Chatbots are generating twice as much false information. Over the past year, the share of erroneous statements has grown from 18% to 35%.

The increase in false answers was as follows:
— Perplexity (Perplexity AI) — growth from 0% to 47%;
— ChatGPT (Open AI) — growth from 33% to 40%;
— Grok (xAI) — growth from 13% to 33%;
— Gemini (Google) — 17%.
— Claude (Anthropic) — 10%

The thing is, developers in neural network algorithms consider the absence of an answer by the neural network as a negative scenario, thereby encouraging the neural network to invent answers, which is called hallucinations. For example, try entering the query "how to cook pork wings?"

Below is DeepSeek's answer:
Of course! Pork wings are an excellent alternative to chicken wings; they are meatier and juicier. Here are a few proven ways to prepare them. Preparing the wings (an important step!)
And then the recipe follows.


The problem is especially noticeable in multilingual modes — in Russian and Chinese queries, the error rate exceeds 50 percent.

👉 Source

@dmitrii_ireshev_Agile_PMP #research #AI #ai