Good morning. It’s Friday, March 28th.

On this day in tech history: In 1979, the Three Mile Island nuclear plant near Middletown, Pennsylvania, experienced a partial meltdown in its Unit 2 reactor, marking the most serious accident in U.S. commercial nuclear power plant history. ​

In today’s email:

• GPT-4o’s Major Upgrade

• Anthropic’s Claude 3.7 Coming Soon?

• Voice AI Just Became Way More Human

• 4 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Make Your Code Brilliant with Diamond

Diamond delivers instant, actionable, codebase-aware AI feedback on every pull request.

• Features: Zero setup, high-signal reviews, one-click fixes, custom rules.

• Pricing: Free for 100 PRs/month | $20/user/month for unlimited | Custom Enterprise plans.

• FAQ: Easy GitHub setup. No data training. Questions? .

Supercharge your team: Catch bugs fast, enforce standards, ship smarter.

"Diamond found bugs no human could spot." – Konrad K., Isometric

OpenAI upgrades GPT-4o and bets on interoperability with Anthropic’s MCP amid viral overload

OpenAI is undertaking a dual-pronged upgrade as it seeks to enhance AI interoperability and refine its flagship models. In a significant strategic move, CEO Sam Altman announced that OpenAI is integrating Anthropic’s Model Context Protocol (MCP) across its ecosystem, including the ChatGPT desktop app and Responses API. MCP—now featuring OAuth 2.1-based authorization, streamable HTTP transport, and JSON-RPC batching—bridges AI models with external business tools, ensuring cleaner, more secure two-way data flows.

At the same time, OpenAI has updated GPT-4o to boost instruction-following and coding capabilities while reducing extraneous emoji use.

However, a viral surge in Studio Ghibli-style image generation has overwhelmed the system, delaying free-tier access as paying users max out capacity, with Altman remarking, “our GPUs are melting.” These upgrades come as OpenAI projects revenue of $12.7 billion this year, a stark leap from last year’s $3.7 billion, with eyes set on $29.4 billion next year. Read more.

Anthropic may soon launch Claude 3.7 Sonnet with 500K token context window

Anthropic appears poised to release Claude 3.7 Sonnet with a massive 500K token context window, a leap from its current 200K limit. Feature flags hint at an imminent rollout, potentially offering enterprise users the ability to process vast datasets and full-scale codebases without fragmenting sessions. This upgrade could reduce reliance on retrieval-augmented generation (RAG), mitigating context distortion in long-form analyses.

Early signs of the update have surfaced in Cursor’s IDE, where a “Claude 3.7 Sonnet MAX” option recently appeared. If broadly implemented, this expansion could redefine AI-assisted coding, legal analysis, and research workflows. It also aligns with the rise of “vibe coding,” a trend popularized by Andrej Karpathy, where AI-driven code generation eliminates the need for manual programming. Details on general availability remain unclear. Read more.

Groq and PlayAI just made voice AI sound way more human — here's how

Groq and PlayAI have partnered to launch Dialog, a next-generation text-to-speech (TTS) model engineered for hyper-realistic, low-latency voice AI. Running on Groq’s high-speed inference platform, Dialog claims to outperform competitors in both responsiveness and natural speech patterns, thanks to its adaptive speech contextualizer (ASC)—a system that maintains conversational flow, adjusting tone and prosody dynamically.

Benchmark tests show Dialog outperforming ElevenLabs, with up to 140 characters generated per second—nearly 10 times faster than real-time. The model debuts with Arabic-language support, positioning Groq to capture a key global market. This comes as the company secures a $1.5 billion Saudi investment to expand AI infrastructure, including a new data center in Dammam. Read more.

• Anthropic makes a breakthrough in opening AI's 'black box'

• Amazon’s AI-powered ‘Interests’ feature automatically finds new products that match your passions and hobbies

• Databricks embeds Anthropic’s Claude and introduces TAO for real-time AI optimization

• Satya Nadella: DeepSeek is the new bar for Microsoft’s AI success

• Google Maps turns screenshots into trip plans with new AI upgrade

• Open Deep Search takes on closed AI search models with a plug-and-play, open-source framework

• Nvidia G-Assist uses 8 billion parameter model and runs without cloud connection

• ‘Naisha’ trailer introduces India’s first AI-powered movie stars

• Amazon's latest AI feature turns your hobbies into hyper-specific shopping suggestions

• China unveils world-first emergency rescue robot with quadrupedal, wheeled, tracked modes

• A breakthrough moment: Researchers discover new class of antibiotics

• Pika introduces AI feature letting users meet their younger selves

• Imandra's new AI coding assistant CodeLogician uses 'reasoning' to guarantee the accuracy of its code

• Luma AI launches Edit Thread in DreamMachine for simplified asset management

• Nvidia is reportedly in talks to acquire Lepton AI

• OpenAI's viral Studio Ghibli moment highlights AI copyright concerns

• OpenAI Discusses Building Its First Data Center for Storage — The Information

• Krisp is using AI to help Indians sound like Americans on calls

📄 Long-Context Autoregressive Video Modeling with Next-Frame Prediction

📄 Dita: Scaling Diffusion Transformer for Generalist Vision-Language-Action Policy ​​

📄 LEGO-Puzzles: How Good Are MLLMs at Multi-Step Spatial Reasoning? 

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!