{"id":36759,"date":"2025-11-11T02:00:00","date_gmt":"2025-11-11T01:00:00","guid":{"rendered":"https:\/\/www.gnews.cz\/?p=36759"},"modified":"2025-11-11T12:50:02","modified_gmt":"2025-11-11T11:50:02","slug":"ais-new-no-thinking-model-beats-gpt-5-security-risks-mcp-and-california-introduce-groundbreaking-laws","status":"publish","type":"post","link":"https:\/\/gnews.cz\/en\/ekonomika\/ai-novy-model-bez-mysleni-porazi-gpt-5-bezpecnostni-rizika-mcp-a-kalifornie-zavadi-prelomove-zakony\/","title":{"rendered":"AI: New model without \u201ethinking\u201c beats GPT-5, MCP safety risks, and California introduces groundbreaking laws"},"content":{"rendered":"<p>Recently, several major events in the world of artificial intelligence have come together - from the release of a powerful open model, to warnings about the security risks of a popular protocol, to the adoption of comprehensive regulation in California.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Ling-1T: A revolution in training, not architecture<\/strong><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Ant Group has introduced&nbsp;<strong>giant open model Ling-1T<\/strong>&nbsp;with one trillion parameters.<\/li>\n\n\n\n<li>Unlike the competition&nbsp;<strong>does not need a special \u201ethinking mode\u201c<\/strong>. His ability to reason is directly built into the immediate response through intensive training on data containing chains of reasoning.<\/li>\n\n\n\n<li>In the tests&nbsp;<strong>outperformed models such as the GPT-5 and Gemini 2.5 Pro<\/strong>&nbsp;in 22 out of 31 benchmarks, particularly in mathematics and reasoning.<\/li>\n\n\n\n<li>The model is&nbsp;<strong>freely downloadable<\/strong>&nbsp;under the MIT license, which blurs the distinction between open and commercial models.<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.gnews.cz\/wp-content\/uploads\/2025\/11\/3.1-\u5c4f\u5e55\u622a\u56fe-2025-11-06-124920-1024x577.png\" alt=\"\" class=\"wp-image-36761\"\/><\/figure>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>&nbsp;<\/strong><\/h4>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Expert Warning: the MCP protocol hides serious security risks<\/strong><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>A new study highlights&nbsp;<strong>exponential increase in security threats<\/strong>&nbsp;for systems using the Model Context Protocol (MCP).<\/li>\n\n\n\n<li>The problem is in the so-called.&nbsp;<strong>compositional risk<\/strong>: the vulnerability increases rapidly with each added MCP server. Just 2 servers means 36% risk, with 10 servers it approaches 92 %.<\/li>\n\n\n\n<li>Attackers can exploit servers that receive data from untrusted sources to&nbsp;<strong>execution of unauthorised orders<\/strong>, such as running code.<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.gnews.cz\/wp-content\/uploads\/2025\/11\/3.2-\u5c4f\u5e55\u622a\u56fe-2025-11-06-125018-1024x574.png\" alt=\"\" class=\"wp-image-36762\"\/><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>California shows the way: first comprehensive AI regulation in the US<\/strong><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>In response to the absence of federal laws, California has adopted a package&nbsp;<strong>four groundbreaking regulations<\/strong>.<\/li>\n\n\n\n<li><strong>SB 53<\/strong>&nbsp;obliges the creators of the most powerful models to publish safety protocols.<\/li>\n\n\n\n<li><strong>SB 243<\/strong>&nbsp;protects minors from the harmful effects of chatbots.<\/li>\n\n\n\n<li><strong>AB 316<\/strong>&nbsp;provides that&nbsp;<strong>legal liability for damage is always borne by the company<\/strong>, not autonomous AI.<\/li>\n\n\n\n<li><strong>AB 853<\/strong>&nbsp;ordered by&nbsp;<strong>clear labelling of AI-generated content<\/strong>.<\/li>\n\n\n\n<li>The laws have provoked mixed reactions, with some firms welcoming them and others warning of a fragmentation of the rules.<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.gnews.cz\/wp-content\/uploads\/2025\/11\/3.3-\u5c4f\u5e55\u622a\u56fe-2025-11-06-125242-1024x576.png\" alt=\"\" class=\"wp-image-36763\"\/><\/figure>\n\n\n\n<h4 class=\"wp-block-heading\"><\/h4>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Scientists discover a more efficient way: better prompts replace expensive training<\/strong><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>A team of researchers has presented an algorithm&nbsp;<strong>GEPA<\/strong>, which greatly enhances the AI agent's abilities. The GEPA algorithm automatically improves AI agent prompts through iterative evolution. LLM analyzes agent failures and then revises prompts to correct those specific errors. It iteratively tests and selects the best performing prompts, improving performance more efficiently than re-learning the model itself.<\/li>\n\n\n\n<li>Instead of demanding fine-tuning of the model&nbsp;<strong>automatically generates and optimizes instructions<\/strong>&nbsp;(prompts) based on agent error analysis.<\/li>\n\n\n\n<li>This method has achieved&nbsp;<strong>better results than traditional learning<\/strong>&nbsp;and at the same time she was&nbsp;<strong>up to 35 times more economical<\/strong>&nbsp;on computing resources.<\/li>\n\n\n\n<li>It is the ideal solution for situations with limited data or computing power.<\/li>\n<\/ul>\n\n\n\n<p><\/p>\n\n\n\n<p><strong><a href=\"https:\/\/www.deeplearning.ai\/the-batch\/tag\/data-points\" target=\"_blank\" rel=\"noopener\" title=\"\">Data Points<\/a> - DeepLearning.AI by Andrew Ng \/ gnews.cz - GH<\/strong><\/p>","protected":false},"excerpt":{"rendered":"<p>Several major events in the world of artificial intelligence have come together recently - from the release of a powerful open model, to warnings about security...<\/p>","protected":false},"author":3,"featured_media":36760,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_lmt_disableupdate":"","_lmt_disable":"","footnotes":""},"categories":[7144,4,3711,2973],"tags":[122,7133,7152,7147,7145,4496,7132,7135,7134,7148,7146,38,41],"class_list":["post-36759","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-ekonomika","category-ekonomika","category-english","category-zahranicni","tag-ai","tag-ant-group","tag-data-points","tag-gepa-algoritmus","tag-kalifornske-zakony-o-ai","tag-kyberneticka-bezpecnost","tag-ling-1t","tag-mixture-of-experts-moe","tag-open-source","tag-optimalizace-promptu","tag-sb-53","tag-usa","tag-world"],"acf":[],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/posts\/36759","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/comments?post=36759"}],"version-history":[{"count":2,"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/posts\/36759\/revisions"}],"predecessor-version":[{"id":37055,"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/posts\/36759\/revisions\/37055"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/media\/36760"}],"wp:attachment":[{"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/media?parent=36759"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/categories?post=36759"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/gnews.cz\/en\/wp-json\/wp\/v2\/tags?post=36759"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}