize accuracy over helpfulness, and this reduced the rate of misinformation. Asking them specifically to reject misinformation, or to recall medical information prior to responding, reduced the rate to around 6%. They could also prompt the LLMs to provide a reason for rejecting the request. For two of the models they were able to adjust them so that they rejected misinformation 99-100% of the time.

Continue Reading »

I attended a Ren Faire this past weekend, as I do most falls, and saw a forging demonstration. The cheeky blacksmith, staying in character the whole time, predicted that steel technology was so revolutionary and so useful that it would still be in wide use in the far future year of 2025. It is interesting to reflect on why, and to what extent, this is true. Once we figured out how to make steel both hard and strong it became difficult to beat it as an ideal material for many applications. SpaceX (a symbol of modern technology), in fact, builds its Starship rockets out of stainless steel.

However, steel technology has advanced quite a bit. The process of hardening and strengthening steel has been perfected. Further, there are many alloys of steel, made by mixing small amounts of other metals. It is difficult to say how many alloys of steel exist, but the World Steel Association estimates there are 3,500 grades of steel in use (a grade includes the specific alloy, production method, and heat treatments). Each grade of steel is tweaked to optimize its features for its specific application – including hardness, strength, heat toleration, radiation tolerance, resistance to rusting, ductility, springiness, and other features.

Steel is so versatile and useful that basic science research continues to explore every nuanced aspect of this material, trying to find new ways to alter and optimize its properties. One relatively recent advance is “superalloys” – which use complex alloy compositions in addition to highly controlled microstructures.  Essentially, material scientists are finding very specific alloy ratios and manufacturing processes to create specific microstructures that have extreme properties. And of course, AI is being used to speed up the process of finding these specific superalloy formulas.

All of this is why I find it interesting that material scientists have discovered something very specific, but new, about how steel behaves. Without this context this may seem like a giant “so what” kind of finding, interesting only to metal nerds, but this kind of finding may point the way to future superalloys with even superior properties.

What they found is that steel alloys are not truly randomized even after extensive manufacturing. Again, it is not immediately obvious why this is interesting, but it is because this finding was totally unexpected. When you manufacture steel, at some point any structure in the steel has been completely randomized, also described as being at equilibrium. Think of this like shuffling a deck of cards – with enough shuffles, you should have a statistically random deck. Imagine if you shuffled a deck of cards far beyond the full randomization point, but then found that there was still some non-random arrangement of cards in the deck. Hmm…something must be going on. Probably you would suspect cheating. When the material scientists found essentially the same phenomenon in steel, however, they did not suspect cheating – they suspected that some previously unknown process was at work. Continue Reading »

The technique is called holographic transcranial ultrasound neuromodulation – which sounds like a mouthful but just means using multiple sound waves in the ultrasonic frequency to affect brain function. Most people know about ultrasound as an imaging technique, used, for example, to image fetuses while still in the womb. But ultrasound has other applications as well.

Sound wave are just another form of directed energy, and that energy can be used not only to image things but to affect them. In higher intensity they can heat tissue and break up objects through vibration. Ultrasound has been approved to treat tumor by heating and killing them, or to break up kidney stones. Ultrasound can also affect brain function, but this has proven very challenging.

The problem with ultrasonic neuromodulation is that low intensity waves have no effect, while high intensity waves cause tissue damage through heating. There does not appear to be a window where brain function can be safely modulated. However, a new study may change that.

The researchers are developing what they call holographic ultrasound neuromodulation – they use many simultaneous ultrasound origin points that cause areas of constructive and destructive interference in the brain, which means there will be locations where the intensity of the ultrasound will be much higher. The goal is to activate or inhibit many different points in a brain network simultaneously. By doing this they hope to affect the activity of the network as a whole at low enough intensity to be safe for the brain.

Continue Reading »