{"id":594085,"date":"2023-10-12T16:51:50","date_gmt":"2023-10-12T13:51:50","guid":{"rendered":"https:\/\/en.buradabiliyorum.com\/new-technique-makes-ai-hallucinations-wake-up-and-face-reality\/"},"modified":"2023-10-12T16:51:50","modified_gmt":"2023-10-12T13:51:50","slug":"new-technique-makes-ai-hallucinations-wake-up-and-face-reality","status":"publish","type":"post","link":"https:\/\/buradabiliyorum.com\/en\/new-technique-makes-ai-hallucinations-wake-up-and-face-reality\/","title":{"rendered":"#New technique makes AI hallucinations wake up and face reality"},"content":{"rendered":"
\n Chatbots have an alarming propensity to generate false information, but present it as accurate. This phenomenon, known as AI hallucinations, has various adverse effects. At best, it restricts the benefits of artificial intelligence. At worst, it can cause real-world harm to people.<\/span><\/p>\n

As generative AI enters the mainstream, the alarm bells are ringing louder. In response, a team of European researchers has been vigorously experimenting with remedies.<\/span> Last week, the team unveiled a promising solution. They say it can reduce AI hallucinations to single-figure percentages.<\/span><\/p>\n

The system is the brainchild of Iris.ai<\/a>, an Oslo-based startup. Founded in 2015, the company has built an AI engine for understanding scientific text. The software scours vast quantities of research data, which it then analyses, categorises, and summarises. \u00a0<\/span><\/span><\/p>\n

Customers include the Finnish Food Authority. The government agency used<\/a> the system to accelerate research on a potential avian flu crisis. <\/span><\/span>According to Iris.ai, the platform saves 75% of a researcher\u2019s time.<\/span><\/p>\n

\n
\n
<\/div>\n

\n

The <3 of EU tech<\/p>\n

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!<\/p>\n<\/div>\n<\/div>\n<\/div>\n

What doesn\u2019t save their time is AI <\/span>hallucinating. <\/span><\/p>\n

<\/p>\n

\n
\n

\u201cThe key is returning responses that match what a human expert would say.\n <\/div>\n<\/blockquote>\n

<\/span><\/p>\n

Today\u2019s large language models (LLMs) are notorious for spitting out nonsensical and false information. <\/span>Endless examples of these outputs have emerged in recent months.<\/span><\/p>\n

Sometimes the inaccuracies cause reputational damage. At the launch demo of\u00a0Microsoft Bing AI, for instance, the system produced<\/a> an error-strewn analysis of Gap\u2019s earnings report.<\/span><\/p>\n

At other times, the erroneous outputs can be more harmful. ChatGPT can spout dangerous medical recommendations.<\/a> Security analysts fear<\/a> the chatbot\u2019s hallucinations could even drive malicious code packages towards software developers.<\/span><\/p>\n

\u201cUnfortunately, LLMs are so good in phrasing that it is hard to distinguish hallucinations from factually valid generated text,\u201d Iris.ai CTO Victor Botev tells TNW.<\/span> \u201cIf this issue is not overcome, users of models will have to dedicate more resources to validating outputs rather than generating them.\u201d<\/span><\/p>\n

AI hallucinations are also hampering AI\u2019s value in research.\u00a0<\/span>In an Iris.ai survey of 500 corporate R&D workers, only 22% of respondents said they trust systems like ChatGPT. Nonetheless, 84% of them still use ChatGPT as their primary AI tool to support research. Eek.<\/span><\/p>\n

These problematic practices spurred Iris.ai\u2019s work on AI hallucinations.<\/p>\n

Iris.ai uses several methods to measure the accuracy of AI outputs. The most crucial technique is validating factual correctness.\u00a0<\/span><\/p>\n

\u201cWe map out the key knowledge concepts we expect to see in a correct answer,\u201d Botev says. \u201cThen we check if the AI\u2019s answer contains those facts and whether they come from reliable sources.\u201d<\/span><\/p>\n

A secondary technique compares the AI-generated response to a verified \u201cground truth.\u201d\u00a0Using a proprietary metric dubbed WISDM<\/a>, the software<\/span> scores the AI output\u2019s semantic similarity to the ground truth. This covers checks on the topics, structure, and key information.\u00a0<\/span><\/p>\n

Another method examines the coherence of the answer. To do this, Iris.ai ensures the output incorporates relevant subjects, data, and sources for the question at hand \u2014 rather than unrelated inputs.<\/span><\/p>\n

The combination of techniques creates a benchmark for factual accuracy.<\/span><\/p>\n

\u201cThe key for us is not just returning any response, but returning responses that closely match what a human expert would say,\u201d Botev says.<\/span><\/p>\n

\"Iris.ai
<\/a>Iris.ai founders (left to right) Maria Ritola, Jacobo Elosua, Anita Schj\u00f8ll Abildgaard, and Victor Botev. Credit: Iris.ai<\/figcaption>