AI Updates on 2025-12-06

Essential AI releases Rnj-1, an 8B parameter base and instruct model pair achieving SWE-bench performance close to GPT-4o, tool use outperforming comparable open source models, and mathematical reasoning on AIME 2025 nearly matching GPT OSS MoE 20B @ashVaswani
Google announces Gemini 3 Pro and Nano Banana Pro in Google Search via AI Mode expanded to more countries in English language @GoogleAI
Google updates Deep Think mode in Gemini App for Google AI Ultra subscribers, improving reasoning capabilities by exploring multiple hypotheses simultaneously @GoogleAI
NVIDIA Nemotron models integrated with Amazon Bedrock, with early adopters like CrowdStrike powering security agents and BridgeWise AI delivering financial insights @NVIDIAAI
Reports suggest OpenAI's GPT-5.2 code red response to Google coming December 9th, earlier than originally planned @apples_jimmy

Meta acquires AI device startup Limitless, expanding its AI hardware capabilities @TechCrunch
AI synthetic research startup Aaru raises Series A at $1B headline valuation @TechCrunch
Ex-Google startup Yoodli triples valuation to over $300M with AI built to assist rather than replace people @TechCrunch
SpaceX reportedly in talks for secondary sale at $800B valuation, which would make it America's most valuable private company @TechCrunch
Bay Area engineering compensation landscape shows OpenAI and Anthropic engineers earning multi-million packages, while AI startup engineers at $200k grind to prompt LLMs and restart after new model releases @deedydas
NVIDIA RTX PRO 6000 GPUs will render 99% of Pixar shots with RenderMan XPU, reshaping Pixar's workflow for Toy Story 5 with bigger scenes and faster rendering @NVIDIAAI

Research shows AI-generated advertising outperforms human-created ads by 19% in click-through rates, but disclosing AI use results in 32% performance drop, raising questions about transparency requirements @AndrewCurran
Ethan Mollick notes AI-created visual ads achieved 20% more clicks than human expert ads, but disclosure of AI creation reduced performance to 31% less than human-made ads @emollick
OpenAI's Nick Turley clarifies there are no live tests for ads in ChatGPT, stating any future ad implementation would take a thoughtful approach respecting user trust @nickaturley
Ethan Mollick raises concerns about xAI's lack of transparency regarding their approaches to AI, safeguards, and what truth-seeking means, particularly important for enterprise use @emollick
Mollick notes odd findings in Grok 4.1 model card including increasing sycophancy rates and high deception scores compared to other models @emollick
Andrew Curran predicts governments will push for backdoor legislation in home robots, demanding mandatory override codes for authorities despite citizens potentially pooling resources for local security @AndrewCurran
Khosla Ventures managing partner Keith Rabois calls AI safety a complete hoax, stating it's bureaucrats finding excuses to interfere with progress @tbpn
Amanda Askell confirms Claude was trained on a real alignment document in supervised learning, with full version and details to be released soon @alexgraveley

Perplexity Finance launches full screen graphs feature @AravSrinivas
NotebookLM mobile app receives updates including Slide Decks and Infographics, Images as Sources, and saved Audio Overview progress @GoogleAI
Google Workspace Studio launches, empowering subscribers to automate work from simple tasks to complex processes with custom AI agents @GoogleAI
Hex CEO Barry McCardel discusses how AI changes data interaction through collaborative analytics workspaces, agent workflows, and conversational interfaces @sarahdingwang
CrowdStrike powers advanced security agents in Charlotte AI AgentWorks using NVIDIA Nemotron models @NVIDIAAI

Google's Gemini 3 Pro demonstrates state-of-the-art multimodal performance across document, screen, spatial, and video understanding, with capabilities to derender complex documents into structured code and generate collision-free trajectories for robotics @googleaidevs
Jeff Dean demonstrates Gemini 3 Pro visual reasoning by having it annotate performance improvements versus competing models, showing large relative accuracy gains across benchmarks @JeffDean
Stanford Professor Yejin Choi presents research on latent collaboration in multi-agent systems and discusses 2026 AI predictions at NeurIPS 2025 @NVIDIAAIDev
Research paper Colors of Growth develops novel approach to measuring long-run economic growth by analyzing systematic variation in color use in European paintings from 1600-1820 @emollick
Ethan Mollick summarizes 2025 AI trends: no slowdown in exponential gains, jaggedness remains main issue, early positive ROI reports, GenAI became industry-level, and AI remains fundamentally weird @emollick
Deep Learning for Code workshop at NeurIPS 2025 focuses on code agents in the agentic era with speakers including Graham Neubig and Dawn Song @Alibaba_Qwen
Stanford researcher notes that deep learning success requires getting 98% of details right, with the last few details having extremely nonlinear impact @arimorcos