Archive — US — techcrunch
Sesame, the startup behind the viral virtual assistant Maya, releases its baseAI model
Sesame has launched its baseAI model, CSM-1B, enabling developers to create advanced voice assistants like Maya. With 1 billion parameters and under the Apache 2.0 license, CSM-1B generates RVQ audio codes but lacks safeguards against misuse. The model's effectiveness in non-English languages is uncertain, and it poses risks similar to other AI voice cloning tools. Sesame aims to revolutionize voice tech and is developing AI glasses.
Key points
- Sesame's new model CSM-1B has 1 billion parameters and is open-sourced under Apache 2.0 for commercial use.
- CSM-1B generates RVQ audio codes from text and audio inputs, using Meta’s Llama framework.
- The model’s capabilities for generating various voices are not fine-tuned and could produce poor results in non-English languages.
- There are no effective safeguards in place against misuse of the technology.
- Realistic voice generation can lead to potential misuse, such as voice cloning without consent.
- Sesame was co-founded by Brendan Iribe and gained attention for its advanced assistant tech, Maya.
- The company has secured funding from notable investors and is also working on AI glasses.
Why it matters
- The rise of AI voice assistants reflects the growing trend of personal AI tools in consumer technology.
- This model aligns with the broader shift towards more realistic and interactive AI systems, mirroring trends in the tech industry through 2025.
- Concerns over ethical use and voice cloning risks relate to ongoing discussions in AI governance and regulation.
- The competitive landscape includes major tech players like Google and Meta, pushing advancements in audio technology.
Sesame’s release of CSM-1B marks a significant step in voice AI technology, but the associated risks underscore the urgent need for ethical guidelines and safeguards in its deployment.