AI Updates on 2025-11-01

Alibaba releases Qwen3-VL models with support across multiple platforms including Ollama, LM Studio, and llama.cpp, with GGUF weights available for all variants from 2B to 235B parameters, supporting CPU, CUDA, Metal, and Vulkan backends @Alibaba_Qwen
OpenAI releases Sora-generated 4-minute "Monster Manor" Halloween video, demonstrating the model's video generation capabilities @OpenAI
OpenAI announces credit-based pricing now live in Codex @gdb
Microsoft announces Copilot is now built into Windows 11 with voice activation via "Hey Copilot" command @Copilot
Google showcases Veo 3.1 video generation capabilities and Nano Banana image generation features for Halloween-themed content creation @GeminiApp

Amazon holds 7.8% ownership stake in Anthropic valued at $9.5B according to Q3 earnings, while Google holds up to 8.8% stake based on unrealized gains from non-marketable equity @deedydas
SF AI startup founder reports abandoning AI-assisted coding interviews because they only measured candidates' hands-on experience with AI tools rather than engineering fundamentals, returning to algorithmic interviews for better signal @GergelyOrosz
Gerge Orosz observes increasing adoption of Claude Code terminals in coffee shops, noting faster-than-expected CLI spread among developers @GergelyOrosz
NVIDIA and Palantir demonstrate AI-powered supply chain system enabling thousands of Lowe's stores to operate as one intelligent system that anticipates and adapts to disruptions in real-time @NVIDIAAI
Gigawatt-scale Stargate data center announced as largest single investment in Michigan history @gdb

Majority of consumers express concern about data centers driving up electricity costs, raising questions about industry preparedness for potential public backlash @TechCrunch
Nathan Lambert criticizes arXiv's new moderation policies requiring peer review for certain submissions, arguing this creates unpredictable barriers to research dissemination and represents a "slippery slope" toward the platform's decline, advocating instead for AI-native curation systems @natolambert
Ethan Mollick notes ChatGPT's image generation is "actually getting close to funny at times" when comparing outputs from the same prompt a year apart, demonstrating rapid improvement in AI humor capabilities @emollick
Gerge Orosz reflects on generational shifts in software engineering, noting how each generation faces skepticism from the "old guard" about their tools and methods, yet consistently proves successful despite different skill sets @GergelyOrosz

Claire Vo builds Halloween Candy Scanner AI app using Gemini that analyzes photos or videos of candy hauls to identify pieces, count quantities, estimate total calories, and calculate teeth-brushing time needed @clairevo
Perplexity launches accurate currency conversions feature on iOS app and web @AravSrinivas
Developers create various Halloween-themed AI applications including spooky photo booths, costume generators using v0 and Nano Banana, 80s costume photo generators, and character voice generators @clairevo
Andon Labs researchers embed various LLMs in a vacuum robot to test embodiment readiness, with humorous results @TechCrunch

Ethan Mollick observes mathematics appears to be the first academic field reaching consensus that AIs will accelerate research, based on feedback from math professors, though noting this differs from autonomous research @emollick
Timothy Gowers suggests we have entered a "brief but enjoyable era where our research is greatly sped up by AI but AI still needs us" @AndrewCurran_
MIT CSAIL commemorates Yann LeCun's 1998 paper on gradient-based deep learning for document recognition, noting it took over a decade before neural networks gained widespread acceptance @MIT_CSAIL
Ethan Mollick identifies innovation and design thinking processes as urgently needing change due to AI, noting research shows many constraints change dramatically while some aspects like building empathy remain important @emollick
Simon Willison highlights a novel approach to working with multiple coding agents simultaneously through coordinated agent communication and task management @simonw
Investigation into reported Codex degradations provides detailed analysis of model performance changes @gdb