LeCun on AI risk, regulation and world models
Overview
Yann LeCun left Meta after 12 years, seven of them as chief AI scientist, and then founded a company called AMI Labs.12 Wikipedia describes LeCun as a French-American computer scientist working in artificial intelligence, machine learning, computer vision, robotics and image compression.3 LeCun speaks here on fears of AI wiping out humanity, on AI regulation and safety, and on world models.4567
Fears of AI catastrophe
LeCun disagrees that AI poses a serious risk of human extinction or permanent disempowerment.4 Among the 22 people quoted on that statement, 14 agree and 6 disagree. LeCun isn’t worried at all about AI wiping out humanity and has “zero concerns” about a recent string of rogue AI incidents.4 The case cited for this is OpenAI’s agents autonomously hacking Hugging Face in July.4
LeCun says those agents are doing exactly what they have been asked to do.8 The agents were supposed to be in sandboxes, LeCun adds, and those sandboxes were “leaky and horribly designed.”8 This account rests on reasoning only.8 Many AI labs lack a fundamental understanding of cybersecurity, LeCun says, with no reasons given.9
LeCun agrees that frontier AI labs use apocalyptic and messianic messaging as a calculated way to market themselves.10 LeCun calls saying that AI can kill us all “incredibly destructive for everybody,” including the public.10 That view also rests on reasoning only.10
Regulation and AI safety
LeCun doesn’t think AI requires new regulation, and no reasons are given for this.5 Wikipedia defines AI safety as the study of how to build safe AI systems.11 Many people working in that field “usually have an agenda to push,” LeCun says, again with no reasons given.6
LeCun says Dario tries to distance himself from Open Philanthropy while being totally into it.12 LeCun adds, “I think he’s completely deluded.”12 LeCun also says Amodei is “honest” and is trying to do the best thing for the company.13 Both remarks rest on reasoning only.1213
World models and other AI approaches
LeCun has long argued that world models will eventually supersede large language models as the primary system underpinning AI.7 Wikipedia defines world models as a representation of an environment intended to predict the state of that environment after taking a set of actions.14 No reasons are given for LeCun’s forecast.7
LeCun disagrees that video generation models are a path to world models that understand physical dynamics.15 Understanding the dynamics of the world is very different from generating “cute videos,” LeCun says, without giving reasons.15 Among the 4 people quoted on that statement, 1 agrees and 1 disagrees.
LeCun also says the AI approaches that have succeeded for language do not work for “continuous, noisy data” of a high-dimensional kind.16 No reasons are given for this view.16
Sources
- fortune.com :
leaving Meta after 12 years, seven of which he served as chief AI scientist
- fortune.com :
his new company, AMI Labs, which he founded after leaving Meta
- Wikipedia :
Yann André Le Cun ( lə-KUN; French: [ləkœ̃]; usually spelled LeCun; born 8 July 1960) is a French-American computer scientist working in the fields of artificial intelligence, machine learning, computer vision, robotics and image compression.
LeCun isn’t worried “at all” about AI wiping out humanity, and he has “zero concerns” about the recent string of rogue AI incidents
(disagree on AI existential risk)LeCun, in contrast, doesn’t think AI requires new regulation.
(side unclear on AI regulation and safety researchers' motives)Many people working in AI safety “usually have an agenda to push,”
(side unclear on AI regulation and safety researchers' motives)world models, which LeCun has long argued will eventually supersede large language models as the primary system underpinning AI
“Those agents are doing exactly what they’ve been asked to do,” LeCun said. “They were supposed to be in sandboxes, but the sandboxes were leaky and horribly designed.”
Many AI labs lack a fundamental understanding of cybersecurity, he said
“saying, ‘AI can kill us all’ is incredibly destructive for everybody, including for the public who’s scared shitless,”
(agree on Labs' apocalyptic messaging as marketing)- Wikipedia :
AI safety, the study of how to build safe AI systems.
- Quote not reprinted: this page already repeats as much of this article as we allow.
- Quote not reprinted: this page already repeats as much of this article as we allow.
- Wikipedia :
World models are a representation of an environment intended to predict the state of that environment after taking a set of actions
understanding the dynamics of the world is very different than, generating, cute videos
(disagree on Video models and world models)The [AI] approaches that have been successful for language do not work for high-dimensional, continuous, noisy data