AssemblyAI TTS-1 is now available on LiveKit Inference, delivering first audio in about 90 ms. It offers over 4,000 voices, supports six languages, includes word timestamps, and costs $15 per 1M characters.
Key points
Latency: first audio in about 90 ms
Voices: more than 4,000 available
Sources
Moderator on-device NSFW model launches with 87.8% recall
Moderator is a 9.7MB on-device NSFW detection model that runs on phones before upload. It flags 6.3% of safe images and catches 87.8% of NSFW images. The model uses five detection heads and can be added via Swift, Kotlin or JavaScript SDKs.
Key points
False-block rate: 6.3% of safe images.
Recall: 87.8% of NSFW images.
Sources
Playwright browser security reminder – no bundled browsers, pinned test builds
Playwright does not bundle browsers; MCP or CLI use the stable Chrome/Edge on your machine with normal updates. The test runner downloads Chrome for Testing, pinned to a specific build, and runs with the sandbox off by default. The library API also launches the pinned build with sandbox off unless overridden.
Key points
MCP or CLI use the stable Chrome/Edge on your machine, receiving normal updates, no config edit needed
npx playwright install downloads Chrome for Testing, pinned to the build tested against, runs with sandbox off by default
Sources
Cencori builds inference infrastructure for Africa
Cencori will build energy-efficient inference infrastructure for Africa, from AI platforms to silicon, including its own ENX chips. It aims to run models on African soil and reduce power use.
Key points
Commitment to build energy-efficient inference infrastructure for Africa.
Plan to create own inference chips called ENX.
Sources
- Original post
- cencori.com - Linked in the post
SurrealDB 4 launch with SurrealDB Earth serverless
SurrealDB 4 launch with SurrealDB Earth serverless, a Cloudflare-based database that scales to any number of agents and users. Start free on Earth, no card needed.
Key points
SurrealDB 4 is released.
SurrealDB Earth is serverless, built on Cloudflare, and scales to as many agents and users as needed.
Sources
- Original post
- sdb.li - Linked in the post
1.3M to independents – payment from platform to creators
$1.3M to independents is announced. The post thanks creators and partners and says more is coming.
Key points
$1.3M to independents for doing what they do best
Thanks to every creator and partners, more to come
Sources
Cencori build energy-efficient inference infrastructure for Africa
Cencori announced it will build energy-efficient inference infrastructure for Africa, covering the platform AI agents run on down to silicon. Most of Africa's AI runs on computers thousands of miles away, which is expensive and slow.
Key points
Commitment: Cencori will build energy-efficient inference infrastructure for Africa, from platform to silicon.
Current situation: Most of Africa's AI runs on computers thousands of miles away, making it expensive and slow.
Sources
J2.3 model upgrade and comparison to Nvidia FLORA
J2.3 is the latest multimodal Bayesian model and is being compared to Nvidia's FLORA. The team plans to retest two straggler regions when J2.4 is released.
Key points
J2.3 compared to Nvidia's FLORA.
Retest of two straggler regions planned for J2.4.
