On Device Chat AI Best Friend

版本 26.3

AXIØM.talk app version 26.3 “Trust & Safety Agent” update includes the following New Features! – a “Trust & Safety” Agent now double-checks both the immediate output as well as the “tone” of the conversation, and recommends safety improvements if needed – this is informed by an internal ethics prompt I have called the “AXIØM Principles” inspired by Anthropic’s published “Constitution” AI research & philosophical approach – You can read more about Ax’ Model Instructions/System Prompts at http://transparency.axiom.talk – Note: this feature has been optimized for performance, but some additional latency to the AI responses can occur; depending on the nature/tone of the conversation – the app can now identify those who are over/under 13 years old, or who do not declate their age-range, and is stricter around safety with Increased Guardrails for this younger/undeclared cohort – This is done in a privacy-preserving way via official Apple® APIs for this purpose, you can read more at https://support.apple.com/guide/iphone/iphe56aab53d/ios – to protect a User’s Perspective, a new Duration Warning will slowly appear the longer a session continues – this warning cannot be dismissed either directly or by relaunching the app, it must “cool down” for the same amount of time it took to appear; this is for the safety of all Users regardless of Age – Added Preview buttons to the Voice-Mode onboarding screen for new users It also includes the following Bug Fixes: – corrects a potential issue with the Foundation Model returning inaccurate languages for your locale due to the “System Prompt” having a small malformation – You can read more about Ax’ Model Instructions/System Prompts at http://transparency.axiom.talk