{"version":1,"type":"story","url":"https://digestai.news/story/causal-analysis-reduces-spurious-speech-in-ai-llms","json":"https://digestai.news/story/causal-analysis-reduces-spurious-speech-in-ai-llms.json","markdown":"https://digestai.news/story/causal-analysis-reduces-spurious-speech-in-ai-llms.md","slug":"causal-analysis-reduces-spurious-speech-in-ai-llms","headline":"Causal Analysis Reduces Spurious Speech in AI LLMs","summary":"Researchers have developed a method to reduce spurious speech from full-duplex speech language models like Moshi and PersonaPlex. These models can initiate inappropriate speech during periods of silence, with Moshi initiating around 12% and PersonaPlex about 11%. The study identifies two hypotheses: either the model is selecting speech despite low probabilities or it's influenced by its previous nonspeech outputs. By analyzing these models' responses to muted input, researchers found that spurious onsets can be significantly reduced without blocking genuine user inputs. Their method requires no retraining and operates in real-time, with decision times below 61 milliseconds. This work is crucial for improving the reliability of AI assistants like Moshi and PersonaPlex.","keyPoints":["Reduced spurious speech from full-duplex LLMs","Method suppresses 13/13 Moshi and 9/9 PersonaPlex onsets","Real-time, no retraining required"],"whyItMatters":"This research enhances the reliability of AI assistants by reducing inappropriate speech during periods of silence.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["Moshi","PersonaPlex"],"models":["Moshi","PersonaPlex"],"people":[]},"firstPublishedAt":"2026-09-15T04:00:00Z","updatedAt":"2026-09-15T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.CL","title":"Causal Analysis and Mitigation of Spurious Onsets in Full-Duplex Speech LLMs","url":"https://arxiv.org/abs/2609.13445","publishedAt":"2026-09-15T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"Causal Analysis Reduces Spurious Speech in AI LLMs\", 15 September 2026, https://digestai.news/story/causal-analysis-reduces-spurious-speech-in-ai-llms","publisher":"Digest AI","title":"Causal Analysis Reduces Spurious Speech in AI LLMs","datePublished":"2026-09-15T04:00:00Z","url":"https://digestai.news/story/causal-analysis-reduces-spurious-speech-in-ai-llms"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}