MarkTechPost (2 hours ago)
Prime Intellect released Prime Mega MoE, a backward pass for DeepSeek’s fused Mega MoE kernel. It makes the kernel trainable and speeds Qwen3 MoE training steps by up to 34% in prime-rl.
Read the full story on MarkTechPost.
“Bowen’s death is what happens when companies race one another and ship a product faster than they can control it,” the family’s lawyers said Tuesday.
CEO Jensen Huang may have been the target of Chinese espionage activity, The New York Times reports.
The company is starting with tests outside of the US.
Accessing sensitive data and making important decisions, AI agents are acting beyond their authority and putting organizations at risk.
Adding sponsored content in more markets could make the feature profitable.
Anthropic is the “Customer B” that made up 34.5% of SpaceX’s third-quarter revenue, Bloomberg reports, overtaking the US government as its biggest client. Anthropic could spend up to $120.5B on SpaceX computing through 2029, but can cancel after early 2027.
Infini-AI-Lab released Hermit, an Apache-2.0 harness adapter framework that trains an external layer of instructions, skills, memory and tools from task feedback, leaving the agent’s model and harness unchanged.
Block’s AI lab released Dessa, a full-duplex speech-to-speech model trained only on human conversation audio. It runs locally on a MacBook Pro M4 Max, but Block has published no size, benchmarks or weights.
AssemblyAI has released AssemblyAI-TTS-1, its first text-to-speech model for voice agents. It costs $15 per 1M characters, offers unlimited concurrency without a contract, and supports 6 languages at launch.
NVIDIA released Nemotron 3.5 Transcribe, an open-weights speech LLM with 4.3B active parameters. It transcribes 40 languages, adds speaker labels and word timestamps, and posts a 4.71% mean WER on the English Open ASR Leaderboard.
71 active sources
26 Turkish · 45 English
All sources
