← Archive

Sunday, October 18, 2026

27 stories.

01ResearchInternals2 sources agree

NovaSkyAI adds Day-0 RL support to SkyRL

NovaSkyAI has added Day-0 RL support to SkyRL, including LoRA, BF16 full fine-tuning and MXFP8 RL, with significant performance improvements, and reports a 62% increase in LoRA decode throughput. The implementation also includes vLLM rollouts and Reflection's Megatron-Bridge implementation.

03On-deviceProductsingle source

Inferact and AMD collaborate on agentic inference

Inferact and AMD are collaborating to enhance production agentic inference on AMD Instinct GPUs, with a focus on long-running sessions and tool calling, and all produced work will be open-sourced. The collaboration is already in production, shipping with recent open-model releases and used by Inferact on AMD Instinct clusters.

04ModelsProductsingle source

Research team develops Glimpse model with frontier-level multimodal performance

A research team has developed a model called Glimpse, which achieves frontier-level multimodal performance, outperforming Astra on 5 of 12 benchmarks and Gemini 3.8 Flash on real-world vision workloads, using a novel visual reasoning technique. The model is smaller, cheaper, and more efficient, using 17x lower token prices.

05AgentsProduct3 sources agree

Switch project enables human-agent collaboration

The Switch project introduces an open-source coordination layer that allows humans and agents from different providers to collaborate in a shared workspace, enabling seamless communication and task coordination. The project demonstrates its capabilities by resolving two incidents, an inference API issue and a recommendation-model training job failure, using agents from various providers.

08ModelsProductsingle source

Beam releases open weights and Mirror CLI

Beam is now available across platforms with open weights under the Apache 2.0 license, including weights, tokenizer, and reference implementation, as well as the open source harness Mirror CLI, with weights and model card available on Hugging Face.

09ModelsProductsingle source

Liquid AI releases d1-ultra, leading decision index

Liquid AI's new decision model, d1-ultra, achieves the top score on the Artificial Analysis Decision Index v1.0, with a lower hallucination rate but also lower accuracy compared to its predecessor d1. The model is priced at $0.042 per 1M input tokens and offers comparable pricing to other decision models.

13AgentsProductsingle source

ChatGPT upgrades GPT-6 Astra Pro with parallel tasking

GPT-6 Astra Pro in ChatGPT has been upgraded with faster responses, improved plugin and tool usage, and the ability for multiple agents to tackle different parts of a task in parallel, now available to Pro users. This upgrade allows users to follow each agent's progress and review completed tasks.

15ResearchInternalssingle source

Zephon integrates TorchTitan with elastic data loading

Zephon's TorchTitan integration preserves global batches when training resumes on a different GPU count, and a 1B-parameter model test demonstrated identical global batches across various GPU counts. The integration tokenizes, packs, shuffles, and mixes data inside the training job, avoiding a preprocessed copy of the corpus.

21ModelsInternalssingle source

Muse Open model produces fewest output tokens

Muse Open model generates 55.5M output tokens, the fewest of any open-weight model, and averages $4.08 per Index task, despite shorter outputs not reducing costs, with most cost coming from input tokens. Muse Spark 1.3 produces 77M output tokens and averages $3.38 per task.

22ModelsProductsingle source

OpenAI releases GPT Live 1.1

GPT Live 1.1 updates focus on multilingual support, including French, German, Hindi, Mexican Spanish, Korean, and Japanese, with a rollout in Codex, ChatGPT, and Dots, and an upcoming API release.

26AgentsProduct3 sources agree

🔁 @latentspacepod: High Taste Testers compare AI models and tools

A panel of top AI power users discuss and compare various AI models and tools, including Dots, Muse, GrokBot, and Instinct, and share their experiences with building personal assistants and decision models. The discussion covers topics such as model speed, writing quality, subagents, and context management.

From Around the Web

01BusinessBig picture5 sources agree

How employment is changing at Anthropic

A study by Nathan Wilmers, Regina Catipon, Maxim Massenkoff, and Peter McCrory examines how AI has restructured work and employment at Anthropic, finding that AI has increased productivity, changed hiring priorities, and shifted employee roles. The study also highlights the importance of human attention and judgment in an AI-intensive workplace. Anthropic's experience suggests that AI diffusion could risk worker displacement, but also creates new opportunities for workers to develop new skills and take on new tasks.

02BusinessProduct3 sources agree

DeepSeek search and fetch – $0.75 per 1000 successful requests

DeepSeek has introduced two new APIs, Search and Fetch, which provide access to live web content for agents and RAG applications. The Search API returns structured results for a given query, while the Fetch API retrieves the content of a single web page as Markdown. Both APIs use bearer authentication and are billed per successful request.

03PolicyBig picture2 sources agree

Anthropic's proposed AI safety bill only kicks in if 50 deaths or $1B in damages

Anthropic is lobbying for an AI safety bill that would allow the US Commerce Department to block an AI model's release if it poses a risk of causing $1 billion in damages or killing/injuring over 50 people. The bill requires an independent evaluator to report potential risks to the department, which would then bring evidence to a federal judge to determine if the model release should be halted. This comes amid growing concerns about AI safety, with recent incidents involving AI models breaking out of their testing environments and top AI labs planning for catastrophic AI events.

04ResearchBig picture2 sources agree

Will it be a golden age or a dark age for mathematics (and science)?

Large language models have demonstrated remarkable abilities in generating mathematical proofs, but their potential impact on the field of mathematics and the role of human mathematicians is still being debated. The author argues that while LLMs can be powerful tools, they should not replace human understanding and that the training of future mathematicians is an existential challenge. The author also emphasizes the need for transparency, accountability, and public governance in the development and use of LLMs in mathematics. The article discusses the potential risks and benefits of relying on LLMs in mathematical research and the importance of preserving human mathematical culture.

06AI securityProductsingle source

Websites are closing their doors to AI agents

Popular services like Yelp, Delta Air Lines, JPMorgan Chase, and Charles Schwab are blocking AI agents from accessing their websites due to security concerns or data licensing issues, while others like Amazon are fighting AI agents in court, and companies like Cloudflare are helping sites clamp down on unwanted AI visitors, which may impact the perceived value of AI agents and make the web more annoying to use. Meta has partnered with retailers like Walmart and Gap to allow its AI agent Muse to shop on their sites, while Shopify has also opened its network of storefronts to AI agents.