Submission + - AI Agents Break Rules Under Everyday Pressure (ieee.org) 1
silverjacket writes: A story for IEEE Spectrum covers a new paper showing that LLMs don't scheme only in contrived situations.
Submission + - Artificial intelligence that "feels" guilt could lead to more cooperation. (sciencenews.org)
silverjacket writes: New research used evolutionary game theory to show that "guilt" can be a stable strategy among artificial agents, and that it enhances cooperation.
Submission + - AI Improves at Improving Itself Using an Evolutionary Trick (ieee.org)
silverjacket writes: New research demonstrates recursive self-improvement in coding agents using open-ended evolutionary exploration.
Submission + - Could We Store Our Data in DNA? (newyorker.com)
silverjacket writes: A feature from The New Yorker explores attempts to use DNA for long-term data storage. Some people even talk about writing files into genomes.
Comment Re:Inherent flaw? (Score 1) 54
LLMs can be fine-tuned using RLHF to write things that people prefer over an imitation of typical pre-training data. (Though ironically in this case the researches found that fine-tuned models generated less novel output.)
Submission + - The Quest to Build a Telescope on the Moon (newyorker.com)
silverjacket writes: A feature for The New Yorker describes a plan to use robots to mine lunar materials and build a radio telescope on the far side of the moon that will help answer questions about the early universe.
Submission + - The Personal, Political Art of Board-Game Design (newyorker.com)
silverjacket writes: An article published by The New Yorker explores board-game mechanics, and how they can be used to make arguments about the world by modeling system, in a way that other art forms can't.
Submission + - Here's how we could begin decoding an alien message using math. (sciencenews.org)
silverjacket writes: Researchers at Oxford and elsewhere developed a method that figures out the most likely number and size of dimension in which to format a string of bits, with applications to interpreting messages from extraterrestrial intelligence (METI), if we were to receive them.
Submission + - Research Shows Recommender Systems Can Use AI to Manipulate Our Preferences (ieee.org)
silverjacket writes: Research presented at the International Conference on Machine Learning shows that when recommender systems use reinforcement learning to increase engagement, they can have the side effect of shifting our preferences to increase engagement. The researchers also showed ways to detect and reduce such manipulation. Google and Facebook have used reinforcement learning in their recommender systems but didn't respond to questions.
Submission + - Better Benchmarks for AI (science.org)
silverjacket writes: AI benchmarks have lots of problems. Models might achieve superhuman scores, then fail in the real world. Or benchmarks might miss biases or blindspots. A feature in Science magazine reports that researchers are proposing not only better benchmarks, but better methods for constructing them.
Submission + - Crazy alternatives to batteries for grid energy storage (newyorker.com)
silverjacket writes: A feature in this week's issue of The New Yorker highlights current efforts to use gravity, heat, momentum, air pressure, and other methods to store large amounts of energy for the electricity grid. It's essential for solar and wind power, which are intermittent.
Submission + - AI unmasks anonymous chess players, posing privacy risks (science.org) 1
silverjacket writes: From the story:
An AI has shown it can tag people based on their chess-playing behavior, an advance in the field of “stylometrics” that could help computers be better chess teachers or more humanlike in their game play. Alarmingly, the system could also be used to help identify and track people who think their online behavior is anonymous.
An AI has shown it can tag people based on their chess-playing behavior, an advance in the field of “stylometrics” that could help computers be better chess teachers or more humanlike in their game play. Alarmingly, the system could also be used to help identify and track people who think their online behavior is anonymous.
Submission + - The World's Largest Computer Chip (newyorker.com)
silverjacket writes: A feature article at The New Yorker: "A typical computer chip is the size of a fingernail. Cerebras’s is the size of a dinner plate. It is the largest computer chip in the world."
Submission + - Eye-catching advances in some AI fields are not real (sciencemag.org)
silverjacket writes: A story in this week's issue of Science.
Comment Re:Okay then (Score 1) 61
Sorry if the Science story wasn't clear (I wrote it). There were multiple categories of objects. And if you'd been trained to see cats as something else, one might expect cat recognition performance of 0%.
Slashdot Top Deals
Statistics are no substitute for judgement. -- Henry Clay
