• AI Academy
  • Posts
  • 🤖 OpenAI Finally Remembers the "Open" in Their Name

🤖 OpenAI Finally Remembers the "Open" in Their Name

Plus: Zuckerberg's Vision Quest and Anthropic's Obvious Advice

Hello AI Enthusiast,

Our AI news newsletter will take a break for the next two weeks, but you'll continue receiving our Educational newsletter every Tuesday with practical AI insights and training tips. Before we pause, this week Meta's Zuckerberg is painting grand visions of "personal superintelligence" while chasing the next platform shift, Anthropic released a safety framework that feels more like a corporate checklist than practical guidance, and OpenAI finally remembered the "Open" in their name by releasing their first public models since GPT-2. Let's break down what's happening.

The Big Picture 🔊

OpenAI Releases First Open-Weight Models Since GPT-2

OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2. These mixture-of-experts models claim to match OpenAI's own o3-mini and o4-mini on reasoning benchmarks while running on consumer hardware. Both show their full chain-of-thought reasoning and support adjustable effort levels.

Helin Yontar
Helin YontarCPO and Polyglot

OpenAI finally had to deliver on the "Open" in their name. They're basically releasing decent models they had sitting around - classic garage cleaning strategy.

The timing makes sense with more companies wanting on-premise solutions for data control, though most aren't equipped to handle the cybersecurity that comes with running their own models.

While more companies are considering on-premise solutions for data control, most still struggle with basic prompting on existing tools. Our Corporate AI Training teaches teams to work effectively with any AI platform - whether it's OpenAI's API, their new open models, or whatever comes next.

Meta’s Vision for "Personal Superintelligence"

Mark Zuckerberg claims Meta will develop "personal superintelligence" - AI assistants that deeply understand individual users through smart glasses and other devices. He contrasts this with competitors focused on work automation, positioning Meta's approach as empowering people rather than having them "live on a dole" of AI output.

Gioele Mottarlini
Gioele MottarliniCOO and Image Addict

Follow the money. After Apple's privacy changes hurt Facebook's ad targeting, Meta learned they can't depend on platforms they don't control. They blew $10 billion annually on VR trying to own the next platform. Now they're betting AI is that shift.

When people chat with ChatGPT instead of scrolling Instagram, Meta loses attention - their whole business.

Anthropic Releases Safety Framework for AI Agents

Anthropic published a framework for developing "safe and trustworthy" AI agents that can handle complex tasks autonomously with minimal human input. The framework outlines five principles: keeping humans in control, ensuring transparency in agent behavior, aligning agents with human values, protecting privacy, and securing against attacks like prompt injection.

Gianluca Mauro
Gianluca MauroFounder and AI Rockstar

The principles are obviously right - "keep humans in control," "be transparent" - but they're missing the meat. How do you actually balance autonomy with oversight? When should an agent ask permission versus just act?

These are genuinely tricky problems that deserve more than bullet points. It feels more like brand-building than practical help for developers facing these challenges.

Bits and Bobs 🗞️

  • OpenAI is introducing a new 'study mode' in ChatGPT to enhance learning by guiding users through problems with step-by-step guidance rather than just providing answers.

  • Anthropic launched Claude Opus 4.1, an upgraded AI model from Claude Opus 4, enhancing agentic tasks, real-world coding, and reasoning performance.

  • Anthropic has enhanced its Claude artifacts, allowing users to upload PDFs, images, code files, and more to AI-powered applications.

  • Google's Gemini 2.5 Deep Think, now available for AI Ultra subscribers, utilizes advanced parallel thinking techniques and extended "thinking time" for enhanced creative problem-solving.

  • Google has introduced Google Earth AI, a collection of geospatial models and datasets designed to address critical global issues.

  • Alibaba and Zhipu AI have developed the Wan2.2 series, to generate video from text and images more efficiently.

  • OpenAI is launching Stargate Norway, its first AI data center in Europe, aiming to boost AI infrastructure with 230MW of capacity using renewable energy.

  • Anthropic has revoked OpenAI's access to its Claude AI models, citing a violation of terms as OpenAI used Claude to compare performance with its own models.

  • Elon Musk's AI company xAI has launched Grok Imagine, an image and video generator, to iOS app subscribers, allowing users to create content, including some NSFW material with restrictions.

On the Podcast 🎧

This week's story shows what happens when someone with zero technical background decides to solve a deeply personal problem with AI. Matsume, a theater maker, used our AI Agent Bootcamp to build a memory companion tool for her stroke-survivor father. Listen Here

From Our Founder’s Channels 🤳

Watch Gianluca's TikTok showing you how to spot when that social media post was probably written by AI instead of a human.

@gianluca.mauro

I’m about to ruin your entire social media experience #Ai #learnontiktok #artificialintelligence #business #machinelearning #product #ux #... See more

LOLgorithms 😂

Not all devices will be able to host OpenAI’s new open-weight models.

That's a wrap on our newsletter! Before you go, here’s a quick recap of our offerings:

  • AI Academy Membership: Get 12 months of access to all our cohort-based programs, live webinars, on-demand courses, and tutorials.

  • AI Agent Bootcamp: Accelerate processes and solve business problems by mastering prompts and building AI Agents, without coding.

  • Practical Introduction to ChatGPT: A free course on using ChatGPT confidently, understanding its workings, and exploring its potential.

  • Customized Corporate Training: Equip your team with the skills they need to unlock the potential of AI in your business.

Catch you next week! 👋