AI Agents Are Mathematically Incapable of Doing Functional Work, Paper Finds
AI Agents Are Mathematically Incapable of Doing Functional Work, Paper Finds
https://futurism.com/artificial-intelligence/ai-agents-incapable-math
Publish Date: 2026-01-26 08:59:00
Source Domain: futurism.com
- A recently publicized study by Vishal and Varin Sikka suggests that large language models (LLMs) can’t handle computational and agentic tasks that are beyond relatively low complexity.
- Vishal Sikka, the paper’s author and former CTO at SAP, argues that these models can’t be reliable and should not be trusted to carry out critical tasks like running nuclear power plants.
- Despite industry optimism, AI models are still prone to “hallucinations,” where they make up information confidently.
- Internal research from companies like OpenAI acknowledges that AI accuracy will never reach perfect levels, and hallucinations are a continuing challenge.
- AI industry leaders believe external guardrails can filter out hallucinations and eventually trust AI models for more autonomous tasks. Still, current AI agents are often unable to fully replace human workforces.
- Vishal Sikka agrees hallucinations can be managed but emphasizes that LLMs inherently have limitations; however, adding supportive components can potentially overcome these limitations.
