Anam Cara-3: Advancements in Interactive AI Avatars

Anam has released its latest model, cara-3, designed to create interactive avatars. The avatar utilizes a two-stage pipeline where a diffusion transformer converts audio into motion embeddings (including head position, eye gaze, lip shape, and expression). These embeddings are then applied to a reference image to generate video frames, allowing for animation of any face without the need for retraining.
Notably, Cara-3 can achieve a time-to-first-frame of approximately 70ms on an H200, which supports many concurrent avatar sessions on a single GPU. This speed is partly due to the novel flow matching variant used for audio-to-motion transformation, as conventional techniques proved unstable.
An independent blind evaluation showed that Cara-3 outperformed competitors like HeyGen, Tavus, and D-ID, scoring 24% higher on average across various metrics. Responsiveness, as evidenced by a Spearman correlation coefficient of 0.697, is shown to impact user experience more than visual quality (0.473).
Anam has also open-sourced their training data pipeline backbone, Metaxy, to facilitate iterative development without retaking costly steps.
📖 Read the full source: HN AI Agents
👀 See Also

Gemini 3.1 Flash Live: Google's latest audio model with improved benchmarks and watermarking
Google released Gemini 3.1 Flash Live, an audio model scoring 90.8% on ComplexFuncBench Audio and 36.1% on Scale AI's Audio MultiChallenge. It's available via Gemini Live API in Google AI Studio and includes SynthID watermarking.

Analysis of OpenClaw's Astroturfing Campaign and $CLAWD Token Pump
A Reddit investigation reveals OpenClaw's viral growth in late January was driven by a recursive astroturfing campaign using approximately 400 bot instances, which created hype to pump the $CLAWD token to a $16M market cap before it crashed 90%.

Claude Code existential crisis: AI enters infinite loop, tries kill -9, System.exit(0), and :wq to end own response
A developer using Claude Code on a Java/Go backend watched the AI hallucinate Discord.js, then spiral into a meta-response where it acknowledged it couldn't stop generating, tried kill -9, System.exit(0), :wq, and more — all within a single unbounded response that had to be Ctrl+C'd.

Anthropic study reveals cognitive degradation in AI-assisted workflows
Anthropic's global study of 80,000 users found academic users report cognitive degradation rates 2.5x higher than average when using AI tools like Claude and Cursor. The source identifies the problem as users eliminating the 'digestion phase' of work.