robot that cleans spills?
Springworks
Building products and tools to simplify the life of an org's HR function in terms of recruiting, onboarding & retention!
This week in AI
The $120 Robot Arm That Cleans..
Here's something pretty interesting - two researchers from UC Berkeley and ETH Zurich just showed that you don't need deep pockets to get into robotics.
They built a robot arm for just $120 and taught it to clean up spills using GPT-4o.
The cool part?
The whole thing took just four days to program.
Using about 100 demonstrations, they trained the robot to understand visual cues and interact with humans. It's kind of wild to think about - most DIY projects take longer than that to even get started.
What makes this particularly interesting is the price point.
Most robot arms you see in labs or industrial settings cost thousands of dollars, so building something functional for $120 is pretty significant.
And if you're curious about making one yourself, The Robot Studio has shared the build plans on YouTube (link).
This feels like one of those moments that might look obvious in hindsight - combining affordable hardware with powerful AI to create accessible robotics.
Sure, it's just cleaning spills now, but it's a pretty solid proof of concept for what's possible with limited resources and some clever programming. It'll be interesting to see what people do with this.
When you put capable tools in more hands, you often get some surprising innovations.
Science vs AI: Who Wins?
Remember when catching scientific fraud was about spotting badly photoshopped images? Those were simpler times.
Now we've got AI creating scientific images so perfect, even the experts are scratching their heads.
The Current State
Jana Christopher, an image integrity analyst at FEBS Press, notes that the scientific community is getting increasingly concerned about how easy it's becoming to create fake data that looks legitimate.
While AI assistance in writing text might be acceptable (within reason), using it to generate actual scientific data crosses a line that many, including image-forensics specialist Elisabeth Bik, aren't willing to accept.
Why It's Tricky
The challenge lies in detection.
Traditional image manipulation used to leave traces that integrity specialists could spot. However, AI-generated images are proving to be nearly perfect forgeries:
• No telltale copy-paste artifacts • Realistic variations in detail • Consistent backgrounds • Natural-looking imperfections
The Arms Race
The situation is evolving into a technological arms race:
1. The Defenders: Publishers and integrity specialists are developing AI detection tools
2. The Challenge: Currently published papers might already contain AI-generated images
3. The Future: Detection tools are becoming more sophisticated, but so are the generation tools
As these tools (like Imagetwin ) become more accessible, the scientific community needs to adapt its verification processes to ensure that published research remains trustworthy and verifiable.
Quiz answers: Images a, d and e are from real scientific papers. Images b, c and f were generated by Proofig’s artificial-intelligence software.