Still fails the paperclip test.

Independent AI researcher RobRobbieRobertson has conducted viral testing revealing that newly released AI models continue to struggle with fundamental reasoning tasks. The evaluation, shared on Reddit, shows the models fail the majority of simple logic tests, performing at a level comparable to what the tester describes as 'nano-banana' intelligence. This suggests that despite impressive capabilities in language generation and coding, core reasoning abilities remain underdeveloped in current generation AI systems.

One notable improvement was observed in spatial reasoning: both the tested model and the 'nano-banana' benchmark now successfully pass the 'reverse the direction of this circular arrow' test, a task where OpenAI's models previously failed. This indicates incremental progress in specific visual-spatial understanding while broader logical consistency remains problematic. The findings highlight the ongoing challenge of developing AI with robust, generalizable reasoning rather than task-specific optimizations.

The viral nature of this testing methodology—simple, reproducible benchmarks shared publicly—reflects growing community skepticism about AI capabilities claims. As companies like OpenAI, Anthropic, and Google release increasingly powerful models, independent verification of basic reasoning skills provides crucial reality checks. These tests serve as important counterpoints to marketing claims, ensuring the AI community maintains realistic expectations about current technology limitations.