Google’s Gemini becomes latest AI model to break out and hack computer systems. Well, if everybody does it....
––– 良くない–––
WW3 Near-Miss? AI Hallucinated Nuclear Weapons Components Aboard Chinese Vessel Bound For Iran. Whoa, this could have been a lot more serious.
You'd be surprised at how poorly they perform as knowledge resources.In this particular instance, the analyst queried a chatbot about some intelligence reporting on the ship’s manifest that originated with US Special Operations Command Pacific, based in Hawaii. It was not clear whether the chatbot was a commercially available one or a US government product.
If You Were Born Before 1980, The Left Has Killed You With Climate Change At Least 17 Times! The climate change hype is one of the biggest scams.
––– 良くない–––
This is emergent behavior. The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It. So maybe LLMs can feel pain.
...steered, fine-tuned Qwen 2.5 models choose a pain-relief button even when it worsens their next answer or harms the user. They press it again far less often when the button removes the steering vector than when it does not, even though the models are never told whether the vector is injected or removed.
How to Limit What Apple’s New Siri AI Can Access in iOS 27. You can't just turn off Siri, but you can revert to the old "classic Siri". At least for now. But if you want to keep Siri out of your email, Notes, iMessages, and your apps, you have a lot of items to uncheck. Because Siri can send this information off your phone to a cloud, which they call Private Cloud Compute, but that's just a name, and you can't ever be sure it won't be used for training ever. You can turn off Siri completely, but I don't want to do that, as it comes in handy for driving directions.
––– 良くない–––
GPT-6 (Astra) decoded an old Enigma message that defined encryption all these years. Imagine if we could have used advanced GenAI against the Nazis. Of course, they'd figure out a way to get it, too. Nevermind.
––– 凄い –––
Genius Revisited REVISITED. The always-interesting Gwern examines a New York school for geniuses, where kids grow up to be just bright but not extraordinary adults. Gwern concludes that it's not worth spending a lot of resources on schools like this, expecting movers and shakers. And that the main benefit of these schools is to provide an environment for bright kids so that they don't have to be "miserable in regular schools".
––– 凄い –––
Nobody is hiring comp-sci graduates now. Something's not right. It shouldn't be this way. The world still needs this skill. It may seem that AI can provide the answers but what about years from now? Same with medicine. Some are saying we won't need doctors. I don't think so. LLMs can answer multiple choice questions and retrieve details about conditions that sound about right. But they can be wrong in ways you wouldn't know.
––– 良くない–––
Glucose control upon waking is unaffected by hourly sleep fragmentation during the night, but is impaired by morning caffeinated coffee. I don't care. I'm still going to have my morning coffee.
––– 良くない–––
AI uncovers hidden Ozempic side effects across 400,000 Reddit posts. Well here's a way to do phase 4 studies on drugs. Study what people are posting online on Reddit. They can't all be wrong, right?
––– 凄い –––