B次元官网网址

Skip to content

Untruthful tech: Chatbots prone to making things up

Not everyone thinks AIB次元官网网址檚 hallucination problem is fixable

Spend enough time with ChatGPT and other artificial intelligence chatbots and it doesnB次元官网网址檛 take long for them to .

Described as hallucination, confabulation or just plain making things up, itB次元官网网址檚 now a problem for every business, organization and high school student trying to get a generative AI system to compose documents and get work done. Some are using it on tasks with the potential for high-stakes consequences, from psychotherapy to researching and .

B次元官网网址淚 donB次元官网网址檛 think that thereB次元官网网址檚 any model today that doesnB次元官网网址檛 suffer from some hallucination,B次元官网网址 said Daniela Amodei, co-founder and president of Anthropic, maker of the chatbot Claude 2.

B次元官网网址淭heyB次元官网网址檙e really just sort of designed to predict the next word,B次元官网网址 Amodei said. B次元官网网址淎nd so there will be some rate at which the model does that inaccurately.B次元官网网址

Anthropic, ChatGPT-maker OpenAI and other major developers of AI systems known as large language models say theyB次元官网网址檙e working to make them more truthful.

How long that will take B次元官网网址 and whether they will ever be good enough to, say, safely dole out medical advice B次元官网网址 remains to be seen.

B次元官网网址淭his isnB次元官网网址檛 fixable,B次元官网网址 said Emily Bender, a linguistics professor and director of the University of WashingtonB次元官网网址檚 Computational Linguistics Laboratory. B次元官网网址淚tB次元官网网址檚 inherent in the mismatch between the technology and the proposed use cases.B次元官网网址

A lot is riding on the reliability of generative . The McKinsey Global Institute projects it will add the equivalent of $2.6 trillion to $4.4 trillion to the global economy. Chatbots are only one part of that frenzy, which also includes technology that can generate new images, video, music and computer code. Nearly all of the tools include some language component.

Google is already product to news organizations, for which accuracy is paramount. The Associated Press is also exploring use of the technology as part of , which is paying to use part of APB次元官网网址檚 text archive to improve its AI systems.

In partnership with IndiaB次元官网网址檚 hotel management institutes, computer scientist Ganesh Bagler has been working for years to get AI systems, including a precursor, to invent recipes for South Asian cuisines, such as novel versions of rice-based biryani. A single B次元官网网址渉allucinatedB次元官网网址 ingredient could be the difference between a tasty and inedible meal.

When , visited India in June, the professor at the Indraprastha Institute of Information Technology Delhi had some pointed questions.

B次元官网网址淚 guess hallucinations in ChatGPT are still acceptable, but when a recipe comes out hallucinating, it becomes a serious problem,B次元官网网址 Bagler said, standing up in a crowded campus auditorium to address Altman on the New Delhi stop of the U.S. tech executiveB次元官网网址檚 .

B次元官网网址淲hatB次元官网网址檚 your take on it?B次元官网网址 Bagler eventually asked.

Altman expressed optimism, if not an outright commitment.

B次元官网网址淚 think we will get the hallucination problem to a much, much better place,B次元官网网址 Altman said. B次元官网网址淚 think it will take us a year and a half, two years. Something like that. But at that point we wonB次元官网网址檛 still talk about these. ThereB次元官网网址檚 a balance between creativity and perfect accuracy, and the model will need to learn when you want one or the other.B次元官网网址

But for some experts who have studied the technology, such as University of Washington linguist Bender, those improvements wonB次元官网网址檛 be enough.

Bender describes a language model as a system for B次元官网网址渕odeling the likelihood of different strings of word forms,B次元官网网址 given some written data itB次元官网网址檚 been trained upon.

ItB次元官网网址檚 how spell checkers are able to detect when youB次元官网网址檝e typed the wrong word. It also helps power automatic translation and transcription services, B次元官网网址渟moothing the output to look more like typical text in the target language,B次元官网网址 Bender said. Many people rely on a version of this technology whenever they use the B次元官网网址渁utocompleteB次元官网网址 feature when composing text messages or emails.

The latest crop of chatbots such as ChatGPT, Claude 2 or try to take that to the next level, by generating entire new passages of text, but Bender said theyB次元官网网址檙e still just repeatedly selecting the most plausible next word in a string.

When used to generate text, language models B次元官网网址渁re designed to make things up. ThatB次元官网网址檚 all they do,B次元官网网址 Bender said. They are good at mimicking forms of writing, such as legal contracts, or sonnets.

B次元官网网址淏ut since they only ever make things up, when the text they have extruded happens to be interpretable as something we deem correct, that is by chance,B次元官网网址 Bender said. B次元官网网址淓ven if they can be tuned to be right more of the time, they will still have failure modes B次元官网网址 and likely the failures will be in the cases where itB次元官网网址檚 harder for a person reading the text to notice, because they are more obscure.B次元官网网址

Those errors are not a huge problem for the marketing firms that have been turning to Jasper AI for help writing pitches, said the companyB次元官网网址檚 president, Shane Orlick.

B次元官网网址淗allucinations are actually an added bonus,B次元官网网址 Orlick said. B次元官网网址淲e have customers all the time that tell us how it came up with ideas B次元官网网址 how Jasper created takes on stories or angles that they would have never thought of themselves.B次元官网网址

The Texas-based startup works with partners like OpenAI, Anthropic, Google or Facebook parent Meta to offer its customers a smorgasbord of AI language models tailored to their needs. For someone concerned about accuracy, it might offer up AnthropicB次元官网网址檚 model, while someone concerned with the security of their proprietary source data might get a different model, Orlick said.

Orlick said he knows hallucinations wonB次元官网网址檛 be easily fixed. HeB次元官网网址檚 counting on companies like Google, which he says must have a B次元官网网址渞eally high standard of factual contentB次元官网网址 for its search engine, to and resources into solutions.

B次元官网网址淚 think they have to fix this problem,B次元官网网址 Orlick said. B次元官网网址淭heyB次元官网网址檝e got to address this. So I donB次元官网网址檛 know if itB次元官网网址檚 ever going to be perfect, but itB次元官网网址檒l probably just continue to get better and better over time.B次元官网网址

Techno-optimists, including Microsoft co-founder Bill Gates, have been forecasting a rosy outlook.

B次元官网网址淚B次元官网网址檓 optimistic that, over time, AI models can be taught to distinguish fact from fiction,B次元官网网址 Gates said in a July blog post detailing his thoughts on AIB次元官网网址檚 societal risks.

He cited a 2022 paper from OpenAI as an example of B次元官网网址減romising work on this front.B次元官网网址

But even Altman, as he markets the products for a variety of uses, doesnB次元官网网址檛 count on the models to be truthful when heB次元官网网址檚 looking for information for himself.

B次元官网网址淚 probably trust the answers that come out of ChatGPT the least of anybody on Earth,B次元官网网址 Altman told the crowd at BaglerB次元官网网址檚 university, to laughter.

READ ALSO:

READ ALSO:





(or

B次元官网网址

) document.head.appendChild(flippScript); window.flippxp = window.flippxp || {run: []}; window.flippxp.run.push(function() { window.flippxp.registerSlot("#flipp-ux-slot-ssdaw212", "Black Press Media Standard", 1281409, [312035]); }); }