Good morning. It’s Wednesday, April 9th.

On this day in tech history: 2003: The Human Genome Project was completed, with 99% of the human genome sequenced to 99.99% accuracy.

In today’s email:

• GPT-4.1, o4-mini, and Full o3 Set for Release

• Delete IP Law for AI?

• Google’s Veo 2

• 3 New AI Tools

• Latest AI Research Papers

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

_{In partnership with NEBIUS}

Build AI Projects Instantly on GPU-Powered Notebooks

Run JupyterLab® in a fully managed environment on Nebius AI Cloud. Benefit from smooth performance on NVIDIA GPU servers — minimum complexity, maximum speed.

• Pre-installed PyTorch for instant model training

• Save your progress with built-in persistent storage

• Customize your environment easily with simple package management

• Choose GPU or non-GPU setups based on your workload

• Save money with affordable NVIDIA GPU infrastructure

Start coding now – no setup required. Pay only for what you use.

GPT-4.1, o4-mini, and Full o3 Set for Release This Week as OpenAI Updates Model Cards

OpenAI is set to release several new models this week, including GPT-4.1, GPT-4.1 mini, GPT-4.1 nano, o4-mini, and the full o3. The updates on the model cards confirm that OpenAI is pushing forward with its release schedule, while GPT-5 is also on the horizon.

However, OpenAI has reduced the time and resources allocated to AI safety testing. The Financial Times reported that staff responsible for evaluating risks and performance now have only days to conduct assessments, a sharp decline from previous practices.

The company is also rolling out A-SWE, an autonomous AI agent built for software engineering tasks like debugging and documentation. Yet past issues with similar tools cast doubt on its reliability. Meanwhile, OpenAI will soon require verified government-issued IDs from organizations to access advanced models via API, limiting each verification to once every 90 days.

Meanwhile, 12 former staffers have filed an amicus brief opposing the company's for-profit transition. The case, which challenges the conversion's legality, will proceed to trial in spring 2026. Read more.

Jack Dorsey and Elon Musk would like to ‘delete all IP law’

Jack Dorsey and Elon Musk’s call to “delete all IP law” reflects a broader clash between tech elites and traditional content regimes, as AI systems increasingly rely on copyrighted and user-generated data. Critics argue this stance disregards creators’ rights, while legal experts warn it erodes distinctions between human and machine output.

Ireland’s Data Protection Commission is also investigating platform 𝕏 for using European users’ public posts to train xAI’s Grok, potentially breaching GDPR. Meanwhile, xAI has modified Grok’s responses to political prompts—removing direct references to Musk and Trump in misinformation queries and recasting controversial topics as matters of opinion.

Despite surrounding controversies, Grok-3 and Grok-3 mini posted strong results in independent evaluations by Epoch AI. Grok-3, a non-reasoning model, topped the GPQA Diamond benchmark at 76%, outperforming GPT-4.5, Claude 3.7 Sonnet, and DeepSeek-R1. Grok-3 mini, with reasoning capabilities, led on OTIS Mock AIME with a 78% score and performed well on MATH Level 5 and the expert-tier FrontierMath. Read more.

Google's AI video generator Veo 2 is rolling out on AI Studio

Google’s AI rollout is accelerating—with polish, precision, and a few blind spots. Veo 2, its latest video generation model, is now available via AI Studio in the US, generating 8-second, 720p clips at 24 fps for $0.35 per second. It interprets both text and sketches, offering fluid, cinematic output that early testers say borders on production-grade. Japan Airlines is already using Veo 2 alongside Gemini to produce 60-second in-flight destination films, scripted with region-specific flair and assembled in under 15 hours using Pencil Pro’s RAG pipelines.

But while the outputs are smooth, governance is fraying. Gemini 2.5 Pro, Google’s top-tier model, was released without a mandatory safety report—raising questions about the widening gap between technical showmanship and responsible disclosure. Read more.

• Google, Nvidia invest in OpenAI co founder Ilya Sutskever’s AI startup Safe Superintelligence

• Alibaba’s Quark surpasses ByteDance’s Doubao, DeepSeek as China’s top AI app

• The delayed Siri Apple Intelligence features are now coming 'in the fall'

• Microsoft is about to launch Recall for real this time

• Watch: Lead Researcher of AlphaGo Reveals Insights on the Path to Superhuman Intelligence in New Podcast

• OpenAI co-founder Sutskever raises $2 billion for AI startup with no product

• Meta’s AI research lab is ‘dying a slow death,’ some insiders say—but Yann LeCun pushes back: ‘It’s a new beginning’

• Watch: Fourier unveils world's first open source humanoid robot, the Fourier N1

• High school student uses AI to reveal 1.5 million previously unknown objects in space

• Universal photonic artificial intelligence acceleration

• Watch: ARC-AGI-2 Overview (in-depth presentation)

• Amazon CEO sets out AI investment mission in annual shareholder letter

• Fintech founder charged with fraud after 'AI' shopping app found to be powered by humans in the Philippines

• World's first interactive 3D holographic display

  

  3 new AI-powered tools from around the web

📄 DeepSeek-R1 Thoughtology: Let's <think> about LLM Reasoning 

📄 FantasyTalking: Realistic Talking Portrait Generation via Coherent Motion Synthesis

📄 C3PO: Critical-Layer, Core-Expert, Collaborative Pathway Optimization for Test-Time Expert Re-Mixing

📄 Seaweed-7B: Cost-Effective Training of Video Generation Foundation Model

📄 Scaling Laws for Native Multimodal Models Scaling Laws for Native Multimodal Models

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!