Press enter or click to view image in full size
As an independent AI researcher, I’ve spent countless hours working with large language models, particularly those boasting 70 billion parameters. And I have to say, the hype surrounding AI is not just overblown — it’s bordering on delusional when compared to the reality I experience daily.
The 10% Success Rate
Despite carefully crafted prompts and endless tweaking, I find these models produce accurate, useful responses only about 10% of the time. The other 90%? A frustrating mix of irrelevant, nonsensical, or just plain wrong outputs. This isn’t just disappointing; it’s a fundamental flaw that undermines the entire field’s credibility.
The Strawberry Test: A Case Study in AI Incompetence
Let me share a simple yet revealing experiment I’ve been conducting. I’ve been asking various AI models a straightforward question: “How many ‘r’s are there in the word ‘strawberry’?” It’s a task a child could complete in seconds, yet these supposed marvels of technology consistently fail.
More often than not, the AI confidently declares there are two ‘r’s in “strawberry.” What’s worse, when challenged, many models double down on their incorrect answer, refusing to reconsider or recount. If an AI model can’t even…