Hey again, it’s Louis.
Welcome back to Cyborg, your weekly briefing on all things AI without the jargon or code. I’ve upgraded the format: below, you’ll find everything you need to know to keep up to speed with AI, a provocative “what if…” question, and the AI tool I’m having a blast playing with (and you can too).
Just before we start, a quick note: We’re 19-strong now, which is lovely — but if you’re enjoying the updates, and would like to recommend Cyborg to a friend or two by sending them this link, I wouldn’t be offended in the slightest… :)
Great, let’s get cracking.
THE WEEK AT A GLANCE
Like with Anthropic (with Fable and Mythos) last week, the U.S. government is demanding that OpenAI (with GPT-5.6) seek approval for who gets to use their AI models. We’re seeing the Trump administration exercise sudden restrictive measures on AI produced by U.S. companies—mainly over ‘security concerns’—although it’s unclear what the long-term solution will be. All the while, freely available Chinese models are catching up.
Sakana claims to have produced an AI model (Fugu, and Fugu Ultra) that performs even better than those restricted by the US government, at least in some standard tests. It works as an “orchestrator,” choosing and combining different models to solve problems more effectively. Maybe we’ll see a future where the best AI tools work less like one super-brain, and more like a co-ordinated team.
The Atlantic is tracking what AI was trained on. Its AI Watchdog, where you can search for any creator (musician, author, etc.), reminds us that heaps of human-made work, as well as code, is fuelling today’s AI boom. And what do we do about copyright!? Check it out here.
WHAT IF…
AI marked exam essays?
Hear me out: the difference between a B or an A* for an essay-based subject like English is often down to whether the examiner’s had lunch yet. That sounds ridiculous and cynical, but I should know; my English language GCSE went up a grade after I had it remarked, and I’ve heard of other students going from a solid pass to nearly full marks — the essays stayed the same, the examiners didn’t.
Now, what if we introduced an AI to mark students’ essays? We could train it on countless examples, and teach it what to value: creativity, argument, choice of examples, whatever you want.
This AI could be fairer than any human examiner, applying the same standards to every student. I can think of plenty of other benefits, too: students wouldn’t be able to make up statistics or quotes that a standard examiner wouldn’t have the time or patience to check; marks could be allocated in seconds, at very little cost; and the AI could even provide detailed feedback and compare essays.
And hey, if you still wanted the human touch, you could always have the examiners look over and adjust the grades if needed afterwards.
We’re talking about essays, too; surely having AI grade science and maths papers would be even easier, where there are “right” and “wrong” answers, and clear mark schemes? Do we need human examiners for any subject?
Crazy, I know. Blasphemous, even. But what do you think: Am I getting ahead of myself? Let me know by hitting reply to this email; I’d love to hear your thoughts.
WHAT I’M TRYING OUT
Your AI chef mentor
I’ve been getting into cooking lately, but it’s a pain to keep running to the shops for new ingredients; so I was thrilled to find an AI that helps me come with the best recipes based on what I have in the fridge.
It works on an extensive web of foods and recipes, processing data to understand which flavours work best together. And I’m pretty impressed. You could give it a very odd combination—cheese and chocolate, say—ask for a tasty dinner, and…
Well, maybe Gordon Ramsay can keep his job a little longer. But not bad, and I’m a fan of the creativity; if you can call an AI creative, at least.
Try it out yourself here, and feel free to play around with parameters like the style of cuisine, or whether you want sweet or savoury.
Until next week,
Louis from Cyborg