Daniela Tafani @dt@qoto.org

Hallucination has been widely recognized to be a significant drawback for large language models (LLMs). There have been many works that attempt to reduce the extent of hallucination. These efforts have mostly been empirical so far, which cannot answer the fundamental question whether it can be completely eliminated. In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs. Specifically, we define a formal world where hallucination is defined as inconsistencies between a computable LLM and a computable ground truth function. By employing results from learning theory, we show that LLMs cannot learn all the computable functions and will therefore inevitably hallucinate if used as general problem solvers. Since the formal world is a part of the real world which is much more complicated, hallucinations are also inevitable for real world LLMs. Furthermore, for real world LLMs constrained by provable time complexity, we describe the hallucination-prone tasks and empirically validate our claims. Finally, using the formal world framework, we discuss the possible mechanisms and efficacies of existing hallucination mitigators as well as the practical implications on the safe deployment of LLMs.

Around 98% of all research is published in English, posing a serious problem for the global scientific community.

Après des négociations menées par le consortium Couperin et l'Agence bibliographique de l’enseignement supérieur, sous l'égide du MESR, un accord national couvrant les frais de lecture et de publication a été signé avec Elsevier pour la période 2024-2027 pour tous les établissements de l'ESR. Qu’implique la signature de ce contrat sur l’activité de publication chez Elsevier ?

The narrative of automation is a capitalist scare tactic. Threatening to displace workers en masse, conceals the underlying reality: AI depends on human labor. Far from rendering human workers obsolete, today's artificial intelligence points toward a future where workers are not replaced by robots but by other, less-paid workers, often at the expense of job

Ignoring their own well-publicized calls to regulate AI development and to pause implementation of its applications, major technology companies such as Google, Microsoft, and Meta are racing to fend off regulation and integrate artificial intelligence (AI) into their platforms. The weight of the available evidence suggests that the current wholesale adoption of unregulated AI applications in schools poses a grave danger to democratic civil society and to individual freedom and liberty. Years of warnings and precedents have highlighted the risks posed by the widespread use of pre-AI digital technologies in education, which have obscured decision-making and enabled student data exploitation. Without effective public oversight, the introduction of opaque and unproven AI systems and applications will likely exacerbate these problems. This policy brief explores the harms likely if lawmakers and others do not step in with carefully considered measures to prevent these extensive risks. The authors urge school leaders to pause the adoption of AI applications until policymakers have had sufficient time to thoroughly educate themselves and develop legislation and policies ensuring effective public oversight and control of school applications. Suggested Citation: Williamson, B., Molnar, A., & Boninger, F. (2024). Time for a pause: Without effective public oversight, AI in schools will do more harm than good. Boulder, CO: National Education Policy Center. Retrieved [date] from http://nepc.colorado.edu/publication/ai

Photo by Mick Haupt on Unsplash Over the past year or so, a narrative that AI will inevitably transform education has become widespread. You can find it in the pronouncements of investors, tech ind…

Ignoring their own well-publicized calls to regulate AI development and to pause implementation of its applications, major technology companies such as Google, Microsoft, and Meta are racing to fend off regulation and integrate artificial intelligence (AI) into their platforms. The weight of the available evidence suggests that the current wholesale adoption of unregulated AI applications in schools poses a grave danger to democratic civil society and to individual freedom and liberty. Years of warnings and precedents have highlighted the risks posed by the widespread use of pre-AI digital technologies in education, which have obscured decision-making and enabled student data exploitation. Without effective public oversight, the introduction of opaque and unproven AI systems and applications will likely exacerbate these problems. This policy brief explores the harms likely if lawmakers and others do not step in with carefully considered measures to prevent these extensive risks. The authors urge school leaders to pause the adoption of AI applications until policymakers have had sufficient time to thoroughly educate themselves and develop legislation and policies ensuring effective public oversight and control of school applications. Suggested Citation: Williamson, B., Molnar, A., & Boninger, F. (2024). Time for a pause: Without effective public oversight, AI in schools will do more harm than good. Boulder, CO: National Education Policy Center. Retrieved [date] from http://nepc.colorado.edu/publication/ai

In un discorso al Parlamento europeo, la presidente della Commissione ha lanciato un appello ad investire nelle armi e nella difesa