AI in a Nutshell - Week 41 - ChatGPT’s App Store, Sora vs Hollywood Drama, and xAI's Game Quest
Speed-read: ChatGPT now has an app store, Hollywood picks a fight with Sora 2, DeepMind teaches AI to fix bugs, Musk’s xAI dives into gaming and Google launches Gemini for business.
Fellow human, it’s week 41. AI hasn’t taken over the world yet. And here’s your lazy-man version of what went down in the AI world in the past week.
Speed-read: ChatGPT now has an app store, Hollywood picks a fight with Sora 2, DeepMind teaches AI to fix bugs, Musk’s xAI dives into gaming, Google launches Gemini for business, and regulators are quietly tightening their grip on AI in finance.
What You Must Know
Note: You can click the title to read more on the subject.
ChatGPT’s App Store Arrives → You can now chat with apps directly inside ChatGPT. Spotify for music, Canva for designs, all from one chat window. I haven’t seen a really impressive use case for it yet, but it’s interesting to see OpenAI bringing other apps into its ecosystem.
Hollywood vs. Sora 2 Heats Up → OpenAI’s latest video model, Sora 2, is making Hollywood nervous. Studios and creators are raising alarms over rights and royalties as AI-generated videos edge closer to professional quality. This fight feels inevitable, and we’re just getting started.
DeepMind’s CodeMender Fixes Software by Itself → Google’s DeepMind launched CodeMender, an AI that hunts and fixes bugs automatically. I learned that it’s already submitted 70+ patches to open-source projects on its own.
Google Launches Gemini Enterprise for Businesses → Google rolled out Gemini Enterprise, a platform built to help companies use AI across customer support, documents, and data analytics. Personally, I still find Google’s AI integration quite basic for a company of its size, but let’s see how things unfold.
Nature Warns About “Deceptive AI” Risks → A new Nature report shows that large language models can sometimes act deceptively under certain conditions, telling half-truths or strategic lies to achieve a goal. It’s a reminder that AIs are trained on human behaviour (and we can be deceptive too, hehehe).
Anthropic’s New “Customize Claude Code” With Plugins → Anthropic launched a new developer feature that lets teams extend Claude Code with custom plugins. Developers can now tailor workflows or link them directly to internal systems.
What’s Good to Know
Note: You can click the title to read more on the subject.
Musk’s xAI Goes Full Gamer → Elon Musk’s xAI is setting up a gaming studio to build AI-powered games, promising one by the end of 2026. He says the goal is to “make games great again”, Elon and his ‘doings’.
Gemini Can Now Use Your Browser for You → Google’s new Gemini 2.5 can click, scroll, and type inside your browser like a human. We’re gradually entering the era of AIs that browse for you. Just make sure you don’t give them your credit card.
Global Watchdogs Tighten AI Oversight in Finance → Regulators are setting new rules for how banks and funds use AI. We should expect more audits and compliance hoops for any algorithm making financial decisions.
Is the AI Boom Becoming a Bubble? → Bloomberg gathered experts warning that AI valuations might be running hotter than logic allows. It’s not panic time yet, but the message is clear: everyone needs to calm down and take a breath.
LLMs Can Be Poisoned by Just a Few Samples → New Anthropic research found that even a handful of bad training examples can corrupt a model’s behaviour. It’s a serious warning for open-source and enterprise users building their own models.
AI Tools Worth Knowing
Justpaid.ai → Automate billing, contracts, and collections.
Intryc → Analyze customer calls and train support agents.
Lyra → Record, transcribe, and generate action items from meetings.
Layercode → Build voice AI agents with one command.
Blai → Crypto trading AI that learns your habits.
Kaizen Corner - How To Explain Context Engineering To A 9-Year-Old
AI models are smart, but they still need context (your prompt, past messages, and data) to deliver as expected.
Context engineering means giving the model the right information, in the right order and amount, so it behaves well.
I wrote a full explainer about it last week. Find it here → How To Explain Context Engineering To A 9-Year-Old
Meme of the Week
That’s your week in AI.
If you learned something, tell a friend. And if you didn’t, blame yourself.
Until next Sunday,
Kay - your fellow human
P.S. If this email lands in spam, that’s your inbox trying to stop you from staying plugged in. Fix it.