Edvance AI Weekly Digest
Reasoning models. Virtual teammates. A new phase of AI is here.
This week at a glance…
1️⃣ Meta launches an AI app – with a head start on knowing you
Meta just dropped a standalone app for its AI assistant, following its rollout across Instagram, WhatsApp, and more. The twist? It already knows a lot about you. By tapping into your Facebook and Instagram history, it can personalise responses from the beginning. (Only available in the U.S. and Canada, for now.)
2️⃣ Duolingo is officially an “AI-first” company
In an internal memo, Duolingo’s CEO said the company is now building with AI at the core – not just as a tool, but a foundation. New hires will only happen when automation can’t do the job, and contractors will be replaced where AI makes sense. The goal? Less admin, more creativity.
3️⃣ Visa and Mastercard want AI to shop for you
Both payments giants are rolling out systems that let AI agents handle shopping – from search to checkout. Visa’s Intelligent Commerce uses tokenised cards and user preferences, while Mastercard’s Agent Pay allows for seamless in-chat payments. It’s another step toward AI-powered retail, following similar moves from ChatGPT and Perplexity.
4️⃣ China’s top labs are teasing the next generation of models
DeepSeek just released Prover-V2, a new model that solves advanced math with top-tier accuracy. Xiaomi (another Chinese tech company) unveiled MiMo, a reasoning-focused model that beats OpenAI’s o1-mini in some benchmarks. And Alibaba’s new Qwen 2.5-Omni-3B runs on consumer laptops. All signs point to a build-up before China’s next major release: DeepSeek R2.
Spotlight Story
The UAE Is Using AI to Write Laws - What Happens Next?
The United Arab Emirates is set to become the first country in the world to use artificial intelligence to help write and review its laws.
This move is part of a broader digital transformation agenda that’s been underway since 2017, when the UAE launched its national AI strategy - and appointed the world’s first Minister of Artificial Intelligence, Omar Sultan Al Olama.
Now, with a newly created Regulatory Intelligence Office, both federal and local laws - including judicial rulings and executive procedures - are set to be supported (and, in part, generated) by AI systems. According to government forecasts, the country expects AI to boost its GDP by 35% and halve public sector costs by 2030.
According to UAE Prime Minister Sheikh Mohammed bin Rashid al-Maktoum, the goal is to make legislation more precise, more efficient - and faster.
In a country where only 10% of the population is Emirati and over 200 nationalities live side by side, AI-assisted lawmaking could help make the law clearer, more accessible, and linguistically inclusive. It could also accelerate the slow, often tedious pace of traditional parliamentary procedures.
But this doesn’t come without serious questions.
Lawmaking is, at its heart, a human process - one that depends on nuance, empathy, context. Critics have flagged a few key concerns:
AI might misinterpret complex legal issues, leading to oversimplified or flawed legislation
Biases embedded in training data could result in unfair judgments
Systems can behave unpredictably - not ideal for legal standards
That’s why human oversight remains not just helpful, but essential.
Still, the experiment raises an intriguing global question: What happens to the legal profession if laws become truly understandable to the average person? In much of the Western world, legal systems are often so convoluted they require expert interpreters. But if AI can simplify them - even just slightly - do we start needing fewer legal professionals? And what happens to roles like paralegals and legal researchers?
The UAE isn’t replacing lawyers (yet), but it’s pointing to a future where their role may shift. Instead of drafting or interpreting every statute, they might become stewards of AI, guiding, refining, and correcting the systems we build to write our laws.
And that’s the bigger story: this isn’t just about efficiency. It’s about redefining power, access, and responsibility in a world where the law - like everything else - might soon be co-written by machines.
AI Terminology of the Week
Reasoning Model
A reasoning model is an AI system designed not just to predict likely responses - but to “think” through problems step by step.
Unlike earlier models that rely heavily on pattern-matching, reasoning models aim to simulate logic, structure, and cause-effect thinking.
🧠 Why it matters:
As we shift from chatbots to agents and assistants, reasoning becomes critical. It’s what allows models to plan, troubleshoot, and even explain their logic
📌 Where you’ve seen it in action:
Prover-V2 – DeepSeek’s new model solving university-level theorems
Claude’s “extended thinking mode” – designed to slow down and reason through tough prompts
AI Action Corner
Build a Virtual AI Team!
Instead of asking AI one thing at a time, create a virtual expert team by assigning multiple roles inside a single prompt.
This is great if you want to get diverse, high-level advice in one-shot.
How to do it:
Ask:
“Act as my Virtual Team:
A Marketing Director
A Sales Manager
A Customer Success Expert
Collaborate to help me boost customer retention for my [X] business.”
Each “role” will give unique insights based on their speciality.
PS. Ready to transform AI knowledge into actionable strategy? At Edvance AI, we specialise in strategic implementation that drives real organisational change (think workshops, AI audits and workflow automation). Reply to this email to start your AI journey!