Hey AI Enthusiasts!

This week’s roundup is packed with fresh breakthroughs and bold moves—from Apple challenging what “reasoning” in AI really means, to OpenAI and Meta racing ahead with new models and ambitious research labs.

Let’s dive in!

In today’s insights:

🍏 Apple says AI “reasoning” is mostly pattern-matching
🚀 OpenAI launches O3-Pro, its sharpest reasoning model to date
🧠 Meta sets up a new lab aimed at achieving super-intelligence

_{Read time: 5 minutes.}

Apple
 🔥 Apple’s new research paper says AI reasoning isn’t all it’s cracked up to be.

The AI Field: Apple just lobbed a reality-check grenade into the AI hype cycle—right on the eve of WWDC. In a peer-review-bound study, its researchers argue that today’s buzzy “reasoning” models buckle as soon as problem complexity rises, no matter how many tokens you give them. 

Details:

• Three performance regimes: On easy tasks, plain LLMs often edge out reasoning-tuned models; on medium tasks, reasoning models win; on hard tasks, everyone crashes to nearly 0 % accuracy.

• Reverse-effort effect: Instead of thinking harder, models shorten their chain-of-thought when puzzles get tougher, wasting unused context window.

• Algorithmic hints don’t help: Even when fed the correct step-by-step algorithm, models still flunk high-complexity instances.

• Competitor call-outs: Apple openly measured models from OpenAI, Anthropic, Google, and DeepSeek—naming names when they collapsed.

Why This Matters: This study is likely to shake up AI benchmarks by scoring reasoning effort, not just final answers. Expect “Apple Intelligence” to emphasize context-based retrieval over fully autonomous agents, injecting realism into AGI timelines and hinting that future systems might rely on symbolic planners rather than just bigger transformers.

👉 Read more here

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI

OpenAI
🚨 OpenAI just dropped O3-Pro: a powerful upgrade to its reasoning AI model

The AI Field: OpenAI just released O3-Pro, a turbocharged successor to its popular O3 AI reasoning model, aimed specifically at tackling complex math, science, and coding tasks with unprecedented accuracy and depth.

Details:

• OpenAI says O3-Pro demonstrates significantly better reasoning skills, especially in advanced mathematics and coding.

• The model introduces enhancements specifically designed for deeper logical understanding, reducing the mistakes typical of earlier AI versions.

• Early testers have highlighted remarkable improvements in O3-Pro's problem-solving accuracy, particularly in scenarios involving multiple steps and logical deductions.

• O3-Pro could impact fields such as scientific research, software development, and advanced analytics, boosting productivity and accuracy.

Why This Matters: This upgrade reinforces OpenAI’s commitment to advancing the frontier of AI reasoning. O3-Pro isn't just about better performance—it's about pushing toward genuinely reliable AI tools that professionals in technical fields can depend on. Expect ripple effects across tech, education, and research industries as AI becomes a more robust partner in tackling complex problems.

👉 Read more here

Meta
🗞️ Meta is Launching an AI Lab Dedicated to Superintelligence

The AI Field: Meta is reportedly setting up a cutting-edge AI research lab aimed specifically at developing superintelligence—AI systems surpassing human-level intelligence—as part of its renewed effort to lead the rapidly evolving AI race.

Details:

• Meta’s new AI lab will focus exclusively on the ambitious goal of superintelligent AI, pushing beyond current generative and predictive models.

• The initiative includes high-level collaborations, tapping into some of the brightest minds in AI research.

• Alexandr Wang, CEO of Scale AI, is rumored to be involved, bringing his expertise and leadership to Meta’s AI ambitions.

• The lab represents Meta’s response to growing competition from leading AI players like OpenAI, Google, and Microsoft.

Why this matters: Meta’s strategic pivot to focus explicitly on superintelligence underscores just how fierce the competition for AI dominance has become. The outcomes of this new lab could significantly reshape the future landscape of technology, ethics, and society as we approach unprecedented levels of artificial intelligence.

👉 Read more here

🧠 Unlock Hidden AI Prompts Used by the Top 1%

Discover a set of ChatGPT prompts trusted by industry leaders and business owners to streamline operations, enhance decision-making, and boost overall productivity. This exclusive collection includes over 100 specialized prompts designed specifically to help businesses thrive using AI.

Get instant lifetime access and start integrating these powerful tools into your daily workflow today.

👉 Explore the Complete Bundle

📝 Fireflies.ai – AI-powered note-taker that transcribes, summarizes, and organizes meetings automatically, ideal for teams enhancing productivity.

🤖 CustomGPT – Build personalized AI chatbots trained on your own content effortlessly, ideal for businesses enhancing customer engagement.

✍️ Writesonic – AI writing tool that quickly generates engaging content like blogs, ads, and emails, perfect for marketers and content creators.

🎬 Pictory AI – Transforms articles into professional-quality videos in minutes, ideal for creators and marketers aiming to boost engagement.

📹 HeyGen – Instantly creates realistic AI avatar videos from text, perfect for businesses and creators looking to enhance visual storytelling.

📣 AdCreative.ai – Generates targeted ad copy and designs by analyzing product details, ideal for small businesses running campaigns.

📅 FeedHive – AI-driven social media management tool offering content suggestions and analytics, great for freelancers managing accounts.

✍️ Copy.ai – AI writing assistant that generates social media posts, blog ideas, and marketing copy, saving time for content creators.

Nvidia’s Blackwell GPUs Halve Training Time for Large Models: Early MLPerf results show Blackwell chips can train complex AI systems roughly twice as fast as their predecessors, reinforcing Nvidia’s hardware dominance.

OpenAI hits a $10 billion annual-revenue run rate: Reuters reports that OpenAI’s revenue pace has nearly doubled since December, underscoring how quickly demand for its models is growing.

Disney and Universal sue Midjourney for AI “rip-offs” of blockbuster characters: A detailed report describes how the studios allege Midjourney enables near-identical recreations of Marvel, Star Wars and other IP, raising major copyright questions for generative AI.

Microsoft’s new Copilot Vision can “see” the apps on your Windows desktop: The Verge explains how the on-device feature lets Copilot visually analyze whatever window you choose, delivering context-aware help without constant screen scraping.

Golden Nuggets

🍏 Apple Questions AI “Reasoning” — A fresh research paper from Apple argues that today’s leading models are often just pattern-matching, not truly reasoning, sparking debate across the field.

🚀 OpenAI Drops O3-Pro — OpenAI’s newest release delivers a hefty boost to its reasoning model, with sharper math, science, and coding skills.

🧠 Meta Builds a Super-Intelligence Lab — Meta is setting up a dedicated research unit to chase AI that surpasses human intelligence, aiming to reclaim ground in the AI race.

Until next time!
Olle | Founder of The AI Field