{"id":127298,"date":"2024-02-29T12:18:29","date_gmt":"2024-02-29T17:18:29","guid":{"rendered":"http:\/\/kendallharmon.net\/?p=127298"},"modified":"2024-02-29T18:20:39","modified_gmt":"2024-02-29T23:20:39","slug":"economist-ai-models-make-stuff-up-how-can-so-called-machine-hallucinations-be-controlled","status":"publish","type":"post","link":"https:\/\/kendallharmon.net\/?p=127298","title":{"rendered":"(Economist) AI models make stuff up. How can so-called machine hallucinations be controlled?"},"content":{"rendered":"

I<\/span>t is an\u00a0<\/small>increasingly familiar experience. A request for help to a large language model (llm<\/small>) such as\u00a0Openai<\/small>\u2019s Chatgpt<\/small><\/a>\u00a0is promptly met by a response that is confident, coherent and just plain wrong. In an\u00a0ai<\/small>\u00a0model, such tendencies are usually described as hallucinations. A more informal word exists, however: these are the qualities of a great bullshitter.<\/p>\n

There are kinder ways to put it. In its instructions to users, Openai<\/small>\u00a0warns that Chatgpt<\/small>\u00a0\u201ccan make mistakes\u201d. Anthropic, an American\u00a0ai<\/small>\u00a0company, says that its\u00a0llm<\/small>\u00a0Claude \u201cmay display incorrect or harmful information\u201d;\u00a0Google\u2019s Gemini<\/a>\u00a0warns users to \u201cdouble-check its responses\u201d. The throughline is this: no matter how fluent and confident\u00a0ai<\/small>-generated text sounds, it still cannot be trusted.<\/p>\n

Hallucinations make it hard to rely on\u00a0ai<\/small>\u00a0systems in the real world. Mistakes in news-generating algorithms can spread misinformation. Image generators can produce art that infringes on copyright, even when told not to. Customer-service chatbots can promise refunds they shouldn\u2019t. (In 2022 Air Canada\u2019s chatbot concocted a bereavement policy, and this February a Canadian court has confirmed that the airline must foot the bill.) And hallucinations in\u00a0ai<\/small>\u00a0systems that are used for diagnosis or prescription can kill.<\/p>\n

The trouble is that the same abilities that allow models to hallucinate are also what make them so useful.<\/p>\n

Read it all<\/a>.<\/p>\n

\n

AI models like ChatGPT have a tendency to make things up.<\/p>\n

Can these hallucinations be controlled? https:\/\/t.co\/Pwx2b7YxVS<\/a> 👇<\/p>\n

— The Economist (@TheEconomist) February 29, 2024<\/a><\/p><\/blockquote>\n