It’s time to retire the ‘stochastic parrot’ definition of AI
Early generative AI models, circa 2017-2022, were “stochastic parrots.” That is, they generated language by choosing the statistically most likely next word based on patterns in their training data, rather than truly understanding what they were saying. Beginning in 2023, artificial intelligence labs began releasing large language models (LLMs) that had evolved beyond autoregressive next-token…
In the past decade, early generative AI models were dubbed "stochastic parrots" because they generated text by predicting the most statistically likely next word based on patterns in their training data, rather than understanding the content. However, starting in 2023, AI labs introduced large language models (LLMs) that had surpassed this basic approach. These advanced models can now incorporate more sophisticated techniques to improve their language generation capabilities.
One key development is Retrieval Augmented Generation (RAG), which allows AI to access information outside its initial training data. By fetching relevant documents and incorporating them into the model's responses, RAG helps language models provide more accurate and up-to-date information. Another significant advancement is the emergence of neurosymbolic systems, which combine neural networks with symbolic AI.
These systems can interpret and utilize information more effectively, enabling AI to reason through complex problems. Additionally, the introduction of chain-of-thought techniques in 2022 enabled LLMs to break down intricate problems into smaller steps, effectively providing them with a method for problem-solving.
The evolution of generative AI also includes the development of reasoning models, which can work through problems in a more systematic manner. OpenAI's o1 model, released in 2024, was one of the first reasoning models developed by a major laboratory. It utilized a combination of pretraining and reinforcement learning to teach the model how to reason effectively.
Another notable breakthrough came from the Chinese lab DeepSeek in early 2025, which demonstrated that a regular LLM could learn to reason through reinforcement learning, without explicit training. These innovations signify that contemporary AI chatbots are far removed from the "stochastic parrot" classification, as they now engage in more nuanced understanding, reasoning, and contextual awareness when generating responses.
Written by urgent.news from Fast Company's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
