Moon's Dark Side launches K1, an advanced visual reasoning AI with end-to-end image understanding and chain-of-thought. It outperforms top models in scientific benchmarks and is available on mobile and web. K1 excels in basic sciences but needs improvements in generalization and complex tasks.

Meta AI introduces Large Concept Models (LCMs) using high-dimensional SONAR embeddings for sentence-level processing. LCMs support over 200 languages and modalities, enhancing coherence, scalability, and zero-shot generalization compared to traditional LLMs.

Tencent launches POINTS1.5, an enhanced version of POINTS1.0 with optimized LLaVA architecture. It outperforms leading open-source models under 10B parameters and excels in OCR, reasoning, and image tasks. POINTS1.5 underscores Tencent’s commitment to advancing AI technology.

Instagram's Adam Mosseri warns that AI-generated images are increasingly realistic, urging careful verification of online content. He advocates for platforms to label AI content and disclose creators. Meta plans policy updates to enhance content reliability with user-driven management similar to community notes and filters.

xAI unveils Grok-2-1212 with lower API costs ($2M input, $10M output), improved accuracy, multilingual support, faster responses, real-time web search, and citation generation, enhancing developer adoption.

Google launches Gemini 2.0 Flash, a multimodal AI enabling real-time video interactions on devices. This advancement intensifies competition with OpenAI and Microsoft, offers faster processing and affordability, and fosters new interactive computing ecosystems.

Google is gradually deploying Gemini-powered Assistant to select Nest Audio and Mini speakers for Nest Aware subscribers. The AI-enhanced Assistant offers detailed responses, camera search, and routine setup, making Google the first major tech company to integrate generative AI into smart home assistants.

OpenAI launched ChatGPT’s Projects feature, enabling task organization with folders, file uploads, custom instructions, and chat categorization. Currently available to Plus, Pro, and Teams users, with free and enterprise access coming soon. Projects enhance efficiency and management.

Meta FAIR and UCSD introduce COCONUT, a Continuous Chain-of-Thought framework enabling LLMs to reason in latent space. COCONUT outperforms traditional CoT in accuracy and efficiency across multiple datasets, enhancing multi-path exploration and computational performance.

Anthropic's "AI for Code" revenue surged tenfold, doubling its enterprise market share to 24%. With Claude adopted by Cursor and integrated into GitHub Copilot by Microsoft, Anthropic is intensifying competition against OpenAI, which remains the leader with $4 B in revenue and advanced AI features.

Peking University and ByteDance launched the Doubao Large Model Joint Lab to advance AI large models. Combining ByteDance’s AI expertise with the university’s research, they focus on training, efficiency, and applications. The lab offers student internships and aims to strengthen China’s AI sector.

Vapi secures $20M Series A led by Bessemer, valuing it at $130M. Specializing in voice AI for sectors like healthcare and finance, Vapi's APIs enable rapid custom voice agent development. The funding will expand engineering and infrastructure, positioning Vapi as a flexible, developer-focused alternative to tech giants.

Microsoft introduces Phi-4, a 14B parameter AI model with superior math abilities, available on Azure AI Foundry for limited research use. Enhanced by high-quality synthetic and human data plus post-training methods. Competes with models like GPT-4o mini, emphasizing efficiency. AI labs report a training data bottleneck.

Deepfakes surged 4x globally from 2023 to 2024. Meta released open-source Meta Video Seal, Watermark Anything, and Audio Seal to embed watermarks in AI-generated content, enhancing detection and originality. Challenges include adoption and balancing watermark visibility.

Google Drive enhances Gemini integration, enabling AI-generated folder summaries. Users can summarize, search, or query contents via buttons or drag-drop. Supports text, PDFs, spreadsheets, presentations, and images. Available to Google One AI Premium and select enterprise users.

On day six of "OpenAI 12 Days," ChatGPT gains video input, screen sharing, and a festive Santa mode. These features enhance user experience and are available to Plus, Pro, and Team users, with broader release planned for next year. Data privacy measures are implemented.

Google introduced Jules, an AI-driven code assistant that autonomously fixes code errors, enhancing developer efficiency and quality. Integrated with Gemini 2.0, Jules competes with GitHub Copilot and other AI tools. Currently in testing, it’s set for wider release in early 2025.

Google launched Gemini 2.0, a multimodal AI enhancing text, image, audio, and code handling with improved depth, creativity, and precision. Accessible via Google Search and the Gemini app, it outperforms Gemini 1.5 Flash in accuracy and capabilities.

Cognition Labs launches Devin, the first AI Programmer, offering comprehensive coding assistance. Available for $500/month with integrations and support, Devin excels in multiple languages, autonomous development, bug fixing, and adapting to new technologies, enhancing efficiency for developers and teams.

Google launches Deep Research, an AI tool using the Gemini chatbot for detailed online reports. Available in English to Gemini premium subscribers, it creates research plans, compiles key findings with sources, and exports to Google Docs. Part of Gemini 2.0, marking Google's move into "agency" AI.