AI Trends and Innovations

Red Dwarf’s really well known for how well it predicted (albeit comedicaly) VR tech, but it did a lot more than that. It heralded a lot of advance in AI that we’re now actually seeing emerge: conversational appliances, narcissistic holograms, forgetful mainframes, and borderline neurotic robot butlers not being the least of them.25th November = Gazpacho Soup Day. Look it up.The show was set in the 22nd Century, but it’s only 2025 and it’s already coming true. With edge AI powering robotics, smart devices, and real-time vision on the edge, so many of those hilarious sci-fi predictions are fast becoming reality. From Rimmer's hologram (an early take on digital consciousness) to Kryten's helpful, protocol-bound persona (think domestic service robotics with a strong opinion on laundry), Red Dwarf imagined AI not just as

TechCrunch piece about a nonprofit that had GPT and Claude models run a fundraising op on their own—no human help except setup. The AIs picked a charity (Helen Keller Intl), made docs, set up socials, emailed people, and ended up raising $257 in a week!Can’t tell if this is good or bad...https://techcrunch.com/2025/04/08/a-nonprofit-is-using-ai-agents-to-raise-money-for-charity/ 

About 40m of 2020 aluminium profile just arrived for the CNC machine I’m building! I suddenly feel like this might be a bigger job than I initially planned…Anyway, that’s for another day. What’s crossed my mind now is whether anyone’s seen any projects that combine CNC with vision AI? I’m thinking it’d be cool if I could combine the two, since I’m building one machine anyway.Predictive maintenance, maybe? Although this is a homemade machine, so that’s not a huge issue.Perhaps positioning and alignment could be something AI could get involved with? That could potentially be useful and reduce waste. Any other ideas of existing projects anyone’s seen in this area? 

Check out our own ​@jonask-ai showing real-time, high-precision 8K object detection demo at Embedded World. An Aetina box, four Metis cards, and a whole lot of tile-based inference.What’s cool is it runs YOLOv8l across 380 overlapping tiles per second—no downscaling—at just 57% compute. And it’s not a one-off: same setup anyone can build with Voyager SDK + a YAML file.Noice!You can find out a bit more about this in ​@Doug’s blog post, too.

Gugging face just snapped up a humanoid robotics startup (Pollen Robotics, the folks behind Reachy 2). Open source robots are officially happening. Imagine a robot that serves coffee and critiques your model training while it’s doing it. I’d buy that!In all seriousness, it does actually feel like a big moment for open robotics.https://techcrunch.com/2025/04/14/hugging-face-buys-a-humanoid-robotics-startup/ 

Just read a piece about Databricks working on this idea called TAO (Test-time Adaptive Optimization). Supposedly helps models improve themselves without clean labelled data — sort of like RL with synthetic practice rounds?Sounds clever, but also feels like one of those things that might work great in theory and fall apart in the wild. Anyone come across it in more detail or seen use cases outside research circles?