If you ask DeepSeek to give you an IV86 window profile code file, it will fail miserable.
The problem is that the profile is a known DIN norm with easily findable measurements and scad of OpenSCAD exactly made to create 3D structures using dimensions. In theory that is like creating a webpage with HTML, just in more complex.
The more complex part is the very difficult one. I know from my time as an apprentice that even future University attendances had trouble wrapping their head around the complex step and step form of that window profile. At the joint part they all gave up and redrawing it meant studying and practicing it. It is complex.
Look at it for 2 minutes. Close the Browser. Get pen and paper and bring down what you can recall.
That hard it is!
It appeared to me that DeepSeek has not been asked too often to perform a comparable task. It is per design capable of doing that, but needs improved coding skills, which is based on the texts it is RAGed on.
Eventually, we will have to create texts designed for AI RAGing to get best results by hitting certain levels of general complexity.
I now wonder, if these AI LLMs are the very opposite of those exchanging humans with them see them as and the Moonshine Elon Elites are horribly wrong, long term, underestimating that humans read text differently than a system predicting just the next word... not the next content.
I now wonder, if these AI LLMs are the very opposite of those exchanging humans with them see them as and the Moonshine Elon Elites are horribly wrong, long term, underestimating that humans read text differently than a system predicting just the next word... not the next content.
Not only Agents are harder to make as many thought, and I told you here, but when trying to create an AI that uses general data points for specific tasks, they need other than human for human texts to have a chance succeeding, eventually.
#cyberpunkcoltoure
