Diffbot has developed an AI chatbot using Meta's Llama 3.3 model enhanced with graph retrieval-augmented generation technology. Unlike large language models, it searches a continuously updated knowledge graph containing over one trillion facts, providing more accurate and transparent responses. Diffbot's model scored 81% on FreshQA and 70.36% on MMLU-Pro benchmarks, outperforming competitors. The model is open-source and available for enterprises to customize and run locally.

Google is consolidating its AI teams into DeepMind to accelerate development and transform research into tools. This includes integrating Gemini model API teams and enhancing public outputs like APIs and open-source projects. CEO Pichai stresses the urgency to expand Gemini's presence in consumer markets in 2025.

OpenAI introduces new personalization settings for ChatGPT enhancing user experience with options like preferred name and chatbot traits. Users report changes in custom instruction menu without seeing new options yet. This update improves interface rather than technology, maintaining previous prompt engineering approach. OpenAI reviews custom instructions for compliance. ChatGPT now serves over 300 million weekly users with features like real-time web search and canvas interface.

xAI launched Grok's standalone iOS app offering real-time web access, text rewriting, summarization, Q&A, and image generation. Previously exclusive to X subscribers, Grok now competes with ChatGPT, Claude, Gemini, and Copilot. xAI also secured $6B in funding from Nvidia and AMD. Android version pending.

Microsoft releases Phi-4, a compact language model on Hugging Face under MIT license. It excels in efficiency, mathematical reasoning, and specialized applications with enhanced security. Phi-4 supports SMEs and promotes sustainable AI development, demonstrating high performance without extensive resources.

Alibaba Cloud releases Qwen Code AI Programmer, supporting VS Code and JetBrains IDEs. It offers conversational collaboration for coding, bug fixing, and unit test generation. With the Qwen2.5-coder model, it provides advanced code handling and intelligent agents. Features include multi-file editing, snapshot functionality, and cross-language support, significantly boosting development efficiency. Over 9 million downloads and 1.5 billion lines of generated code make it a leading AI programming t

OpenAI faces multiple challenges in 2025 including legal disputes, conspiracy allegations, and economic struggles. CEO Sam Altman's sibling sued him for abuse, a former employee's death raised suspicions, and ChatGPT was used in a terror plot. Financial losses and sustainability concerns add pressure. Competitors and ongoing scandals threaten the company's future. OpenAI must prove its business model viable amidst these issues.

Grok, X's AI chatbot, will introduce an unrestricted mode featuring offensive and controversial responses, reflecting Musk's vision of an unfiltered AI. Currently, Grok is somewhat reserved on political topics and leans slightly left on social issues. Musk aims to make Grok more politically neutral by improving its training data. The unrestricted mode is not yet live.

Microsoft rolls back Bing Image Creator to older DALL-E model after user complaints about deteriorated performance and image quality following December upgrade. Issues included lack of detail, inaccurate content reflection, and problematic anime-style character rendering. The company aims to address both technical and subjective critiques.

AI researcher Francois Chollet co-founded ARC Prize Foundation to develop benchmarks for assessing human-level AI intelligence. The foundation will expand the ARC-AGI test, focusing on closing the gap between AI and human capabilities. Despite criticisms, the nonprofit aims to create challenging tasks for AI systems and collaborate with industry leaders to establish AGI standards. Future ARC-AGI versions will pose significant challenges even to advanced models like OpenAI's o3.

Microsoft released Phi-4, a 14 billion parameter language model based on Transformer architecture optimized for efficiency. Available on Hugging Face, it outperforms larger models in benchmarks. Google and Meta also recently open-sourced compact models with advanced techniques like quantization.

During CES, Meta's Yann LeCun criticized the current definition of AGI and argued that expanding text-driven LLMs is insufficient for achieving human-level intelligence. He noted that while LLMs excel at sentence completion, they lack complex brain functions and reasoning abilities. LeCun emphasized that current AI systems are narrow and fail outside specific tasks, citing autonomous vehicles as an example. He also mentioned diminishing returns in LLM expansion and highlighted the potential of g

Highlights from CES 2025 include Yukai Engineering's Nékojita FuFu robotic cat cooling device, Kirin Holdings' electronic spoon enhancing flavor, Acer's Nitro Blaze 11 handheld gaming console, Lenovo's ThinkBook Plus Gen 6 Rollable laptop, Spicerr AI-powered spice dispenser, Birdfy Smart Bird Bath Pro, Meticulous smart coffee maker with robotic lever, Razer's temperature-controlled gaming chair, In Case of Death digital legacy iPad package, and Yukai Engineering's Mirumi plush robot for wallets.

IBM study shows retail and consumer goods executives prioritizing AI investments with a 52% expected increase next year aiming to boost innovation and efficiency. However, significant gaps in AI governance remain, with only 25% fully implementing risk management tools. Successful integration requires aligning AI initiatives with brand goals and fostering cross-functional collaboration.

An innovative system integrates Google Docs, Brave Search API, and AI agents like Gemini 2.0 to automate research and report writing. It features two principal agents for data gathering and report organization, operating autonomously until task completion. This AI-driven setup promises enhanced efficiency, reduced workload, and refined outputs for various applications including business analysis, academic research, and blog writing. Future enhancements focus on improving search accuracy, report

Inworld AI, NVIDIA, and Streamlabs launched the world's first intelligent streaming assistant at CES 2025. This multimodal AI serves as an interactive co-host, producer, and technical assistant, enhancing live streams with real-time highlights, audio-visual effects, and seamless technical support. It integrates NVIDIA's ACE technology, Inworld AI's cognition capabilities, and Streamlabs' streaming expertise, offering creators customizable personalities to match their style and boosting audience

Google announces integration of Gemini AI into Google TV to enhance user experience with smarter features including streamlined content search, interactive Q&A on various topics, and smart home controls. Rollout expected late 2025.

Nvidia's GR00T platform, launched last March, gained support from major firms like Boston Dynamics and Figure AI. At CES, CEO Jensen Huang introduced Blueprint, a feature focusing on imitation learning for teaching robots by replicating human actions. This method is effective for automating tasks in factories and warehouses using remote operation and digital twins via Apple's Vision Pro.

Researchers from Meta AI introduced a scalable storage layer to enhance factual accuracy and reduce fictional content in large language models. This approach uses sparse activation and key-value lookups, offering computational efficiency. Experiments showed that storage-enhanced models outperformed dense models and matched MoE models' performance with fewer resources, suggesting significant potential for next-generation AI architectures.

At CES 2025, Uber and Nvidia announced a collaboration to advance autonomous driving using Cosmos, a generative world model tool, and DGX Cloud, a cloud-based AI supercomputing platform. Uber, focusing on partnerships rather than in-house development, plans to leverage these tools for more efficient scaling of autonomous driving services in cities like Austin and Atlanta. Previously, Uber faced challenges including legal issues and the 2018 pedestrian fatality, leading to the sale of its ATG div