The #1 AI Newsletter for Business Leaders

Join 400,000+ executives and professionals who trust The AI Report for daily, practical AI updates.

Built for business—not engineers—this newsletter delivers expert prompts, real-world use cases, and decision-ready insights.

No hype. No jargon. Just results.

Subscribe free—trusted by leaders

Today:

• China Claims Self-Improving AI Breakthrough

• OpenAI Launches ChatGPT Study Mode

• Meta Allows AI in Interviews

• NotebookLM Adds AI Video Overviews

• Bowen Zhang Leaves Apple for Meta

"AlphaGo Moment" For Self Improving AI... can this be real?

A new Chinese paper claims a breakthrough in AI self-improvement, introducing ASI Arch, an autonomous system that designs and tests its own neural architectures. It reportedly ran 2,000 experiments and produced 106 state-of-the-art attention models, suggesting a "scaling law for scientific discovery." 

While some experts doubt its validity due to questionable data filtering, the approach could accelerate recursive AI innovation if verified, pushing AI closer to self-directed research and faster advancements.

WATCH THE VIDEO ON YOUTUBE

OpenAI Launches ChatGPT Study Mode

OpenAI has launched "study mode" in ChatGPT, a feature that teaches step by step instead of just giving answers. It uses guided questions, personalized feedback, and interactive prompts to help students understand complex topics. Built with input from teachers and learning experts, study mode aims to boost critical thinking and active learning. OpenAI plans to refine the feature using student feedback and integrate it more deeply into future AI models.

Why It Matters

1. AI in Education: It shows how AI can move beyond answering questions to becoming a real teaching assistant.
   

2. Model Design for Learning: Study mode tests how AI can mimic human tutoring strategies, informing future AI development.
   

3. User Feedback Loop: Iterative improvement of study mode highlights how direct user feedback can shape smarter, more effective AI systems.

Meta Allows AI in Interviews

Meta will allow certain job candidates to use AI tools during coding interviews, reflecting real-world developer workflows where AI-assisted coding is becoming standard. Internal messages reveal that employees are testing “mock AI-enabled interviews.” This move highlights Silicon Valley’s push to embrace AI as a core part of engineering roles. Meta aims to hire developers skilled at combining human logic with AI-assisted coding for greater efficiency and productivity.

Why It Matters

1. AI-Integrated Hiring: Marks a shift toward evaluating coding skills alongside AI-assisted problem-solving.
   

2. Industry Trend: Signals that major tech companies now consider AI coding tools standard in development.
   

3. Workforce Evolution: Encourages engineers to adapt to AI-powered workflows, shaping future software engineering practices.

NotebookLM Adds AI Video Overviews

Google’s NotebookLM now includes “Video Overviews,” which turn user documents into AI-generated slide presentations with voiceovers, images, and charts to explain complex topics. This complements its Audio Overviews and allows more customization, such as setting learning goals or tailoring content for specific audiences. The update also introduces a flexible Studio panel supporting multiple outputs like mind maps and reports, with NotebookLM increasingly positioned as a learning and educational tool.

Why It Matters

1. AI-Driven Content Creation: It shows how AI can turn static documents into engaging multimedia lessons.
   

2. Education and Learning: Highlights AI’s growing role as a teacher’s assistant for breaking down complex subjects.
   

3. User-Centric AI Tools: Demonstrates how AI can adapt to individual learning needs and multiple formats simultaneously.

🧠RESEARCH

The Geometry of LLM Quantization: GPTQ as Babai's Nearest Plane Algorithm

This paper shows that GPTQ, a method for shrinking large AI models, works the same as a well-known math algorithm called Babai’s nearest plane. This link gives GPTQ a clear geometric meaning and proven error limits, helping researchers build better, faster models using ideas from lattice-based math.

Rep-MTL: Unleashing the Power of Representation-level Task Saliency for Multi-Task Learning

Rep-MTL improves multi-task learning by focusing on how tasks interact in the shared representation space rather than only adjusting losses or gradients. It measures task importance, reduces harmful interference, and encourages helpful knowledge sharing. Tests on multiple benchmarks show Rep-MTL delivers better performance and efficiency compared to traditional methods.

Agentic Reinforced Policy Optimization

ARPO is a new reinforcement learning method designed for AI agents that use tools over multiple steps. It tackles uncertainty that arises after tool interactions by adapting how the agent explores and learns. Tested on 13 benchmarks, ARPO outperforms existing methods while using fewer tool interactions, making training more efficient.

🛠️TOP TOOLS

Adobe Audio Enhancer - Clean up audio for free with Adobe Podcast AI

Mangoai.co - Create AI Videos in Minutes with AI avatars and voices.

AKOOL – AI Face Swap App - Premium Quality AI Face Swap Tool

Let’s Enhance! - Get clear, high-res images with AI

LeiaPix - The AI platform converting images & videos into 3D

📲SOCIAL MEDIA

🗞️MORE NEWS

• Apple has lost its fourth AI researcher in a month, as Bowen Zhang left to join Meta’s new superintelligence team. The departure highlights Apple’s ongoing struggle to keep talent for its AI platform.

• Meta is aggressively recruiting AI experts for its new superintelligence lab, now targeting Mira Murati’s startup. Over a dozen staffers have been approached, with one candidate reportedly offered a staggering package worth over $1 billion.

• Anthropic is close to raising $3–5 billion, pushing its valuation to $170 billion, nearly triple its March valuation of $61.5 billion. Iconiq Capital leads the round, with possible Middle Eastern investment despite CEO Dario Amodei’s concerns.

• AI chip startup Groq is negotiating a $600 million funding round that would value it at nearly $6 billion—double its 2024 valuation. The raise, led by Disruptive, follows major partnerships with Bell Canada and Meta.

• Writer has launched “Action Agent,” an autonomous AI system that completes complex business tasks across 600+ tools without human help. It outperforms OpenAI benchmarks, offers enterprise-grade security, and is free for existing customers, targeting Microsoft and Google’s enterprise AI market.

• Bragi is partnering with OpenAI to add generative AI features to headphones and audio products. Its upcoming ChatAI app, launching summer 2025, enables voice commands for tasks like note-taking, aiming to create fast, personalized audio experiences.

• Microsoft is negotiating with OpenAI to secure long-term access to its AI technology, even after OpenAI reaches advanced general intelligence (AGI). The deal may include a larger Microsoft stake and aims to overcome regulatory and structural hurdles.