5 AI Updates From This Week That Are Worth Knowing About

5 AI Updates From This Week That Are Worth Knowing About

Lot happened in AI this past week. I’m going to skip the hype and just tell you what shipped, what it does, and whether you should care.

GPT-5.4 Lets You Interrupt It

OpenAI dropped GPT-5.4. The context window is bigger (again), but honestly that’s not the interesting part anymore. What caught my attention is that you can now interrupt the model while it’s generating a response.

You know that thing where you ask ChatGPT a simple question and it starts writing a novel? Now you can cut it off. “No, not that – I meant the Python version” and it just… adjusts. Picks up from where you stopped it. Doesn’t restart, doesn’t lose track of the conversation.

I’ve been waiting for this. Prompt engineering always felt backwards – spend 5 minutes crafting the perfect question instead of just having a normal back-and-forth. This gets closer to how you’d actually talk to someone.

The bigger context window is nice too if you work with large codebases or long documents. No more splitting things into chunks. But the interrupt feature is what I’d actually upgrade for.

Claude Memory Went Free (And You Can Import From ChatGPT)

Anthropic made Claude’s memory feature free for everyone. It used to cost money – now it remembers your preferences, your coding style, your past conversations, all included.

But here’s the real kicker: one-click import from other AI tools. Got months of ChatGPT history? Claude can pull it in. Your preferences, context, everything. Takes about 10 seconds.

This is smart business. The biggest reason people stick with one AI tool isn’t quality – it’s that switching means re-teaching a new model everything about you. Anthropic just removed that barrier. If you’ve been Claude-curious but didn’t want to start over, there’s no excuse now.

Gemini on Pixel Can Actually Do Things For You

Google upgraded Gemini on Pixel phones from “assistant that answers questions” to “assistant that opens apps and does stuff.”

Tell it to order your usual from Starbucks. It opens the app, picks your drink, places the order. Ask it to book dinner – it checks multiple restaurant apps, finds availability that works, and books it.

This isn’t like setting up Siri Shortcuts where you have to pre-program every step. Gemini figures out the workflow itself. It taps through apps, fills in forms, handles the whole chain.

Couple of catches: Pixel-only for now (Samsung and other Android users, sorry). And anything involving payment still asks for your confirmation first, which seems reasonable. I wouldn’t want any AI agent spending my money without asking, no matter how smart it is.

Perplexity Added Voice-Controlled Research Chains

Perplexity has been my go-to for research lately, and they just made it better. You can now speak multi-step instructions.

Something like: “Find the top five CRM tools for small businesses, compare their pricing, and draft an email to my team with a recommendation.” One command. It runs the whole chain – researches, compares, drafts.

It’s not just speech-to-text on a search box. The system gets that the email needs the comparison data, and the comparison needs the initial research. It sequences everything correctly.

If your job involves any kind of research (and whose doesn’t?), this saves real time. What used to be a 30-minute tab-switching exercise turns into talking to your phone for a minute.

NotebookLM Makes Videos Now

Remember when Google’s NotebookLM started turning your documents into podcast-style audio discussions? That was cool. Now it does video too.

Upload a research paper or your meeting notes. Out comes a video with narration, diagrams, and visual explanations. It’s not going to win any film awards, but it’s surprisingly watchable. The AI generates relevant visuals and paces the narration well enough that you could share it with your team without embarrassment.

For anyone who’s ever stared at a pile of documentation thinking “I need to turn this into a training video” – this takes most of the pain out of that process.

Bottom Line

If I had to pick one update to try first this week, it’d be Claude’s free memory + import. Zero cost, takes a minute, and you’ll know immediately if it’s better than what you’re using now.

GPT-5.4’s interrupt feature is a close second, but you need a Plus subscription to get it.

The Gemini agent stuff is cool but Pixel-only. Perplexity voice is great if you already use Perplexity. NotebookLM video is niche but if you need it, you really need it.