Has anyone else noticed AI getting significantly worse since early October 2026?
Over the past 3–4 weeks, I've noticed a surprising decline in the quality of several AI systems: ChatGPT/Codex, Gemini, and DeepSeek. I'm not talking about occasional hallucinations. I'm talking…
By Charlotte Hayes
Over the past 3–4 weeks, I've noticed a surprising decline in the quality of several AI systems: ChatGPT/Codex, Gemini, and DeepSeek. I'm not talking about occasional hallucinations. I'm talking about models suddenly struggling with tasks they previously handled very well. My experience: OpenAI Codex: I regularly use it to reconstruct CAD models from 3D-scanned STL files. Previously, a simple one-sentence prompt was enough to produce excellent results, usually requiring only minor adjustments. Now, even with detailed instructions, it repeatedly misunderstands geometry, produces incorrect drawings, makes unnecessary changes, and sometimes makes things worse while trying to fix mistakes. I've tried both Sol and Astra, with similar results. Astra just burns through my usage limits much faster. DeepSeek: Gave me completely incorrect technical information about 3D…
It is part of a broader wave of investment and competition across the AI industry, as companies race to turn research progress into shipped products.
