● Covered by 2 sources · 2 reportsLow impact2 neutral
Block unveils Dessa, an AI model trained solely on human conversation
Key points
- Block released Dessa, an AI model trained from scratch on human conversation.
- Dessa is a speech-to-speech model designed to run locally.
- The model aims to improve AI understanding of human timing and emotion.
- Early access will be provided to a small group of AI safety researchers.
- Dessa was trained on a single NVIDIA GPU rack.
- Dessa is a full-duplex model, capable of simultaneous speaking and listening.
- Dessa's knowledge stems from its conversational data.
- Emotion is built directly into Dessa's model.
Block Introduces Dessa AI Model
Jack Dorsey’s Block has announced Dessa, an artificial intelligence model developed by training it exclusively on human conversation. This approach differs from models that rely on written text, aiming to replicate how humans learn speech.
The company shared examples of Dessa's interactions, which sometimes exhibit awkward or uncanny characteristics, such as mistaking a speaker for someone else or going on unexpected tangents during conversations.
Technical Capabilities and Design
Dessa is described as a true speech-to-speech model, meaning all its knowledge originates from spoken human language. It is designed to operate locally on devices, specifically noted to run on a MacBook Pro M4 Max or better hardware.
The development team stated their motivations for creating Dessa were curiosity, enhancing understanding between humans and AI, and improving safety in AI interactions.
Addressing Current AI Limitations
Block's team highlighted that current voice AI often converts speech to text, losing nuances like human timing and emotion. Dessa seeks to address this by maintaining the richness of spoken interaction.
The model's focus on understanding emotional cues in voices is intended to prevent situations where AI might fail to recognize a user's distress, thereby improving safety in AI-human communication.
Future Access and Research
Block plans to provide early access to Dessa to a select group of AI safety researchers. This step suggests an emphasis on evaluating and refining the model's safety aspects and its ability to interpret human communication effectively.
Updates
- Dessa was trained on a single NVIDIA GPU rack.
- Dessa is a full-duplex model, capable of simultaneous speaking and listening.
- Dessa's knowledge stems from its conversational data.
- Emotion is built directly into Dessa's model.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
How outlets covered it
Block's research lab released Dessa, a new voice model trained on human speech that runs locally and exhibits human-like expressions including laughs, pauses, and cries. Dessa is a full-duplex model, capable of simultaneous speaking and listening, and aims to provide more natural interactions by building emotion directly into its speech generation.
Block, led by Jack Dorsey, announced Dessa, an AI model trained exclusively on human conversation to improve speech-to-speech interaction. The model aims to understand human timing and emotion better, running locally on devices like the MacBook Pro M4 Max.