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