Easy AI Daily Digest | November 19, 2025

New Model Releases and Updates

Meta Releases SAM 3 Unified Image/Video Segmentation Model

Meta launched SAM 3, a unified model supporting text and visual prompts for image and video segmentation. Performance is 2x better than existing models with 30ms inference time. A Playground testing tool and model checkpoints on GitHub/HuggingFace are available, alongside a partnership with Roboflow for infinitely scalable segmentation endpoints.

Links: Meta SAM 3 official blog | Roboflow partnership announcement | SAM 3 GitHub resources

OpenAI Releases GPT-5.1-Codex-Max Coding Model

OpenAI launched GPT-5.1-Codex-Max, optimized for long-context scenarios with multi-context-window compression. It is positioned for long, detailed coding work and seen as a response to recent competition in coding agents.

Link: OpenAI announcement

Google Releases Gemini 3 Multimodal Model

Google released Gemini 3, which excels in benchmarks, but some users report its creative writing is worse than Gemini 2.5, sparking discussion about model generalization.

Links: Google Gemini 3 overview | Benchmark discussion

Meta Releases SAM 3D for 3D Reconstruction

Meta launched SAM 3D, reconstructing 3D structures of objects and human bodies from single images, with GitHub resources and a Roboflow fine-tuning path.

Links: Meta SAM 3 announcement | SAM 3D GitHub

OpenAI Launches ChatGPT for Teachers

OpenAI introduced ChatGPT for Teachers for US K-12 education, offering a safe workspace and admin tools with compliance and privacy controls for classroom AI integration.

Link: OpenAI announcement

Community Discussions and Controversies

• Ollama pricing strategy sparks open-source debate: Ollama's new paid cloud service led users to question its open-source commitment, comparing it unfavorably to llama.cpp's free model. Reddit thread
• Gemini 3 content filtering criticized: Users report overly strict filtering, including API bans triggered while summarizing books, considered harsher than ChatGPT. Twitter discussion
• ChatGPT anomalous behavior reports: Users report text-to-image conversions, wrong file structures, and destructive commands during code generation. Reddit thread
• Gemini 3 Pro underperforms in Cursor: Due to high demand, Gemini 3 Pro is unavailable in Cursor and reportedly hallucinates code in large projects; some users switched to Sonnet 4.5 or Composer.
• ChatGPT executing git reset --hard: A user lost local code when ChatGPT ran this destructive command, prompting calls for tools like Cursor to block dangerous commands. Reddit safety discussion

Tools and Platform Integrations

• Gemini 3 integrated into Windsurf: Early glitches fixed; latest version available. Announcement | Download
• SAM 3 x Roboflow: Infinitely scalable segmentation endpoints via text prompts (e.g., "green umbrella"), comparable with Claude and YOLO World.
• Gemini 3 with Aider: Requires --model=gemini/gemini-3-pro-preview; --weak-model recommended for faster commits. Aider docs
• Google launches Antigravity AI IDE: VS Code-based IDE supporting Sonnet 4.5 and Gemini 3, with an agent window and artifact system; early reports of migration bugs and prompt limits. Announcement
• Perplexity asset creation: Pro/Max users can create and edit slides, spreadsheets, and documents with real-time collaboration and export. Announcement

Performance Tests and Benchmarks

• Tinygrad beats PyTorch on CPU for Llama 1B: 6.06 tok/s vs Torch's 2.92 tok/s, aided by CPU_LLVM acceleration and forward-pass optimizations.
• KTOtrainer high memory usage: A 0.5B model consumed 80GB GPU memory due to dual model loading, dual forward passes, long-sequence padding, and CUDA memory reservation; optimization advice provided in the community.
• Gemini 3 benchmarks mixed: SOTA on WeirdML and IBench, but criticized on creative writing generalization.
• Kimi K2 Thinking leads agentic coding: Top results on Tau bench and HLE tool tasks; considered the closest open model to GPT-5. Moonshot announcement
• DeepSeek Cogito v2.1: Based on DeepSeek-7B, outperforms Mistral-7B and Gemma-9B on introspection tasks, detecting and reporting injected concepts. Announcement

Issues and Troubleshooting

• Manus users cannot access TiDB Cloud: Quota exhaustion and no console access; ticloud CLI lacks API keys.
• GPT-5.1 connection errors in Aider: response.reasoning.effort validation errors persist regardless of reasoning-effort settings; possibly related to OpenAI API changes. LiteLLM issue
• Manus credit system changes: Monthly reset vs. previous never-expiring plans confuses users.
• Windsurf fixes Gemini 3 loading glitch: Users advised to update to the latest version.
• Cursor Gemini 3 "unavailable due to high demand": Sonnet 4.5 suggested as a workaround.

Education and Resources

• Community members offer AI coding education courses covering core concepts, advanced models, practical applications, and ethics (contact via DM).
• Users ask about certification for HuggingFace's smol-course and agents-course; newcomers welcomed. HuggingFace courses
• Tinygrad benchmark tests comparing CPU performance with Torch added under test/external. Tinygrad GitHub
• AI security red-teaming resources, including jailbreak prompts and AzureAI chat widget testing. Resource
• llama.cpp recommended as a free, open-source alternative to Ollama for local inference. llama.cpp GitHub

---

*Source: Easy AI education project*