🤔Apple Challenges AGI Claims

Today we will discuss:

• Sponsored: Picsart—Creative APIs to add AI-powered visual magic quickly and at scale

• AI isn’t reasoning its way to AGI yet—Apple researchers say today’s models collapse under complex tasks

• Don’t miss this week’s Workflow Wednesday—work smarter with AI using SheetsResume to generate a polished resume in minutes, plus write sharper messages with Grammarly AI in Gmail and Slack, and clean up stale GitHub branches automatically with GPT and Make.com; now available at only $20/month! 

• Meta targets scale AI in Billion-Dollar play—Zuckerberg’s AI ambitions may push Meta into one of the biggest private AI deals ever

• Spotlight: Alumnium—The only open-source AI test automation library for both web and mobile

• In other AI news—Elon Musk’s feud with Donald Trump could complicate xAI’s $5B debt deal, Alibaba releases new open-source embedding models to cement its global lead, and RedNote enters the AI race with its own model it claims outperforms Alibaba and DeepSeek

• Saved the best for last—10 must-try AI tools

Add Smart Visual Processing in 3 Minutes

Ship faster with tools your users will notice.

Creative APIs from Picsart let you add visual magic like AI Background Removal or Auto Crop in just a few lines of code.

REST APIs, fully documented, built for scale.
Try it live. No fluff. No waiting.

💡 Explore the docs
⚡ 200 free credits. No card. Just go → Sign up now

Top AI models fail to match human-like reasoning Apple finds

Key Points 

• Apple tested top models on custom puzzles and found their reasoning broke down as complexity increased.

• The study shows models fail to apply consistent logic especially beyond basic benchmarks or familiar patterns.

📰News - Despite growing claims that artificial general intelligence or AGI is just a few years away Apple researchers say there’s still a significant gap between today’s AI models and human-like reasoning. 

In a new paper titled The Illusion of Thinking the team tested leading large language models and found their reasoning abilities to be inconsistent, shallow, and often misleading.

🧐Beyond the Benchmarks - Models like OpenAI’s o1 and o3-mini Claude Sonnet and DeepSeek’s R1 and V3 were put through custom designed puzzle tasks that went beyond standard math and coding benchmarks. While these models often got the right answer early on they tended to overthink the problem and arrive at the wrong conclusion especially as complexity increased.

The researchers found that large reasoning models appear to mimic reasoning patterns rather than actually understand or generalize them. They also don’t reliably use explicit algorithms and struggle with exact computation. In other words the models often look like they are reasoning but fall apart when the task requires deeper logic.

These findings challenge the optimistic timelines from AI leaders like Sam Altman and Dario Amodei who have said AGI is just around the corner. Apple’s results suggest we may be underestimating how much progress is still needed before AI can truly reason like a human.

Read the full story here.

We’ve just launched the 22nd edition of Workflow Wednesday for AI-minded professionals like you—actionable AI workflows delivered straight to your inbox.

This week’s topic: Integration

Join us here

Meta may join big tech’s billion-dollar AI investment club

Key Points 

• Meta is reportedly in talks to invest over $10 billion in Scale AI, its biggest external AI deal yet.

• Meta and Scale already collaborate on military AI tools, including a defense-focused version of the Llama 3 model.

♨️News - Meta is in discussions to invest more than $10 billion in Scale AI, in what could become its largest external AI investment to date. If the deal goes through, it would also rank among the biggest private fundraising rounds ever in the AI space.

Scale AI is known for its data labeling services, which are used by companies like Microsoft and OpenAI to train their models.  Last year, Scale brought in $870 million in revenue and is projecting $2 billion this year.

👨‍💻Why this matters - This move would be a rare one for Meta, which has mostly relied on internal research and an open development approach to advance its AI efforts. While other tech giants like Microsoft, Amazon, and Google have funneled billions into AI startups, often by offering access to cloud computing, Meta has stayed largely in-house. Unlike those companies, Meta does not have a cloud business, and it is not yet clear what form this investment would take.

🤓See also - Meta and Scale already work together on Defense Llama, a specialized version of Meta’s Llama 3 model built for military applications. The two companies also share an interest in defense tech. 

Just last week, Meta announced a partnership with defense contractor Anduril to co-develop military tools, including an AI-powered headset for virtual and augmented reality use. With CEO Mark Zuckerberg committing up to $65 billion toward AI projects this year, Meta’s ambitions in the space are only getting bigger.

Read the full story here.

Alumnium is changing how devs automate testing. 

It’s the only open-source, AI-powered library that supports both web and mobile — no more writing brittle scripts or juggling multiple frameworks.

Fast-growing, dev-loved, and built for real-world testing.

If you're into clean automation and smarter tooling:
Give it a ⭐ on GitHub → github.com/alumnium-hq/alumnium