📅 AI Industry Digest — October 29, 2025
Model Updates & Releases
#### Cursor 2.0 Released with Composer-1 Agent Model and New UI Cursor 2.0 ships with the Composer-1 coding agent model (4x faster), a new Agents tab for managing multiple agents, plus a built-in browser, voice-to-code, and automated code review.
> Links: Cursor announcement tweet | Cursor blog
#### OpenAI Releases gpt-oss-safeguard Open-Weight Models OpenAI launched gpt-oss-safeguard 20B/120B models for policy-based safety classification, fine-tuned from gpt-oss under the Apache 2.0 license. They run on Ollama, LM Studio, and other inference stacks. The ROOST community was also launched.
> Links: OpenAI announcement | Ollama support
#### MiniMax M2 MoE Model Ranks Global Top 5 MiniMax released a 230B-parameter M2 MoE model with only 10B active parameters, outperforming M1 and Claude Opus 4.1. It excels at long-horizon tool use (shell, browser, MCP) and supports Cursor, Claude Code, and more.
> Link: MiniMax announcement
#### Windsurf Launches Falcon Alpha Model Windsurf's Falcon Alpha is a high-performance agent model optimized for speed; the team is seeking user feedback. Cascade now supports Jupyter Notebooks across all models.
> Link: Windsurf announcement
#### Modular Mojo MAX Adds HuggingFace and Torchvision Support
Modular Mojo MAX now supports HuggingFace/Torchvision models via torch_max_backend.export_to_max_graph (e.g., VGG11), with GPU execution.
> Link: Modular announcement
---
Tools & Frameworks
#### Aider-CE Adds Navigator Mode and Community RAG
Aider-CE introduced Navigator mode; community PRs added RAG (Retrieval Augmented Generation) and GitHub Copilot model support (requires the github_copilot/ prefix), plus a directory-switching bug fix.
#### DSPy Over LangChain for Structured Task Optimization Users report DSPy is better suited for structured tasks (e.g., chat), optimizing prompts and avoiding LangChain's prompt-pattern breakage across model upgrades; one team fully migrated to DSPy.
#### GitHub Integrates MCP Registry for Unified Server Discovery GitHub plans to integrate the MCP Registry: developers can self-publish MCP servers to the OSS community registry and sync to the GitHub MCP Registry (currently 44 servers).
> Link: GitHub MCP Registry
---
Community Discussions
#### Manus.im Users Complain About Fast Credit Burn, Switching to Claude Users report heavy credit consumption (e.g., 5,600 credits in 3 hours building an Android IRC app) and find Anthropic Claude better value ($20 completed 3 projects), asking for free alternatives.
#### Eleuther Community Discusses Petals Progress, Shift to LlamaCPP RPC The Petals project (Llama 70B) lost momentum as it couldn't keep up with new architectures; LlamaCPP RPC is the current alternative. Schmidhuber's HGM model code was also discussed.
> Link: Petals GitHub
#### Nous Research: API Parameter Removal and Inference Rule Changes Users discuss GPT-5 and Anthropic Claude removing temperature/top_p parameters — speculated to prevent probability leakage or reflect reasoning-model design — with some pushback.
> Link: Anthropic migration docs
---
API & Service Updates
#### OpenAI Quietly Raises Sora Browser Rate Limits and Speed Users report improved Sora app browser rate limits and generation speed, though some say limits are unchanged and quality is stable.
> Link: OpenAI Sora
#### GitHub Copilot Offers Unlimited RAG and Multi-Model Access The $10/month GitHub Copilot subscription includes unlimited RAG, gpt-5-mini, gpt4.1, grok-code-1-fast, and free embedding models.
> Link: GitHub Copilot docs
#### Anthropic Claude Restricts Parameters to temperature OR top_p Claude's API now accepts only one of temperature or top_p, not both, affecting developer prompt tuning.
> Link: Anthropic announcement
---
Research & Papers
#### KBLaM vs RAG: Compressed Knowledge vs Raw Content Community debate on KBLaM (knowledge compression via embeddings, risk of quality loss) versus traditional RAG (preserves raw content, no approximation error, higher storage cost).
> Link: KBLaM paper
#### Schmidhuber Releases HGM Model Code and Paper Jürgen Schmidhuber's HGM model code and paper are now public, drawing community interest in its geometric structure and relation to intelligence in representation learning.
> Links: HGM GitHub | HGM paper
#### Grokking and Representation Learning Discussion of the paper *Towards Understanding Grokking: An Effective Theory of Representation Learning*, on models suddenly generalizing late in training.
> Link: Grokking paper
---
Events
#### Nextdata Hosts Data 3.0 Virtual Event, Launches Nextdata OS CEO Zhamak Dehghani discusses delivering AI-ready data products with agentic co-pilots, unifying multimodal data, and replacing manual orchestration with self-managing data products.
> Link: Event registration
#### DSPy Meetup in San Francisco on November 18 The DSPy community will hold a San Francisco meetup on framework progress and future direction; multiple members have confirmed attendance.
> Link: DSPy meetup
---
*Source: Easy AI teaching project*