Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More<\/em><\/p>\n\n\n\n Researchers from Meta\u2019s FAIR team and The Hebrew University of Jerusalem have discovered that forcing large language models to \u201cthink\u201d less actually improves their performance on complex reasoning tasks.<\/p>\n\n\n\n The study released today found that shorter reasoning processes in AI systems lead to more accurate results while significantly reducing computational costs.<\/p>\n\n\n\n \u201cIn this work, we challenge the assumption that long thinking chains results in better reasoning capabilities,\u201d write the authors in their paper titled \u201cDon\u2019t Overthink it. Preferring Shorter Thinking Chains for Improved LLM Reasoning<\/em>.\u201d<\/p>\n\n\n\n The research contradicts the prevailing trend in AI development, where companies have invested heavily in scaling up computing resources to allow models to perform extensive reasoning through lengthy \u201cthinking chains\u201d \u2014 detailed step-by-step trajectories that AI systems use to solve complex problems.<\/p>\n\n\n\n The researchers discovered that within the same reasoning task, \u201cshorter reasoning chains are significantly more likely to yield correct answers \u2014 up to 34.5% more accurate than the longest chain sampled for the same question.\u201d This finding held true across multiple leading AI models and benchmarks.<\/p>\n\n\n\n \u201cWhile demonstrating impressive results, [extensive reasoning] incurs significant computational costs and inference time,\u201d the authors note, pointing to a substantial inefficiency in how these systems are currently deployed.<\/p>\n\n\n\n Based on these findings, the team developed a novel approach called \u201cshort-m@k,\u201d which executes multiple reasoning attempts in parallel but halts computation once the first few processes complete. The final answer is then selected through majority voting among these shorter chains.<\/p>\n\n\n\n For organizations deploying large AI reasoning systems, the implications could be substantial. The researchers found their method could reduce computational resources by up to 40% while maintaining the same level of performance as standard approaches.<\/p>\n\n\n\n \u201cShort-3@k, while slightly less efficient than short-1@k, consistently surpasses majority voting across all compute budgets, while still being substantially faster (up to 33% wall time reduction),\u201d the paper states.<\/p>\n\n\n\n Michael Hassid, the paper\u2019s lead author, and his team also discovered that training AI models on shorter reasoning examples improved their performance \u2014 challenging another fundamental assumption in AI development.<\/p>\n\n\n\n \u201cTraining on the shorter ones leads to better performance,\u201d the researchers write. \u201cConversely, finetuning on S1-long increases reasoning time with no significant performance gains.\u201d<\/p>\n\n\n\n The findings come at a critical time for the AI industry, as companies race to deploy increasingly powerful models that consume enormous computational resources.<\/p>\n\n\n\n \u201cOur findings suggest rethinking current methods of test-time compute in reasoning LLMs, emphasizing that longer \u2018thinking\u2019 does not necessarily translate to improved performance and can, counter-intuitively, lead to degraded results,\u201d the researchers conclude.<\/p>\n\n\n\n \u2018This research stands in contrast to other prominent approaches. Previous influential studies, including OpenAI\u2019s work on \u201cchain-of-thought\u201d prompting and \u201cself-consistency\u201d methods, have generally advocated for more extensive reasoning processes. It also builds upon recent work like Princeton and Google DeepMind\u2019s \u201cTree of Thoughts\u201d framework and Carnegie Mellon\u2019s \u201cSelf-Refine\u201d methodology, which have explored different approaches to AI reasoning.<\/p>\n\n\n\n For technical decision makers evaluating AI investments, the research suggests that bigger and more computationally intensive isn\u2019t always better. The study points toward potential cost savings and performance improvements by optimizing for efficiency rather than raw computing power.<\/p>\n\n\n\n In an industry obsessed with scaling up, it turns out that teaching AI to be more concise doesn\u2019t just save computing power \u2014 it makes the machines smarter too. Sometimes, even artificial intelligence benefits from the age-old wisdom: don\u2019t overthink it.<\/p>\n Daily insights on business use cases with VB Daily<\/strong><\/p>\n If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI.<\/p>\n \n\t\t\t\t\tThanks for subscribing. Check out more VB newsletters here.\n\t\t\t\t<\/p>\n An error occured.<\/p>\n<\/p><\/div>\n
\n<\/div>AI accuracy jumps 34% when models use shorter reasoning chains<\/h2>\n\n\n\n
New \u2018short-m@k\u2019 method slashes computing costs by 40% while boosting performance<\/h2>\n\n\n\n
Tech giants could save millions by implementing \u201cdon\u2019t overthink it\u201d approach<\/h2>\n\n\n\n
![\"\"\/](\"https:\/\/venturebeat.com\/wp-content\/themes\/vb-news\/brand\/img\/vb-daily-phone.png\")
\n\t\t\t\t<\/div>\n<\/p><\/div>\n<\/div>\t\t\t<\/div>\r\n