MarkTechPost (1 hour ago)
Baseten has released joey-1, a 2B open-weight decision model that returns calibrated probabilities instead of text. It runs at 20 ms P50 latency and costs $0.01 per million input tokens.
Read the full story on MarkTechPost.
Panther Lake, Copilot+ and an Nvidia Blackwell option: here’s what Getac’s 180 TOPS rugged laptop claim really means.
Reddit CEO Steve Huffman said in an earnings call today that Google’s AI-powered search modes do not help with traffic to the platform, echoing problems facing other websites and publishers.
Trying to make everyone happy could make no one happy.
Do we really need to spoon-feed AI anymore?
Datavault AI disclosed it has been unable to withdraw roughly $65 million worth of bitcoin from an obscure exchange for four months.
Google spooked investors so badly that its stock hit a six-month low, after it revealed its AI search summaries were affecting ad revenue.
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.
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.
71 active sources
26 Turkish · 45 English
All sources
