Microsoft’s most capable Phi-4 AI model now rivals models 10x its size

Presented By Notion

Want unlimited AI, ASAP?

Get up to 6 months of Plus plan + unlimited AI free!

Launch and scale your startup faster with Notion.

Visit the Notion for Startups page to get the offer.

Get Notion AI for free

AI & Tech News

1. Gemini’s Next Upgrade Makes ChatGPT’s Memory Look Outdated (Source)

Google is rolling out major upgrades to Gemini that rival ChatGPT’s latest memory features. Soon, Gemini will remember past conversations and, more significantly, pull insights from Gmail, Calendar, YouTube, and other Google apps using a feature called “pcontext.”

This integration aims to make Gemini far more personalized and proactive—anticipating your needs and acting like a true digital assistant. Privacy controls will be in place, and more details are expected at Google I/O on May 20.

2. Zuckerberg Just Declared War on the Entire Ad Industry with ‘Infinite Creative’ AI (Source)

Mark Zuckerberg just revealed Meta’s bold vision to eliminate traditional advertising as we know it. In a candid interview, he said businesses will soon need no ad creatives, no targeting strategy, and no measurement tools Meta's AI will handle everything: generating ad content, targeting, iterating, and reporting results.

This concept, dubbed "infinite creative," threatens to disrupt agencies, brands, and decades of advertising infrastructure. While small businesses may benefit, industry leaders are alarmed—saying Meta is effectively asking advertisers to “trust the results we spit out

3. Google Now Puts Ads Inside AI Chatbot Conversations (Source)

Google has started placing ads inside chatbot conversations through its AdSense for Search network, partnering with AI startups like iAsk and Liner. This strategic shift helps Google maintain dominance in digital ads as generative AI disrupts traditional search.

The company aims to monetize AI interactions, even if it loses search traffic to AI-powered platforms like Perplexity. While startups benefit from new revenue streams, Google’s expanding influence—especially amid antitrust scrutiny—raises big questions about the future of ad-funded AI chat.

4. Google Upgrades Gemini Chatbot With Powerful Image Editing (Source)

Gemini chatbot now allows users to edit AI-generated and personal images natively within the app. Rolled out with support for 45+ languages, the update brings multi-step editing—letting users change backgrounds, swap objects, and enhance photos contextually. 

While powerful (and similar to ChatGPT’s recent update), the feature sparked concerns over deepfakes. In response, Google adds invisible watermarks and is testing visible ones to ensure transparency.

5. Meta Unveils Llama 4-Powered AI App to Rival ChatGPT (Source)

Meta has launched a standalone AI app powered by its new Llama 4 model, offering real-time, voice-first interaction and tight integration with Facebook, Instagram, and WhatsApp. It features full-duplex speech, personal responses using user preferences, and even smart glasses support.

A Discover feed allows sharing and remixing AI content, while upgraded web tools include image generation and document editing. With privacy controls and visible mic indicators, Meta aims to blend utility with user control—marking a bold step in AI-powered social experiences.

6. Startups Are Building Advanced AI Models Without Data Centers (Source)

Startups Flower AI and Vana have developed a new AI model called Collective-1 using a completely different method—training it on GPUs distributed across the internet instead of relying on massive data centers.

Flower AI created the technology to enable this decentralized training, while Vana provided private user data from platforms like X, Reddit, and Telegram, with user permission. The model, currently at 7 billion parameters, is small compared to leading models like ChatGPT, but the team is already working on models with 30 billion and 100 billion parameters. To make this possible, they built an open-source tool called Photon, which improves the efficiency of distributed training.

This approach could significantly lower the cost of building powerful AI, making it accessible to smaller companies, universities, and countries without expensive infrastructure. It also opens the door to training models using private or sensitive data that wouldn't normally be centralized—like in healthcare or finance—while giving users more control and even ownership over how their data is used. Experts believe this shift in how AI is trained could disrupt the current power held by big tech companies in the AI space.

Thanks for your time!

Shailesh