Large language models, dubbed "black boxes," remain mysterious. Anthropic's study on AI algorithm behavior sheds light on decision-making processes. Researchers aim to comprehend neural networks and the inputs triggering outputs. Anthropic's methods aim to understand AI reasoning.

Slack has faced criticism for using customer data to enhance machine learning capabilities, but clarified that data is not used for advanced AI products. User data is anonymous and not shared with third parties for training. Clarifications have been made regarding data usage.

Microsoft launched Phi Silica, a small language model for NPUs in Copilot+ PCs, boasting 3.3B parameters and efficiency in token generation. Developers can access Phi Silica API for innovative AI experiences within the Windows ecosystem. Copilot+ PCs with dedicated AI processors are set for mid-June release, with Microsoft and Intel investing in AI integration for end users.

iFlytek announced free iFlytek Starfire Lite API to promote AI technology democratization. iFlytek also launched pricing for Starfire Pro/Max API at 0.21 yuan per 10,000 tokens. The new iFlytek Starfire Large Model V3.5 brings significant improvements in text generation, language understanding, knowledge question answering, and more. The goal is to make AI technology more accessible and beneficial to various industries.

Adobe introduces new Generative Remove feature in Lightroom, powered by Firefly AI. Simplifying editing workflow for all users, it removes unwanted objects with realistic pixel-level perfection. Other AI-based features include Lens Blur tool and improved mobile editing. Generative Remove includes Content Credentials for transparency. Early access in Lightroom ecosystem.

Microsoft introduced Phi-3-vision, a multimodal model combining language and visual capabilities. With 4.2 billion parameters, it offers insights from charts and diagrams, enhancing AI applications. The Phi-3 family includes Phi-3-small, Phi-3-medium, and Phi-3-mini, available on Azure. Phi-3-vision excels in OCR, chart analysis, & diagram understanding, surpassing larger models in visual reasoning. Compact design enables deployment on devices, offering low-latency AI experiences with cost-effec

Wearable startup Humane is seeking to sell its AI Pin project for $750M-$1B. Despite promising innovation, the device has faced criticism for performance issues. Market competition and user feedback pose challenges for the company's future.

Google announced the launch of AI-powered search with integrated shopping ads in the US. Ads will be labeled clearly and triggered by user keywords. Users welcome ads as helpful shopping recommendations. Google plans to fully promote this model, providing advertisers with easy ad display in search results. This initiative aims to enhance the user experience and benefit both consumers and advertisers.

Zhipu·AI launched new multimodal model CogVLM2 with significant performance improvements. It supports 8K text and 1344*1344 images, surpassing GPT-4V in tests. Enhanced visual-language fusion boosts processing efficiency for large-scale data. CogVLM2 excels in various benchmarks, showing promise for AI technology advancement.

Wondercraft introduces "Parrot Mode," allowing precise dialogue control using AI voices. Co-founder Oskar Serrander emphasizes its importance for creators and advertisers, reducing feedback loops. Wondercraft also launches "Workspaces" for team collaboration and expands audio resources with new voices and music tracks.

Microsoft's Copilot intelligent assistant is set to receive a major upgrade with GPT-4o, marking a significant advancement in AI. The update will enhance collaborative experiences with more intelligence and efficiency.

Intel unveils new laptop processor Lunar Lake, promising unparalleled AI experiences for users. Lunar Lake boasts 1.4x faster performance than Snapdragon X Elite, thanks to its NPU capable of 40 TOPS. Intel aims to ship 40 million AI PC processors, showcasing its innovation and dominance in the market. Stay tuned for more details at Computex and Hot Chips events.

Scarlett Johansson declined OpenAI's offer to voice ChatGPT, but a voice similar to hers, named Sky, was created without her consent. After public backlash, OpenAI apologized and removed Sky's voice. Johansson seeks transparency and legal protection.

Neuralink gains FDA approval for next brain chip implant, addressing previous issues. Arbaugh's chip malfunctioned, causing data leakage. Neuralink plans to implant chips in up to 10 more patients this year, aiming to address past malfunctions and improve technology. Arbaugh regains abilities with software updates and remains supportive of Neuralink. Surgeons use robot "R1" to implant chip wires directly into patients' brains. Previous animal testing faced ethical concerns.

Microsoft unveiled a new AI chatbot for laptops that remembers your actions and assists. Competing against Google and OpenAI, Microsoft integrates AI into Windows OS to enhance user experience and privacy protection.

Microsoft to unveil latest AI advancements at Build conference, aiming for AI integration in Windows to enhance user experience. Google and OpenAI pose competition with recent AI events. Analysts positive on Microsoft's AI strategy to boost PC market. Microsoft's AI PCs to use AMD, Intel, Qualcomm chips with NPUs for complex AI tasks, ahead of tech trend in AI hardware.

OpenAI adjusts strategy after departure of key scientists, emphasizing AI safety. New Head of Science appointed to lead safety alignment work. CEO emphasizes commitment to secure AI deployment. Focus on global AI regulation for safety and responsibility.

ElevenLabs released Audio Native, an audio player for generating human-like narrations from online content. Users can customize voices, appearance, and pronunciation. Deployment options include embedding players, using existing projects, or API integration. Subscription to ElevenLabs' Creator plan is required. Audio Native enhances content accessibility and engagement, but AI technology is evolving towards local implementations for faster, more efficient, personalized experiences.

Former OpenAI chief scientist faces challenges in researching existential AI risks as the company disbanded his team. Superalignment research now integrated into other efforts, led by co-founder John Schulman. The company emphasizes the need for safety and alignment in AI development. The dissolution of the Superalignment team has raised concerns about the future of controlling superintelligent AI systems.

Google launched the Frontier Safety Framework to proactively detect and mitigate risks posed by advanced AI systems. It focuses on capability identification, model evaluation, and mitigation plans. Other institutions like OpenAI and Anthropic are also engaged in AI safety research. The framework aims to enhance AI technology responsibly by 2025.