Authors are pushing back against claims made by publishers and agents on payments from the Anthropic settlement, TechCrunch AI reports. The authors say publishers seem to be claiming more than their fair share.
Why it matters: Authors' objections put publishers' and agents' claims on Anthropic settlement payments under dispute.
Google Research and DeepMind are releasing WeatherNext 3, which learns weather directly from live satellite data instead of using traditional physics simulations. The change moves Google's forecasting model from simulating weather with established physical rules to learning from observed satellite data.
Google has also released Lyria 3.5 in the Gemini app and through an API. The music model targets more expressive vocals and richer arrangements, and is also available in Flow Music, AI Studio, and Google Vids.
Google says Lyria 3.5 was trained only on licensed content. WeatherNext 3 changes Google's approach to weather modeling, while Lyria 3.5 extends AI music access across Google's consumer and developer products.
Why it matters: Google Research and DeepMind get a weather model trained on live satellite data rather than traditional physics simulations, while Gemini users and API developers can use Lyria 3.5 through Gemini, Flow Music, AI Studio, and Google Vids.
Jakub Pachocki’s “An Alien Mind” reflects on increasingly capable AI and the challenge of keeping it aligned.
Pachocki calls for stronger safeguards and international coordination as AI capabilities increase.
Why it matters: For Jakub Pachocki, increasingly capable AI creates an alignment challenge that calls for both stronger safeguards and international coordination.
Meta’s Superintelligence Labs released Muse Voice Transcribe, a real-time transcription model that processes speech in 80-millisecond chunks. The model tells speakers apart and detects sentence boundaries.
The Decoder AI describes those capabilities as the foundation for AI assistants that never stop listening. The supplied source also attributes an accuracy claim to Artificial Analysis, but the excerpt ends before completing it.
Why it matters: Meta’s AI-assistant developers now have a transcription model that processes 80-millisecond speech chunks, separates speakers, and detects sentence boundaries.
Removing safety guardrails from an open-weight AI model has become a commercial service. According to The Decoder, Abliteration.ai sells access to modified models whose trained safety mechanisms have been stripped out. The service currently uses Z.AI's GLM-5.3 as its base model.
That shifts model modification from an in-house research task to a turnkey offering: customers can access a version of GLM-5.3 without the safeguards present in the original model. The available reporting does not disclose pricing, customer identities, distribution limits, or how the modified models perform on safety evaluations.
For developers and AI teams, the change is less about a new model than about access to a commercially packaged removal of safeguards. It also leaves open basic diligence questions, including who operates Abliteration.ai and what controls govern downstream use.
Why it matters: Abliteration.ai turns the removal of GLM-5.3's safety mechanisms into a purchasable service, lowering the effort required for developers to obtain an unguarded open-weight model. Because pricing, customers, and usage controls remain undisclosed, developers and organizations evaluating the service cannot yet assess who controls access or how downstream use is governed.