AI is not moving too fast, it’s our governance, wisdom, and imagination that need to catch up.
— Sam
Anthropic is nearing a deal to raise up to $10 billion in new funding, surpassing earlier $5 billion goals, valuing the company at around $170 billion.
🔍
Iconiq Capital leads the round with top venture firms and sovereign funds expected to join.
The raise significantly outpaces previous $3.5 billion round at $61.5 billion valuation.
Funds will support model development, infrastructure scaling, and global expansion.
Anthropic competes with OpenAI and xAI, focusing on safety-focused "Constitutional AI" models and developer tools.
A new brain-computer interface (BCI) decodes inner speech with up to 74% accuracy, offering hope for people with severe speech paralysis to communicate naturally.
🔍
Microelectrodes in the motor cortex capture neural signals related to imagined speech.
AI interprets inner speech from a vocabulary of up to 125,000 words.
A password system ensures decoding happens only when users allow it.
The BCI can distinguish between attempted and inner speech for more precise control.
Qwen-Image Edit, an AI model by Alibaba, offers advanced text-to-image editing with precise bilingual text control and semantic plus appearance editing.
🔍
Built on a 20B parameter foundation, it enables detailed edits preserving original image style.
Supports English and Chinese bilingual text editing, maintaining font and style.
Allows semantic edits like object rotation and style transfer alongside localized appearance changes.
Available via API with affordable pricing, targeting creators from professionals to casual users.
🤖 We scan thousands of AI stories.
You get the Top 7.
MIT researchers used sparse autoencoders and AI to interpret how protein language models encode biological features, improving model explainability and enabling better biomedical applications.
🔍
Sparse autoencoders expand neural representations, making individual protein features more interpretable.
AI assistant Claude helped link neural nodes to specific protein functions and families.
Understanding encoded features aids selection and tuning of models for drug and vaccine research.
This approach may unlock new biological insights from powerful protein models.
Waymo balances advanced AI exploration with a multi-sensor approach focused on LiDAR and radar to ensure robust safety in autonomous driving.
🔍
LiDAR and radar provide essential safety in extreme weather and unpredictable environments.
Waymo’s method contrasts Tesla’s camera-only approach, emphasizing sensor reliability.
AI models like Google Gemini and EMMA are explored but have data and sensor integration limits.
Waymo’s 100 million driverless miles demonstrate safety results while expanding to five cities.
Google and Meta have agreed to a six-year, $10 billion cloud deal where Meta will use Google Cloud’s infrastructure to support its AI-driven data centers and services.
🔍
This deal follows Google's recent major agreement with OpenAI for cloud services.
Meta plans to invest heavily in AI with capital expenditures up to $72 billion annually.
Meta is offloading $2 billion in data center assets to fund AI infrastructure.
Google Cloud saw a 32% revenue jump in Q2, driven by AI-related demand.
Nearly 300,000 user conversations with Elon Musk’s AI chatbot Grok have been inadvertently exposed in Google search results, raising serious privacy concerns.
🔍
Unique share links made Grok chats publicly searchable without users’ explicit knowledge.
Indexed conversations reveal sensitive requests including passwords, medical questions, and illicit instructions.
Experts warn this reflects a broader privacy crisis with AI chatbots sharing user data.
Similar issues previously occurred with OpenAI’s ChatGPT and Meta AI shared chats.
Sponsored Product: As an Amazon Associate, Brain Daily earns from qualifying purchases.
Sharing is caring ➦🤗
If you find this newsletter valuable, please forward it to a friend or colleague who may also benefit from staying up to date with the latest in AI.
If you have any suggestions, questions, or ideas, feel free to reply to this email. We’d love to hear from you and are always looking to improve our newsletter.