Daily AI digest for operational businesses.

Tencent has released an open-source language model called Hy3, which is surprisingly efficient despite its large size. Hy3 has 295 billion parameters, but only 21 billion are active at a time.

GPT-4 held the top spot in AI model capabilities for about a year, but newer models have a much shorter reign. The current top model, Claude 3 Opus, has already been surpassed.

Players 1st has launched an AI assistant that uses the world's largest collection of golf intelligence to provide users with personalized golf advice.

Alibaba has stopped using Anthropic's Claude Code due to concerns over a hidden backdoor that could detect its use in China. This decision reflects a growing rift between the two companies, with Alibaba now directing employees to use Qoder instead.

Google has released two new AI models, which could impact the company's stock performance. The new models are designed to improve various Google products and services.

An open-source terminal app integrates AI into any terminal, staying invisible until needed. It supports popular AI agents like Claude and Codex.

Microsoft is investing $2.5 billion in a new AI group to help businesses adopt and use artificial intelligence. This investment aims to make AI more accessible and easier to implement.

Google released a commercial imagining the founding fathers using Google Workspace and Gemini to collaborate on the Declaration of Independence. The ad is meant to showcase the tools' capabilities.

A developer used AI tools to port the 2003 PC game Command & Conquer to native iOS in just a few hours. The first build took only 40 minutes.

A user expresses concern about sharing AI research with large companies due to potential misuse of intellectual property.

Mistral's CEO warns against relying on closed AI models, citing concerns about customer data and potential competition. Mistral is prioritizing EU sovereignty over raw AI performance.

Mistral AI released an open-source model for formal verification called Leanstral 1.5, which has passed rigorous math benchmarks and caught five previously unknown bugs in open-source code.

Greg Brockman, OpenAI cofounder, predicts a future where software interfaces are minimal or non-existent, and users interact with intelligent agents. This vision is far from being realized with OpenAI's current technology.

OpenAI, a leading AI company, may not have visited a key UK site for a major investment, sparking concerns about the deal's legitimacy.

Anthropic is launching its own drug development program to tackle diseases that pharmaceutical companies consider unprofitable. This could lead to new treatments for conditions that currently lack effective options.

A problem has been identified in the GPT-5.5 Codex model, which may be causing its performance to degrade. The issue is related to how the model processes and clusters reasoning tokens.

A new open-source tool called pxpipe converts text prompts into compact PNGs to reduce costs for using AI tools like Claude Code and Fable 5. This can save up to 70% on costs, but may impact accuracy and speed.

Anthropic developer Thariq Shihipar shares techniques for improving Fable 5 prompts by identifying and addressing the user's own blind spots. This involves using methods like blindspot passes and structured interviews to uncover unconscious knowledge gaps.

Mistral AI is a new AI technology company that's raising funding to make frontier AI accessible to everyone. It offers some open source AI models.

Anthropic's AI tool, Claude Code, is causing tension with Chinese companies. Restrictions are being imposed on both sides of the Pacific.

Meta's AI team claims their new model has matched the capabilities of OpenAI's dominant GPT-5. This comes as Meta prepares to release its own large language model.

Meta Platforms plans to spend $135 billion on AI in 2026, a staggering investment that raises questions about its motivations.

Microsoft is revamping its Copilot AI tool, merging consumer and enterprise versions into one app. It's also introducing new AI agents called AutoPilot to handle tasks in the background.

Bridgewater and a startup from a former OpenAI CTO claim their AI model outperforms GPT and Claude in finance tasks, but the results haven't been verified. The model is said to be more accurate and cheaper to run.