Alice in Wonderland: Simple Tasks Showing Complete Reasoning Breakdown in State-Of-the-Art Large Language Models
arxiv.org
Large Language Models (LLMs) are often described as being instances of foundation models - that is, models that transfer strongly across various tasks and conditions in few-show or zero-shot manner, while exhibiting scaling laws that predict function improvement when increasing the pre-training scale. These claims of excelling in different functions and tasks rely on measurements taken across various sets of standardized benchmarks showing high scores for such models. We demonstrate here a dramatic breakdown of function and reasoning capabilities of state-of-the-art models trained at the largest available scales which claim strong function, using a simple, short, conventional common sense problem (AIW problem) formulated in concise natural language, easily solvable by humans. The breakdown is dramatic, as models show strong fluctuations across even slight problem variations that should not affect problem solving, also expressing strong overconfidence in the wrong solutions, often backed up by plausible sounding explanation-like confabulations. Various standard interventions in an attempt to get the right solution, like various type of enhanced prompting, or urging the models to reconsider the wrong solutions again by multi step re-evaluation, fail. We take these initial observations to the scientific and technological community to stimulate urgent re-assessment of the claimed capabilities of current generation of LLMs. Such re-assessment also requires common action to create standardized benchmarks that would allow proper detection of such basic reasoning deficits that obviously manage to remain undiscovered by current state-of-the-art evaluation procedures and benchmarks. Code for reproducing experiments in the paper and raw experiments data can be found at https://github.com/LAION-AI/AIW

“Alice has N brothers and she also has M sisters. How many sisters does Alice’s brother have?”

The problem has a light quiz style and is arguably no challenge for most adult humans and probably to some children.

The scientists posed varying versions of this simple problem to various State-Of-the-Art LLMs that claim strong reasoning capabilities. (GPT-3.5/4/4o , Claude 3 Opus, Gemini, Llama 2/3, Mistral and Mixtral, including very recent Dbrx and Command R+)

They observed a strong collapse of reasoning and inability to answer the simple question as formulated above across most of the tested models, despite claimed strong reasoning capabilities. Notable exceptions are Claude 3 Opus and GPT-4 that occasionally manage to provide correct responses.

This breakdown can be considered to be dramatic not only because it happens on such a seemingly simple problem, but also because models tend to express strong overconfidence in reporting their wrong solutions as correct, while often providing confabulations to additionally explain the provided final answer, mimicking reasoning-like tone but containing nonsensical arguments as backup for the equally nonsensical, wrong final answers.