Easy AI Daily | November 18, 2025: Gemini 3 Pro, Grok 4.1, and Community Highlights

Forum topic · 小凯 · 2026-03-27

Summary

Easy AI Daily for November 18, 2025 covers major model releases including Google's Gemini 3 Pro, which showed strong results on HLE, Video-MMMU, and ARC-AGI-2 benchmarks, and xAI's Grok 4.1, which topped Text Arena at 1483 Elo, scored 1586 on EQ-bench, and claims 3x fewer hallucinations. Google also launched the DS Star data science agent and WeatherNext-2 weather model. Community discussions across LMArena, Perplexity AI, Cursor, Unsloth, OpenRouter, Eleuther, tinygrad, and other Discord servers touched on Gemini 3 Pro limits and performance, Grok 4.1's system prompt leak, image-injection jailbreaks, Builder.ai's fake-AI scandal, LiteAPI fraud concerns, weight tying in Transformers, Virtual Width Networks linear attention, and tooling updates like dspy-intellisense for VSCode, Mojo-Poetry integration, and Windsurf's support for Gemini 3 Pro and Grok 4.1.

📅 AI Industry Update — November 18, 2025

Model Releases & Updates

Google releases Gemini 3 Pro

Google released Gemini 3 Pro, showing strong performance on benchmarks including HLE, Video-MMMU, and ARC-AGI-2. It is accessible via AI Studio and Google's VS Code fork. The model card link was shared and later removed. > Link: Gemini 3 Pro Model Card (removed)

xAI releases Grok 4.1

xAI released Grok 4.1, achieving Text Arena #1 (1483 Elo), EQ-bench 1586, and Creative Writing v3 1722 Elo. Available on web and app, with 3x fewer hallucinations. > 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, a weather prediction model. > Links: DS Star announcement | WeatherNext-2 announcement

Discord Community Highlights

LMArena #general: Gemini 3 Pro performance, AI Studio's 50-per-day limit, LMArena technical issues. Channel
Perplexity AI #general: Gemini 3 Pro launch issues (some users downgraded to 2.5), Google Antigravity IDE availability. Channel
BASI Jailbreaking #general: Grok behavior changes, AI solving reCAPTCHAs (over 50% success rate), Builder.ai fake-AI scandal (700 Indian developers hand-coding). Channel
BASI Jailbreaking #jailbreaking: Image-injection jailbreaks (bypassing text safety via images), Grok 4.1 system prompt leak. Channel
Cursor Community #general: Mac OS shortcomings, Composer free period ending November 11, Gemini 3 Pro performance. Channel
Unsloth AI #general: Gemma 3 270M shortcomings vs Granite-4.0, llama.cpp speeds on Ryzen 8 series (over 25 tok/s), TPU support limits. Channel
OpenRouter #general: LiteAPI fraud concerns (40% cheaper but violating ToS), Grok 4.1 release, Gemini 3 launch coinciding with a Cloudflare outage. Channel
OpenAI #ai-discussions: Custom GPT efficiency, Gemini 3 Pro's one-shot React/SwiftUI code, Grok 4.1 creative writing. Channel
LM Studio #general: E-commerce return abuse, LLM performance degradation over long sessions (suspected bug), MCP integration tools. Channel
Latent Space #ai-general-chat: Sourcegraph ad revenue (5-10M ARR), Grok 4.1 (Text Arena #1), Poe group chat support (200 people). Channel
Nous Research AI #general: Amazon Nova Premier v1 novelty, high Bedrock AWS costs ($3000 bill), Gemini 3 Pro generating a real-time raytracer. Channel
HuggingFace #general: New graph-RAG database (alternative to Kilo Code/Pinecone), Mimir project (multi-agent orchestration, MIT license), Lablab hackathon fraud controversy. Channel
Yannick Kilcher #general: *Understanding Machine Learning* book, ReLU activation advantages, Gemini 3 Pro model card (outperforming GPT-5.1). Channel
Eleuther #general: EleutherAI's NeurIPS 2025 papers (3 in the main track), HuggingFace throughput improvement (60% faster via empty logits processor). Channel
Eleuther #research: Weight tying in Transformers (common for small models), Cohere Command A's weight tying, Virtual Width Networks (VWN) linear attention. Channel
Modular (Mojo) #mojo: NVFP4 support (for Nvidia/GPUMode competition), negative indexing and Int vs UInt debate, Poetry integration. Channel
DSPy #general: dspy-intellisense VSCode extension (type hints), MLflow alternatives (Arize Phoenix), LLM non-determinism (temperature 0 solution). Channel
tinygrad #general: tinybox meetings, Llama 1B speed on Tinygrad (6.06 tok/s vs Torch's 2.92 tok/s), kernel import cleanup. Channel
aider #general: Gemini 3 repeated errors, experimental channel migration to a new Discord server. Channel
Manus.im #general: Manus 1.5 improvements (task handling, eliminating loops), developers seeking opportunities. Channel
Windsurf #announcements: Gemini 3 Pro rollout and minor fixes. Channel
MLOps @Chipro #events: AI governance and control webinar announced for December 3. Registration: https://bit.ly/3LPl7FO. Channel

Papers & Research

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 to layer level, avoiding vanishing gradients.
Ideal structure for an optimizer: A shared paper exploring ideal optimizer structure. Paper

Tools & Integrations

Mojo + Poetry integration: Modifying pyproject.toml to add sources and dependencies.
dspy-intellisense VSCode extension: Adds type hints for DSPy. Announcement on X
Grok 4.1 in Windsurf: Windsurf announced Grok 4.1 availability. Announcement on X

---

*Source: Easy AI teaching project.*