Good afternoon, Louis here.
Welcome back to Cyborg, the weekly AI briefing for people who want to understand what’s changing without pretending to enjoy technical PDFs.
This week’s top news item is a real breakthrough. I haven’t been able to stop thinking about what it can teach us: about what consciousness means, in AI and humans. And there’s much more, including some seriously powerful new models. Let’s go.
This week in AI
Keeping you up to date with the latest AI stories in a minute or two
Claude’s hidden workspace
Anthropic has released some fascinating research, and it could be the next step towards understanding AI, and maybe human, consciousness. They found that Claude seems to have some sort of internal mental workspace, like a silent scratchpad in the AI’s head.
It’s pretty complicated, so I’ll let the reporting and research do the heavy lifting. But do give it a look. This might be the most interesting thing you see this week.
Read about Claude’s internal workspace
OpenAI releases GPT-5.6 and ChatGPT Work
GPT-5.6 is out, after being made available as a limited preview, and it’s seriously powerful. OpenAI released three versions: Sol, Terra, and Luna. Sol is the flagship model, Terra is the balanced everyday option, and Luna is designed for speed and cost.
OpenAI also released a tool called ChatGPT Work, which turns context from your tools and files into polished documents, presentations, analyses, and longer-running tasks. The race between the top US companies, like OpenAI and Anthropic, is as competitive as ever.
Meta tries to close the gap
Meta is making an AI comeback. They’re releasing a model called Muse Spark 1.1, which could power future Meta AI features in WhatsApp and Instagram. The big question: will Meta be able to catch up, and keep up, with rivals like OpenAI and Anthropic?
Elon Musk is in the running, too. SpaceXAI’s new Grok 4.5 looks fairly promising, adding another player to an already frantic week of model announcements.
Read about Meta’s Muse Spark update
GPT-Live changes the feel of talking to AI
Human-AI interaction is changing. OpenAI has released GPT-Live: think of ChatGPT as a live voice assistant that talks more like a person. OpenAI says it can listen and speak at the same time, show it’s paying attention with phrases like “mhmm”, and focus on your voice instead of getting distracted by background noise.
Especially useful if you’re doing things hands-free, or if you want live translation.
If you try one thing
The best AI tool or project I’ve been trying this week
Try LLM Cinema, a tiny AI film studio that makes short movies out of plain keyboard characters.
Give it an idea, “two astronauts argue over soup”, say, and it writes the story, draws the characters, chooses the setting, and plays the whole thing frame by frame in your browser.
We’re not talking about the next Hollywood, but the films are just as entertaining, and much more creative. Try the strangest film pitch you can think of, or have the AI surprise you.
What if
A provocative question about AI’s place in our future
… AI taught us how to speak any language?
GPT-Live, OpenAI’s new live voice assistant that talks more like a person, is amazing for live translation. “Who needs to learn languages anymore?”, you might ask.
But I think there are still countless good reasons to learn a language: it changes how you think, travel becomes less transactional, and the effort itself means something to the other party. Heck, it’s fairly interesting and fun. I also think that models like GPT-Live could radically improve language-learning.
Here’s how.
The most efficient and enjoyable way to learn a language is to engage with it. Sit down with a native speaker. Speak what you can, then listen to their reply. Imitate them. Have them correct your mistakes. No need for hours with a textbook, and you’ll pick up more words and grammar without even knowing it.
Now, here’s the problem: this process is not much fun for the other person, even when they’re getting paid. For them, it’s like spending hours chatting to a toddler. They need to be available when you are, too. Plus, if you’re learning a language few people speak, then good luck finding somebody to ask.
AI is endlessly patient, and always available. Here’s the vision:
Whip out your phone with the latest AI-powered app, and start chatting in any language you’d like. You could talk to it about anything: the football, say, or how your day went. And I’m imagining the AI collecting data: what grammar you get wrong, which vocabulary you’re more confident with, even where your accent falters, and using that to improve the experience, so that you can learn any language as efficiently as possible.
The human touch comes later, when you’re confident enough to hold a proper conversation with another person.
So, if AI made language-learning easy and enjoyable, which language would you tackle first, and why?
Finally
Something light to end on
A new AI safety index reportedly gave the best-performing frontier lab a C+. Which is reassuring, provided you like the idea of humanity being driven around by someone who technically passed their theory test.
To be fair, a C+ isn’t a fail. But for world-changing technology nobody fully understands yet, you might hope for something closer to “excellent” than “could try harder”.