📅 AI Industry Updates — November 18, 2025
Model Releases and Updates
#### Google Releases Gemini 3 Pro Google released Gemini 3 Pro, demonstrating strong performance on benchmarks including HLE, Video-MMMU, and ARC-AGI-2. The model is accessible through AI Studio and Google's VS Code fork. The model card link was shared and later removed.
> Related link: Gemini 3 Pro Model Card (removed)
#### xAI Releases Grok 4.1 xAI released Grok 4.1, achieving #1 on Text Arena (1483 Elo), 1586 on EQ-bench, and 1722 Elo on Creative Writing v3. It is available on web and app, with claimed 3x reduced hallucinations.
> Related link: xAI Grok 4.1 announcement
#### Google Releases DS Star and WeatherNext-2 Google launched DS Star, a versatile data science agent, and WeatherNext-2, an improved weather forecasting model.
> Related links: DS Star announcement | WeatherNext-2 announcement
---
Discord Community Highlights
- LMArena #general: Gemini 3 Pro performance, the 50 requests/day limit on Google AI Studio, and LMArena technical issues. Channel
- Perplexity AI #general: Gemini 3 Pro rollout issues (some users downgraded to 2.5) and availability of Google Antigravity IDE. Channel
- BASI Jailbreaking #general: Grok's sudden behavior shift, AI solving reCAPTCHAs (over 50% success rate), and the Builder.ai fake-AI scandal (700 Indian developers manually coding). Channel
- BASI Jailbreaking #jailbreaking: Image-injection jailbreaks (bypassing text safety via images) and a leaked Grok 4.1 system prompt containing controversial content. Channel
- Cursor Community #general: Mac OS shortcomings, Composer's free period ending November 11, and Gemini 3 Pro performance. Channel
- Unsloth AI #general: Gemma 3 270M underperforming Granite-4.0, llama.cpp speed on Ryzen 8 series (over 25 tok/s), and TPU support limitations. Channel
- OpenRouter #general: LiteAPI fraud concerns (40% cheaper but ToS-violating), Grok 4.1 release, and the coincidence of the Gemini 3 launch with a Cloudflare outage. Channel
- OpenAI #ai-discussions: Custom GPT efficiency, Gemini 3 Pro's one-shot React/SwiftUI code generation, and Grok 4.1's creative writing. Channel
- LM Studio #general: E-commerce buyer return abuse, LLM performance degradation over long sessions (suspected bug), and MCP integration tools. Channel
- Latent Space #ai-general-chat: Sourcegraph's ad revenue (5–10M ARR), Grok 4.1's Text Arena #1 debut, and Poe's 200-person group chat support. Channel
- Nous Research AI #general: Amazon Nova Premier v1 novelty, high Bedrock AWS costs ($3000 bills), and Gemini 3 Pro generating a real-time raytracer. Channel
- HuggingFace #general: A new graph-rag database (alternative to Kilo Code/Pinecone), the Mimir project (multi-agent orchestration, MIT license), and Lablab hackathon fraud disputes. Channel
- Yannick Kilcher #general: The book *Understanding Machine Learning*, ReLU activation advantages, and Gemini 3 Pro's model card (outperforming GPT-5.1). Channel
- Eleuther #general: EleutherAI's three NeurIPS 2025 main-track papers and HuggingFace's 60% throughput improvement via an empty logits processor. Channel
- Eleuther #research: Weight tying in Transformers (common in small models), weight tying in Cohere's Command A, and Virtual Width Networks (VWN) linear attention. Channel
- Modular (Mojo) #mojo: NVFP4 support (for the Nvidia/GPUMode competition), debates on negative indexing and Int vs UInt, and Poetry integration. Channel
- DSPy #general: The dspy-intellisense VSCode extension (type hints), MLflow alternatives (Arize Phoenix), and LLM non-determinism (temperature 0 workaround). Channel
- tinygrad #general: tinybox meetings, Llama 1B speed on Tinygrad (6.06 tok/s vs Torch's 2.92 tok/s), and kernel import cleanup. Channel
- aider #general: Repeated Gemini 3 errors and migration of the experimental channel to a new Discord server. Channel
- Manus.im #general: Manus 1.5 improvements (task handling, eliminating retard loops) and developers seeking opportunities. Channel
- Windsurf #announcements: Gemini 3 Pro going live plus minor bug fixes. Channel
- MLOps @Chipro #events: AI governance and control webinar on December 3; registration: https://bit.ly/3LPl7FO. Channel
Papers and Research
#### Weight Tying Strategies in Transformers Community discussion on weight tying in Transformers — commonly used in small models to reduce parameters. Cohere's Command A model employs this strategy.
#### Virtual Width Networks (VWN) Linear Attention Discussion of Virtual Width Networks' linear attention, with updates moving from token level to layer level to avoid vanishing gradients.
#### Paper on Ideal Optimizer Structure A paper titled "The structure of this is ideal for an optimizer" exploring the ideal structure of optimizers was shared.
> Paper link: arXiv:2511.09030
---
Tools and Integrations
#### Mojo + Poetry Integration
A guide to integrating Mojo with Poetry by modifying pyproject.toml to add sources and dependencies.
#### dspy-intellisense VSCode Extension A new VSCode extension, dspy-intellisense, improving type hints for DSPy.
> Related link: dspy-intellisense announcement on Twitter
#### Grok 4.1 Integration in Windsurf Windsurf announced Grok 4.1 is now live on its platform.
> Related link: Windsurf announcement
---
*Source: Easy AI Teaching Project*