Speech & audioTechnical note
Teaching a model how to speak
A closer look at conversational speech fine-tuning with CSM and LoRA. The real question is what happens after the loss curve.
Back to the journalConversations / Field notes
ConversationsConversation
The human side of artificial intelligence
A conversation on personalization, privacy, and human–AI interaction.
What makes an AI interaction feel personal? This Vyvo Voices episode explores that question through a conversation with Hakan Kozaklı.
Context beyond a single prompt
Published on September 26, 2025, the episode covers human–AI interaction, personalization, privacy, and emotional context in Vyvo AI. The complete conversation is hosted by vLife Digital.
A useful lens for listening is to separate three questions: what the person is asking, which context is relevant, and which context the system is allowed to use.
01 / An interaction lensEditorial interpretation
- IntentUnderstandWhat is being asked?
- ContextSelectWhat is relevant?
- ControlRespectWhat is allowed?
Relevance and restraint
More remembered information does not automatically make a response more useful. The engineering question is which details help with the present task, and how a person can understand and control their use.
Watch the full conversation
This page is an illustrated introduction to the episode. Follow the original publication for the complete conversation and its context.
Vyvo Voices / 26 September 2025Open the original episodeSources & further reading
Language modelsTechnical note
From fluent to useful: a note on DPO
Fluent language is a starting point. Preference optimization asks a more specific question: which response is actually better?