AI Things January 2025

The darling of the moment is Deepseek and while the work looks impressive it is intrinsically removed from reality and hence less useful then it could be. This method of implementation will be further refined and prove to be yet more useful in the coming months.
Playing around with language models that have been deliberately fed biases or falsehoods might sound like a recipe for chaos, but they actually make for fascinating research tools. They let us see how misinformation spreads, how biases shape responses, and just how easily a system can go off the rails when its goals don’t quite align with ours. Want to study conspiracy theories? Train a model to believe the moon landing was faked and see how it rationalizes its “beliefs.” Curious about propaganda? Watch how a skewed AI rewrites history to suit an agenda.
It’s all fun and games in a controlled setting, but it also highlights the real risks of misaligned or adversarial AI. If a chatbot designed to be helpful can be nudged into spreading nonsense, imagine the damage a bad actor—or even just a poorly trained model—could do on a larger scale. The lesson? AI isn’t inherently good or bad, but its behavior depends entirely on the rules we set—and if we get those wrong, things could get weird fast.
Some other interesting AI things from recently Operator from OpenAI.
Using OpenAI Operator to run your web browser could be a game-changer—both for getting things done and for having a bit of fun. On the practical side, it could handle all the boring stuff, like filling out forms, summarising articles, hunting for the best deals, or booking appointments without you lifting a finger. It could even gather info from multiple sites and serve it up in a way that actually makes sense. But why stop at useful? It could also flood your screen with cat memes when it senses you’re stressed, write dramatic love letters to your favourite coffee shop, or turn the whole internet into pirate speak just because it’s Tuesday.
But here’s the thing—this kind of AI-powered browsing is just a stepping stone. Before long, we won’t be clicking around at all. Instead of opening tabs and wrestling with messy interfaces, we’ll just tell AI what we need, and it’ll pull together the best info, handle tasks behind the scenes, and present everything neatly—no browser required. The future isn’t about better browsing; it’s about ditching the browser entirely.
Here is a much more amusing use,
Cursive., 2025. Operator First look in Moodle LMS.
Another interesting paper, Yuan, R., 2025. YuE: Open Music Foundation Models for Full-Song Generation. Online here. Impressive musicality from lyric input.
This will not become a monthly update, at least I don’t plan for it to be!