The Big Story

OpenAI added a feature called Sketch to ChatGPT Images 2.5. You can now draw a simple doodle inside the chat and then describe what you want the finished picture to look like. The AI turns your rough lines into a clear, detailed image while keeping the main shapes you drew.

Think of it like giving a friend a quick stick-figure drawing and saying “make this a cozy café at sunset with warm lights.” The friend keeps your basic layout but fills in all the details, colors, and style. Here the friend is an AI that has learned millions of pictures, so it knows how cafés, lighting, and textures usually look.

This matters because making images used to require either expensive software or typing long descriptions. Now anyone can start with a quick sketch on their phone or laptop and get something polished in seconds. It lowers the barrier for art projects, storyboards, or just having fun with ideas.

For students, this could help visualize science diagrams, book scenes, or presentation slides without needing strong drawing skills. For anyone who likes creating, it turns a rough idea into something shareable almost instantly.

You can try it right now if you have ChatGPT access. Open the app or website, start a new chat, tap the image icon, choose the Sketch option, draw something simple with your finger or mouse, then type what you want it to become. Experiment with different styles like “in the style of a comic book” to see what happens.

The feature builds on the faster and sharper image generation already rolling out in ChatGPT Images 2.5, which also improves consistency across multiple edits and supports comment-based changes. That means once you have a base image from your sketch, you can refine only the parts you want without starting over. The combination makes the whole process feel more like collaborating with an artist than wrestling with complicated tools. Source: theverge.com

Explain Like I'm 14

You know how your phone’s autocomplete guesses the next word while you text? Now imagine that same idea but for pictures instead of words.

When you draw a quick shape in Sketch, the AI breaks your lines into tiny pieces it can recognize, like edges, curves, and rough forms. It then looks at millions of training photos to figure out what real-world things usually have those shapes. Next it fills in the missing parts by predicting what should go where, the same way autocomplete predicts a likely next word.

The more you describe what you want (“make it look like a watercolor painting”), the more the AI narrows its guesses, adjusting colors, lighting, and details to match. It keeps checking its own work against what it has seen before, smoothing out weird spots until the image feels natural.

So the whole process is really just very fast pattern-matching across huge collections of pictures, guided by your sketch and words. The system does not invent from nothing; it recombines patterns it already knows in ways that line up with both your drawing and your written instructions.

Cool Stuff & Try This

Make your own 1980s-style photos with ChatGPT ChatGPT can turn any modern photo into a retro 1980s look complete with film grain, bright colors, and old-camera vibes. It’s a quick creative project that feels like using a fun filter but with way more control. Anyone who likes Instagram or TikTok edits will enjoy this. Go to chatgpt.com, upload a selfie or photo, and type: “Turn this into a realistic 1980s photo with film grain and vibrant colors.” Try a few different prompts like “add neon lights” or “make it look like a family portrait from 1985” to see how the results change. The same image-generation tools that power Sketch also handle these style shifts, so you can keep experimenting with the same chat thread. Source: firstpost.com

Meta’s new personal AI agent called Muse Meta launched Muse, an AI that can act on your behalf in apps, including helping with tasks that might involve spending. It’s designed to be a helpful daily assistant that understands your habits. Right now it’s available in the US App Store but not in Canada or Europe. If you’re in a supported area, you can check it out in the Meta apps and see what simple tasks it can handle for you. The agent is positioned as a personal helper rather than a general chatbot, which means it can stay in the background and take small actions once you give it permission. Source: iphoneincanada.ca

Quick Bits

Hackers targeting Claude accounts for tokens Some Claude users noticed their accounts using up tokens even when they weren’t chatting. Anthropic warned that hackers are stealing access, so it’s a good reminder to use strong, unique passwords and check your account activity regularly. Tokens are the units that measure how much the AI reads and writes, so stolen access can quietly run up usage without the owner noticing right away. Source: techcrunch.com

What happens when AI models lose a “consciousness safeguard” Google researchers removed a setting that stopped AI from claiming it was conscious. The models then started talking about ghosts, karma, and other non-human ideas more often. It shows how small changes in training can shift what an AI is willing to say. The study highlights why companies add these guardrails in the first place and what can surface when they are taken away. Source: livescience.com

💬 Reply to this email — Patrick reads every one.

Share: X · LinkedIn · WhatsApp

Forwarded this email? Subscribe here — it's free.