{"id":16248,"date":"2025-11-19T01:25:06","date_gmt":"2025-11-19T00:25:06","guid":{"rendered":"https:\/\/haimagazine.com\/uncategorized\/is-grok-getting-softer-new-model-from-xai\/"},"modified":"2025-11-20T16:15:43","modified_gmt":"2025-11-20T15:15:43","slug":"is-grok-getting-softer-new-model-from-xai","status":"publish","type":"post","link":"https:\/\/haimagazine.com\/en\/ai-news-2\/is-grok-getting-softer-new-model-from-xai\/","title":{"rendered":"\ud83d\udd12 Is Grok getting softer? New model from xAI"},"content":{"rendered":"<p>xAI is rolling out Grok 4.1. This new model is designed to bring a more natural and engaging conversation style, greater empathy (similar to GPT 5.1) and creativity, all while keeping the responses highly credible.<\/p><h4 class=\"wp-block-heading\"><strong>What&#8217;s changed?<\/strong><\/h4><p>According to xAI, the biggest advancements are in dialogue fluidity, emotional context sensitivity, and personality coherence. The model is now better at understanding the user&#8217;s subtle intentions, while remaining honest and as helpful as possible. These features were achieved through an extensive phase of reinforcement learning.<\/p><p>The creators also promise that they&#8217;ve worked on two annoying AI features:<\/p><ul class=\"wp-block-list\"><li><strong>Lying:<\/strong> The model is designed to be more resistant to pressure and less likely to deviate from the truth, even when we try to fool it with tricky questions.<\/li>\n\n<li><strong>Nodding along:<\/strong> The bot&#8217;s habit of agreeing with everything we say just to be nice, even when we&#8217;re talking nonsense, has been toned down.<\/li><\/ul><h4 class=\"wp-block-heading\"><strong>Two weeks of quiet testing<\/strong><\/h4><p>Before the official launch was announced, xAI conducted a little experiment. From November 1 to 14, 2025, some users unknowingly chatted with the new Grok. The preliminary versions of Grok 4.1 gradually handled an increasing percentage of the actual traffic. In blind comparative tests, users preferred the responses from Grok 4.1 over the previous production version in 64.78% of cases.<\/p><p>Grok 4.1 ranks really high in popular charts like LMArena, both in the mode with extra reasoning and in the quick version without thinking.<\/p><figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"428\" src=\"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-005947-1024x428.png\" alt=\"\" class=\"wp-image-16195\" srcset=\"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-005947-1024x428.png 1024w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-005947-300x125.png 300w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-005947-768x321.png 768w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-005947-1536x641.png 1536w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-005947-600x251.png 600w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-005947.png 1590w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">Source: LMArena<\/figcaption><\/figure><p>xAI really highlights the progress in tasks that require empathy and soft skills. Internal tests have shown that Grok 4.1 gets a better grasp of what we really mean when we ask a question and can adjust the tone of the conversation accordingly.<\/p><p>This should also be evident in creative tasks. When we ask it to write a story or an email, the style should be more coherent and less chaotic. In the EQ-Bench3 benchmark, the new version scores among the highest of any publicly available models. It performs similarly well in creative writing tests (Creative Writing v3).<\/p><figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"573\" src=\"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010347-1024x573.png\" alt=\"\" class=\"wp-image-16197\" srcset=\"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010347-1024x573.png 1024w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010347-300x168.png 300w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010347-768x429.png 768w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010347-600x336.png 600w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010347.png 1123w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">Source: xAI<\/figcaption><\/figure><h4 class=\"wp-block-heading\"><strong>Fewer hallucinations<\/strong><\/h4><p>Anyone who has used AI knows that models sometimes go off track and invent a fact or two. xAI has tackled this issue head-on. By implementing additional steps in the post-training phase, Grok 4.1 is noticeably making fewer factual errors in responses to informational queries \u2014 both in internal tests and on the public FActScore benchmark.<\/p><figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"511\" src=\"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010748-1024x511.png\" alt=\"\" class=\"wp-image-16199\" srcset=\"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010748-1024x511.png 1024w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010748-300x150.png 300w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010748-768x383.png 768w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010748-600x299.png 600w, https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/11\/Zrzut-ekranu-2025-11-19-010748.png 1141w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">Source: xAI<\/figcaption><\/figure><h4 class=\"wp-block-heading\"><strong>Safer, yet still in Grok&#8217;s style<\/strong><\/h4><p>The model has undergone extensive safety testing according to the xAI Risk Management Framework. Grok is designed to refuse when we ask for something illegal (like a recipe for a dangerous substance), but it shouldn&#8217;t censor us on topics that are simply controversial or debatable.<\/p><p>So it seems xAI is attempting a tricky art: it wants to civilize the chief rebel among chatbots without killing its spirit. Time will tell whether Grok has truly mellowed or just learned to better hide its horns from the censors.<\/p><p><\/p>","protected":false},"excerpt":{"rendered":"<p>xAI has rolled out Grok 4.1, their latest language model update. The company boasts significant improvements in understanding intentions, speech coherence and error reduction, all aimed at enhancing interactions with users.<\/p>\n","protected":false},"author":230,"featured_media":12433,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_lock_modified_date":false,"footnotes":""},"categories":[813],"tags":[707,816,814],"popular":[],"difficulty-level":[36],"ppma_author":[884],"class_list":["post-16248","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news-2","tag-ai-5","tag-grok-2","tag-xai-2","difficulty-level-easy"],"acf":[],"authors":[{"term_id":884,"user_id":230,"is_guest":0,"slug":"karolina-ceron","display_name":"Karolina Cero\u0144","avatar_url":"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/07\/PXL_20250419_110132091.MP4-scaled.jpg","first_name":"Karolina","last_name":"Cero\u0144","user_url":"","job_title":"","description":"Wsp\u00f3\u0142tw\u00f3rczyni newslettera AI Flash, studentka psychologii i pasjonatka sztucznej inteligencji. Interesuj\u0119 si\u0119 wp\u0142ywem nowych technologii na cz\u0142owieka, a w wolnych chwilach eksperymentuj\u0119 z generatywn\u0105 grafik\u0105 w Midjourney."}],"_links":{"self":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16248","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/users\/230"}],"replies":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/comments?post=16248"}],"version-history":[{"count":1,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16248\/revisions"}],"predecessor-version":[{"id":16249,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16248\/revisions\/16249"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media\/12433"}],"wp:attachment":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media?parent=16248"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/categories?post=16248"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/tags?post=16248"},{"taxonomy":"popular","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/popular?post=16248"},{"taxonomy":"difficulty-level","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/difficulty-level?post=16248"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/ppma_author?post=16248"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}