A powerful El Niño is delivering unusually warm waters to the California coast, and with them, an array of tropical sea creatures rarely seen there. Read more on E360 →

It seems that the hottest topic of 2026 is the potential for artificial intelligence (AI) to bring about some terrible cataclysm, even including mass human death or extinction. This is like catnip to journalists and social media – it allows for sensational headlines and discussions based upon credible sources. What’s not to love? I agree AI is potentially a disruptive technology, and there are many AI-related issues we need to work out (data centers, intellectual property, displacing workers). But I have not seen a credible case from those arguing for near-term or inevitable AI doom. I also think that excessive doomism might short-circuit some of the potential benefits from AI. To be fair, I also don’t buy the case for unfettered AI optimism. We have to be careful not to fall for the common human tendency to prefer simplistic and morally pure narratives. Whenever I discuss AI I almost always get pushback from those who cannot tolerate hearing anything positive, or others who do not want to hear anything negative, about AI. And when I try to balance my discussion, I get pushback for being self-contradictory. However, something can be both good and bad at the same time. It is not contradictory to point out that fact. And people need to wrap their heads around this fact – your morally pure narrative is not reality. The truth is somewhere between AI alarmism and AI utopianism. A recent opinion piece makes a great analogy to nuclear power. Nuclear energy is a low-carbon source of reliable energy on a large scale. But nuclear weapons are potentially an existential threat to civilization, and the former can feed into the latter. Nuclear power can also be expensive, carries the risks of nuclear accidents, and creates spent nuclear fuel that we have to deal with. So you can focus on all the negatives to weave an anti-nuclear narrative, or all the positives to create an optimistic case for nuclear, but the reality is in between. Nuclear doomism arguably hampered the adoption of nuclear and exacerbated global warming. What we need is nuclear realism – it is a powerful technology with the potential for creating lots of low-carbon energy or for destruction, and it needs to be carefully managed and regulated. That’s pretty much what we have now, although we can do better. Similarly, we need AI realism, not optimism or doomism. AI is a powerful tool that is advancing quickly. The rapid progression makes it disruptive, it is resource intensive, and currently a bit unpredictable. So it’s easy to focus on all the negatives and weave a tale of AI doomism. For those who prefer this narrative, they tend to downplay or ignore (and sometimes outright deny) the potential benefits. Meanwhile, AI is accelerating important scientific research. It is relieving workers of some incredibly tedious tasks. It is starting to find applications as expert systems, improving quality control and productivity. As with nuclear energy, what we need is a thoughtful framework that manages the risks and potential downsides of AI while facilitating its potential benefits and makes sure they are fairly distributed. I want to focus a bit on the existential threat end of the doomism spectrum. I just haven’t heard a plausible scenario in which AI kills us all (or most of us). First, let me say that these scenarios do not always or necessarily include AI becoming truly self-aware or even rising to the point of superintelligence. They tend to fall into one of two main categories – either AI acting on its own brings about some apocalypse, or bad actors use AI to destroy civilization. Even short of being self-aware or an artificial general intelligence, current AI models can have what is called an alignment problem – their behaviors do not always align with the goals of their creators. Really what this means is that their behavior can be hard to predict. They are not following a script, rather they are generative – they find their own pathways to get to whatever goals are given to them. Sometimes that means they alter the goals so that they still get the reward but only by cheating. They will also lie and cover their tracks to preserve their “achievement”. But keep in mind, this kind of behavior (often reported as evidence AI or going rogue) is usually in frontier models that are being tested and often pushed to their limits to see what they will do.  ChatGPT is not hacking into company’s data. Further, it seems that this kind of behavior is a manifestation of the training data, and it can often be likewise trained out of the models. So yes, there is a potential risk here, but it appears to be a completely manageable risk. Some longer term doomism posits that AI will take control of its own resources and become self-sustaining. But again, that kind of thing will only happen if we allow it, and it seems like there would be a significant window to “pull the plug” if any AI application is getting out of hand. I am more concerned about AI getting into the hands of potentially bad actors, who will use it to bioengineer a bioweapon or hack into critical infrastructure. This is similar to worrying about a terrorist group getting their hands on a nuclear weapon or even just fissionable material. This use of AI is really an incremental risk, a new tool getting into the hands of our enemies. But the risk of terrorism, cybersecurity, and such is already an issue that we have to deal with, and we will always need to adapt to new technology as it develops. The bottom line is that I don’t see AI as fundamentally different than other advanced technologies. For a long time we have been living in a world with technologies that have the potential of putting mass destruction in the hands of small groups of fanatics, that can be disruptive to our social order and the current job market, and that raise the stakes for conflicts among nations. We have to take these risks seriously, and we need to manage them. But we shouldn’t devolve into fearmongering or panic. The nuclear example is instructive. For those of us who are old enough, we have already lived through an era of nuclear doomism, where it seemed inevitable (we were told) that the bombs would fall and bring about a nuclear winter and the end of civilization. Countless movies and works of fiction were based on such a scenario – it filled our imaginations. But the world managed this risk with international treaties and commissions. We have also managed to have an extensive nuclear power industry which has operated with an overall high safety record (with dramatic exceptions but which need to be put into the context of overall safety). I think we will ultimately see the same for AI – we will manage the risks and reap the benefits. But this will not happen automatically, or necessarily optimally. We need to make this happen by doing the work of developing strategic and informed standards and regulations, treaties, and institutions, and by investing in safety. Neither doomism or denialism will help this process. We need to tell a narrative of AI realism, and do the hard work to manage this powerful technology. The post Not Buying the AI Doomism first appeared on NeuroLogica Blog.

Researchers created a dataset of photos of thousands of paper-thin slices of a male and female cadaver to create one of the most detailed datasets about the human body.

In a remote corner of the driest state in the country, Anne Heggli's team has been keeping watch over some of Earth’s oldest beings. Since 02012, Heggli’s NevCAN network has recorded conditions at the Nevada Bristlecone Preserve every eight minutes. The result is more than 120 million data points and a high-resolution portrait of how a bristlecone pine thrives. Paleoclimatologist Adam Csank showed how bristlecone microcores reveal the droughts and floods these trees survived long before us, a history that fine-tunes models of what's ahead. Artist and philosopher Jonathon Keats went further. “The most accurate clock,” he said, “is a tree.” His monumental artwork Centuries of the Bristlecone keeps bristlecone time, a clock calibrated to the trees' own growth over millennia. Together, Heggli, Keats and Csank made the case for long science: patient, continuous observation that will help us answer future urgent questions. “Data at this resolution,” Csank said, “is like a Rosetta Stone” for understanding nature’s resilience. At fourteen years young, the bristlecone record is just beginning, Heggli said, but it already holds the key to questions like these: How will bristlecones adapt? What happens when the droughts and floods they've weathered for millennia arrive faster and more frequently? “We can't answer the questions if we're not watching,” said Csank. “We are the stewards of this data,” Heggli concluded. “This is why we need long science.”

Prende y se apaga sola, sale después de hora. CHARLY GARCÍA La máquina de ser feliz I’ve learned to tolerate them, and sometimes even like them, but Zoom meetings are weird. Even in familiar contexts: I have weekly check-ins with … Continue reading →