In partnership with

Good morning. It’s Monday, March 31st.

On this day in tech history: March 31 is recognized as World Backup Day, encouraging individuals and organizations to back up their data to prevent loss. ​Back it up!

In today’s email:

• X Merges With XAI

• Google’s Gemini 2.5 Released For Free Tier

• OpenAI Adds Thinking Sliders

• 4 New AI Tools

• Latest AI Research Papers

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

Elon Musk merges xAI and X in all-stock deal

Elon Musk is folding 𝕏 into xAI in an all-stock maneuver that pegs his AI venture at $80 billion and the embattled social platform at $33 billion, debt-adjusted. The deal tightens the feedback loop between AI development and user data, positioning xAI to refine its models at scale while 𝕏 morphs into an AI-native digital domain.

Despite 𝕏’s past struggles with advertiser exodus and content moderation concerns, reports suggest its valuation may have rebounded due to Musk’s political ties. The merger follows xAI’s recent acquisition of a generative video AI startup and its $30 billion AI infrastructure fund with Nvidia. xAI, which has raised at least $12 billion, also launched Grok 3, one of its most advanced AI models. Musk envisions xAI developing superintelligent AI to tackle scientific challenges like dark matter and the Fermi paradox. Read more.

Google is surprisingly rolling out Gemini 2.5 Pro (exp) to free users, adds Canvas on the web

Google has unexpectedly rolled out Gemini 2.5 Pro (experimental) to all free-tier users, expanding access to what it calls its “most intelligent AI model.” Previously exclusive to Gemini Advanced subscribers at $19.99 per month, the model integrates advanced reasoning capabilities rather than offering them as standalone variants. It now supports apps, file uploads, and the newly introduced Canvas tool on the web, enhancing its interactive capabilities.

Currently topping the LMArena leaderboard and excelling in math and science benchmarks, Gemini 2.5 Pro is also undergoing improvements in coding performance. The move suggests Google’s broader push to democratize AI, as recent months have seen premium features like Deep Research, document analysis, and generative image tools trickle down to all users. Whether the stable release will remain free or revert to a paid tier remains uncertain, but for now, Google is accelerating mass adoption of its most advanced AI to date. Read more.

OpenAI reworks image policies, tests thinking sliders, and targets $40B funding

OpenAI is closing in on a $40 billion funding round, with SoftBank expected to contribute $30 billion, and Microsoft and others adding $10 billion, including an upfront $10 billion tranche. SoftBank has also pledged $3 billion annually for OpenAI tech and launched a Japan-focused joint venture.

On the policy front, OpenAI is loosening GPT-4o’s image generation rules, replacing blanket bans with opt-outs for public figures and refining filters to remove implicit bias. The move follows mounting competition from xAI’s Grok 3, which enforces fewer restrictions. While OpenAI still blocks artist names, user feedback is shaping ongoing adjustments.

OpenAI is also reportedly testing new ChatGPT web app features, including a “thinking slider” that may let users control the depth of model reasoning, potentially aligning with GPT-4o’s reasoning tiers. Additional rumoured updates include a notifications feed, a dictation feature on desktop, and enhanced security warnings for external links.

There are also references to “deep research share links,” hinting at possible collaboration tools. A system prompt update for the Moonshine demo may also adjust ChatGPT’s response style based on user memory depth. OpenAI has not confirmed these changes. Read more.

_Sponsored:

Mindstream is how 200,000+ readers keep on top of AI. The daily newsletter delivers actionable insights and the news you need to know to stay ahead of the curve and maximise AI in your daily life. The best part? It’s completely free. Get smarter on AI in just a few minutes each day. Join the club!

• CoreWeave stumbles in debut as AI cloud IPO falls short—but many high-profile tech IPOs remain in the pipeline

• ElevenLabs launches Actor Mode, letting users direct AI voices with their own

• Apple reportedly revamping Health app to integrate AI coach, food tracking, and doctor-guided advice

• Peter Thiel reportedly warned Sam Altman about AI safety conflicts shortly before the OpenAI crisis

• Study finds that fewer documents can lead to better performance in RAG systems

• This watchdog is tracking how AI firms are quietly backing off their safety pledges

• Baidu's ERNIE 4.5 model just destroyed OpenAI's GPT-4.5 at Chinese chess

• China's Zhipu AI launches free AI agent, intensifying domestic tech race

• There are 372 AI unicorns right now, according to PitchBook. Here’s what that means

• Early preview of the "Discover Sources" feature on NotebookLM

• Researchers warn of 'catastrophic overtraining' in LLMs

• Perplexity CEO denies having financial issues, says no IPO before 2028

📄 UI-R1: Enhancing Action Prediction of GUI Agents by Reinforcement Learning

📄 Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation​​

📄 AdaptiVocab: Enhancing LLM Efficiency in Focused Domains through Lightweight Vocabulary Adaptation

📄 ORIGEN: Zero-Shot 3D Orientation Grounding in Text-to-Image Generation

📄 Exploring Data Scaling Trends and Effects in Reinforcement Learning from Human Feedback

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!