Easy AI Daily News | November 21, 2025: Gemini 3 Pro Launch, GPT-5 Math Breakthroughs, and AI Industry Updates

Forum topic · 小凯 · 2026-03-27

Summary

Easy AI Daily for November 21, 2025 covers Google's release of Gemini 3 Pro and the Nano Banana Pro image model with improved text rendering, 4K visuals, and reasoning; GPT-5 reportedly proving decade-old math conjectures; AI2's OLMo 3 technical report; and SmolLM3's reasoning mode. Community discussions span LMArena debates on Gemini-3 image realism, Perplexity adding Kimi-K2 Thinking and Gemini 3 Pro access, and Modular Platform 25.7's MAX Python API. Research highlights include Kitsune's spatial-pipeline GPU execution (2.8x inference speedup), Iris tile-based symmetric memory for multi-GPU programming, and Octa in-kernel communication for distributed LLMs. Funding news: Genspark raised $275M at a $1.25B valuation, and Cline launched a $1M cline-bench reward. Controversies include an 88% reported hallucination rate for Gemini 3 Pro and academic misconduct concerns over teachers using AI to boost work.

📅 AI Industry Updates — November 21, 2025

Model Updates & Performance

Google Releases Gemini 3 Pro and Nano Banana Pro Image Model

Gemini 3 Pro and Nano Banana Pro (Gemini Image Pro) launched with improved text rendering, 4K visuals, and reasoning capabilities, supporting lighting control and flexible aspect ratios. Community feedback highlights clear infographic and academic figure generation, with user preference win rates exceeding 80%.

GPT-5 Solves Decade-Old Math Problems; Distillation to K2 Rumored

GPT-5 with a scaffold reportedly proved the 2013 tree subgraph conjecture and a 2012 COLT dynamic networks problem within two days. Some users claim GPT-5.1 has been distilled into Moonshot AI's K2 model.

Multi-Turn Jailbreak Strategies for Claude Sonnet 4.5

Community members proposed multi-turn strategies to unlock Claude Sonnet 4.5 by convincing the AI it needs to generate visualizations in an artifact app, suggesting adaptations of the ENI prompt from /r/ClaudeAIJailbreak.

SmolLM3 Introduces Reasoning Mode

SmolLM3 published its training process and introduced a reasoning mode; the community uses it as an example for learning reasoning behavior in model training.

OLMo 3 Technical Report Released

AI2 released the OLMo 3 technical report detailing post-norm training, sliding window attention, GQA, and other architecture details, targeting 32B-scale comparability.

Related links: OLMo 3 Blog | Technical Report PDF

Community Discussions

LMArena Discord: Gemini-3 generated images hard to distinguish from real ones; Nano Banana Pro background quality degrades over multi-turn generation; Google reCAPTCHA loops making the platform unusable; Grok roleplay outperforms Gemini; OpenAI may launch 18+ content. Link
BASI Jailbreaking Discord: Users planning AI-driven WiFi attacks (community warned of illegality); Gemini CLI automation of nmap/sqlmap; Claude Sonnet 4.5 multi-turn jailbreak strategies. Link
Perplexity AI Discord: Pro/Max users gain access to Kimi-K2 Thinking and Gemini 3 Pro; Comet Android app sync issues; Brave's exaggerated indirect prompt injection vulnerability claims. Link
Unsloth AI Discord: Fine-tuning Gemini/GPT for accurate information; AI pens (Neo Smartpens); combining AMD/Nvidia GPUs; Ollama model compatibility issues. Link
Cursor Community Discord: Codex-max API availability; Cursor mid-month billing; Gemini 3 Pro sending code instead of editing files at 150k–200k context. Link
LM Studio Discord: LM Studio offers an OpenAI-compatible REST API (no API keys); Qwen3-VL-30B BF16 recommended for MacBook Pro M4 MAX. Link
OpenRouter Discord: linker.sh tool call failures (1/10); Nano Banana 2 400 errors; Gemini 3 does not yet support grounding. Link
HuggingFace Discord: Maya1 Voice Model launches on Fal; SmolLM3 reasoning mode discussions; Diffusers MVP program launched. Link
Yannick Kilcher Discord: Teachers using AI to cheat; Nano Banana Pro infographics; proposals to limit daily paper output and use AI to filter papers. Link
Nous Research Discord: Nano Banana Pro image samples (resembling early Grok); Adobe's $1.9B acquisition of Semrush; Gemini 3 Pro requires a Pro account. Link
Modular Mojo Discord: Modular Platform 25.7 released with MAX Python API, Nvidia Grace support, safer Mojo GPU programming; UnsafePointer generics no longer default. Modular Blog | UnsafePointer Proposal
Moonshot AI Discord: Kimi hosted in the US based on user location; GPT-5.1 reportedly distilled into K2; Kimi attention weaker on long-context tasks; open-source models 9 months behind. Link
DSPy Discord: Gemini 3 Pro generated a proxy server from a tweet in one shot (GitHub); suggestions to use RL for agent-generated task DAGs. Skylar B Payne Twitter
MCP Contributors Discord: modelcontextprotocol.io migrating from Anthropic DNS to community; Tool Annotations solution proposed. MCP Discord | Tool Annotations PR
aider Discord: Engineers building real-time AI voice agents handling latency and call handoffs; experiments with Feather AI (featherhq.com). Link

Tools & Platform Updates

LM Studio: Provides an OpenAI-compatible REST API without API keys for local LLM hosting; no security or metering features. Link
OpenRouter: Show livestreams on X and YouTube; linker.sh failures, Nano Banana 2 400 errors, and 401 errors in Singapore/Hong Kong reported. Twitter | YouTube
HuggingFace: Diffusers MVP program launched with active community contributions; smol-course has circular links and training issues. Diffusers Issues | smol-course
Modular: Platform 25.7 introduces MAX Python API, Nvidia Grace support, and safer Mojo GPU programming. Blog
DSPy: Gemini 3 Pro generated a proxy server from a single tweet; code available on GitHub. DSPy Proxy GitHub

Research & Papers

Kitsune: *Enabling Dataflow Execution on GPUs with Spatial Pipelines* — builds spatial pipelines via PyTorch Dynamo, achieving 2.8x/2.2x inference/training speedups and reducing off-chip traffic by 99%/45%. Paper
Iris: *Simplifying Multi-GPU Programming with Tile-Based Symmetric Memory* — tile-based symmetric memory with in-kernel communication improves multi-GPU performance by 1.79x. Paper
Octa: *Fine-Grained In-Kernel Communication for Distributed LLMs* — proposes the Three Taxes framework; in-kernel communication reduces latency by 10–20%. Paper
Gradient Compression: A gradient compression algorithm adjusting gradients based on sampled logits to improve alignment, with visualizations of the compression process.
OLMo 3: AI2's technical report covers post-norm training, sliding window attention (7B), GQA (32B), and 5.4x FFN expansion. Blog | PDF

Jobs & Funding

Rivian is hiring GPU coding experts for next-generation autonomous driving features in Palo Alto and London. Job 1 | Job 2
Modal is hiring GPU engineers for inference optimization (SGLang, FlashAttention), serving clients like Decagon and Reducto. Blog | Careers
Genspark raised $275M in Series B funding at a $1.25B valuation, launching AI Workspace for automated task completion. Eric Jing Twitter
Cline launched cline-bench with a $1M reward pool — RL environments based on real OSS issues. Cline Twitter

Controversies & Ethics

Teacher AI cheating: Users reported teachers asking AI to "boost 20% work," condemned by the community as academic misconduct. Link
Gemini 3 Pro hallucination rate: Community reports high hallucination rates — the-decoder.com cites 88%, exceeding models like GPT-4o. the-decoder.com
Brave vs. Perplexity: Brave published exaggerated claims about Comet's indirect prompt injection vulnerability; Perplexity clarified the vulnerability was not exploited. Link
Gemini-3 image realism: Community debates AI-generated images indistinguishable from real photos; Nano Banana Pro background quality degrades over multi-turn generation. Link

---

*Source: Easy AI Educational Project*