<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:media="http://search.yahoo.com/mrss/">
<channel>
<title>AI &amp; Computing | DexScience</title>
<link>https://dexscience.eu/ai/</link>
<description>Daily science news from trusted primary sources: space, biology, chemistry, geology, physics, AI, climate and health. Written in plain English and checked against the original research.</description>
<language>en</language>
<atom:link href="https://dexscience.eu/ai/feed.xml" rel="self" type="application/rss+xml"/>
<item>
<title>Google DeepMind releases Gemini 4 Neon, a mid-tier model it says cuts cost per task</title>
<link>https://dexscience.eu/ai/google-deepmind-releases-gemini-4-neon-a-mid-tier-model-it-says-cuts/</link>
<guid isPermaLink="true">https://dexscience.eu/ai/google-deepmind-releases-gemini-4-neon-a-mid-tier-model-it-says-cuts/</guid>
<pubDate>Thu, 19 Nov 2026 16:44:25 GMT</pubDate>
<category>AI &amp; Computing</category>
<description>Google says Neon pairs strong scores on coding, business and science tests with a lower cost per task. DexScience has not independently checked its figures.</description>
<media:content url="https://dexscience.eu/media/2026/11/google-deepmind-releases-gemini-4-neon-a-mid-tier-model-it-says-cuts/og.jpg" medium="image" width="1200" height="630"/>
</item>
<item>
<title>Google DeepMind opens Genie 4 world model to some Google AI Ultra subscribers</title>
<link>https://dexscience.eu/ai/google-deepmind-opens-genie-4-world-model-to-some-google-ai-ultra/</link>
<guid isPermaLink="true">https://dexscience.eu/ai/google-deepmind-opens-genie-4-world-model-to-some-google-ai-ultra/</guid>
<pubDate>Thu, 19 Nov 2026 16:38:50 GMT</pubDate>
<category>AI &amp; Computing</category>
<description>Users can now change a generated world mid-session by typing a new prompt, and Genie 4 adds real-time audio, the company says.</description>
<media:content url="https://dexscience.eu/media/2026/11/google-deepmind-opens-genie-4-world-model-to-some-google-ai-ultra/og.jpg" medium="image" width="1200" height="630"/>
</item>
<item>
<title>MIT flapping-wing microrobot flies outside motion capture with its own sensors and computer</title>
<link>https://dexscience.eu/ai/mit-flapping-wing-microrobot-flies-outside-motion-capture-with-its-own/</link>
<guid isPermaLink="true">https://dexscience.eu/ai/mit-flapping-wing-microrobot-flies-outside-motion-capture-with-its-own/</guid>
<pubDate>Thu, 19 Nov 2026 16:34:19 GMT</pubDate>
<category>AI &amp; Computing</category>
<description>The featherweight, microcassette-sized robot carried a 244-milligram electronics package and landed on a sunflower, though it still draws power through a tether.</description>
<media:content url="https://dexscience.eu/media/2026/11/mit-flapping-wing-microrobot-flies-outside-motion-capture-with-its-own/og.jpg" medium="image" width="1200" height="630"/>
</item>
<item>
<title>Auditors mostly fail to spot deliberately sabotaged AI safety experiments in red-team test</title>
<link>https://dexscience.eu/ai/auditors-mostly-fail-to-spot-deliberately-sabotaged-ai-safety/</link>
<guid isPermaLink="true">https://dexscience.eu/ai/auditors-mostly-fail-to-spot-deliberately-sabotaged-ai-safety/</guid>
<pubDate>Wed, 18 Nov 2026 18:34:11 GMT</pubDate>
<category>AI &amp; Computing</category>
<description>In a preprint study, human auditors did not do significantly better than chance at telling sabotaged research from honest work. Top AI models did somewhat better, within noise.</description>
<media:content url="https://dexscience.eu/media/2026/11/auditors-mostly-fail-to-spot-deliberately-sabotaged-ai-safety/og.jpg" medium="image" width="1200" height="630"/>
</item>
<item>
<title>Nature Machine Intelligence paper presents interpretable graph learning to rank cancer genes</title>
<link>https://dexscience.eu/ai/nature-machine-intelligence-paper-presents-interpretable-graph-learning/</link>
<guid isPermaLink="true">https://dexscience.eu/ai/nature-machine-intelligence-paper-presents-interpretable-graph-learning/</guid>
<pubDate>Wed, 18 Nov 2026 18:29:19 GMT</pubDate>
<category>AI &amp; Computing</category>
<description>A new peer-reviewed study describes an interpretable, distance-resolved graph learning approach for prioritizing candidate cancer genes.</description>
<media:content url="https://dexscience.eu/media/2026/11/nature-machine-intelligence-paper-presents-interpretable-graph-learning/og.jpg" medium="image" width="1200" height="630"/>
</item>
<item>
<title>Researchers describe an open, agentic AI co-scientist aimed at biomedical discovery</title>
<link>https://dexscience.eu/ai/researchers-describe-an-open-agentic-ai-co-scientist-aimed-at/</link>
<guid isPermaLink="true">https://dexscience.eu/ai/researchers-describe-an-open-agentic-ai-co-scientist-aimed-at/</guid>
<pubDate>Wed, 18 Nov 2026 18:26:18 GMT</pubDate>
<category>AI &amp; Computing</category>
<description>The Nature Medicine paper comes from a large team centered on Washington University in St. Louis and Lawrence Berkeley National Laboratory. Its findings were not in the material we had.</description>
<media:content url="https://dexscience.eu/media/2026/11/researchers-describe-an-open-agentic-ai-co-scientist-aimed-at/og.jpg" medium="image" width="1200" height="630"/>
</item>
<item>
<title>Self-updating AI text detector caught over 93 percent of disguised machine writing in tests</title>
<link>https://dexscience.eu/ai/self-updating-ai-text-detector-caught-over-93-percent-of-disguised/</link>
<guid isPermaLink="true">https://dexscience.eu/ai/self-updating-ai-text-detector-caught-over-93-percent-of-disguised/</guid>
<pubDate>Wed, 18 Nov 2026 18:22:11 GMT</pubDate>
<category>AI &amp; Computing</category>
<description>MIT and Cornell researchers let a detector keep learning from unlabeled text after release. A commercial tool caught about 27 percent of the same disguised text.</description>
<media:content url="https://dexscience.eu/media/2026/11/self-updating-ai-text-detector-caught-over-93-percent-of-disguised/og.jpg" medium="image" width="1200" height="630"/>
</item>
</channel>
</rss>
