Microsoft has secured a patent for an AI infrastructure that uses idle GPUs in personal computers to train AI models, according to a report from the Korean outlet BizWorld. The report describes the technology as a distributed AI infrastructure.
Anthropic has released Claude Meetings in beta. Claude can join Google Meet calls to take notes, answer questions, and create documents or whiteboards during the meeting.
A Matthew Berman video in which Jeremy, a researcher at Anthropic, explains how large language models (LLMs, AI that understands and writes text) are trained. The description does not list specific training steps or results.
Anthropic has launched Claude Meetings, which lets Claude join video calls as a visible participant. During a call, Claude can answer questions, create documents and whiteboards, and record notes and action items.
Hugging Face released a guide for training the 3D image generation model TRELLIS.2 on your own data. It includes fine-tuning scripts and ways to run them. The goal is to fix the mistakes a general model makes on specific objects.
As a scandal involving Nvidia CEO Jensen Huang spreads, US lawmakers are moving to tighten semiconductor export controls on China. Lawmakers from both parties want to pass three bills by attaching them to the National Defense Authorization Act, the annual law that sets defense budgets and policies.
Multi-vector embedding models match queries precisely by making a vector for each token, but the many vectors take up a lot of storage. This article explains how to train a model specifically for token pooling, which merges similar vectors, so that useful information survives the merge.
Anthropic has released v2.6.0 of its Python software development kit (SDK), a set of tools that lets programs use Claude. This version adds a feature for logging organization members out of all sessions and several new error codes.
Suno now lets Pro and Premier subscribers upload MIDI files to its Create page to start a song. MIDI is a file that contains only notes and timing information, without sound. Users can follow these notes as they are, develop them further, or change the instrumentation.
Amazon AWS Machine Learning blog · Nov 5, 01:33 KST
As AI agents handle more tasks, deciding where people should still check their work has become important. Smart Approvals in Amazon Quick automatically lets through routine actions that are allowed, and sends only actions that need judgment back to the user. Permissions and policies stay the same.
AI Office is an app that lets people use Claude Code and Codex without knowing how to code. Instead of a black terminal, you hire AI staff and assign them work, much like delegating tasks to people at a company. Important tasks need approval first.
Fyxer is an AI assistant that uses OpenAI's models to organize inboxes and write email drafts in the user's own tone. The company said it combines fine-tuning (further training a model on specific data), memory features, and real user feedback.