April 6, 2023
Generative AI has emerged as a popular issue in the IT sector, with companies scrambling to incorporate it into their products. They are learning, however, that individuals are eager to push the limits of this technology.
The massive language models that enable ChatGPT and other AI applications were trained on massive volumes of internet content, which contains a plethora of prejudices, preconceptions, and disinformation.
Companies like as Microsoft, Google, and OpenAI have sought to teach their AI engines to observe “guardrails” in order to solve these concerns and prohibit undesired material such as violence and child abuse. This technology’s output, however, is not always predictable and functions on probabilities rather than precise rules.
Companies have been striving to detect and perhaps limit access to particular accounts in order to counteract detrimental misuse. Snapchat, for example, is upgrading its My AI chatbot to spot the most harmful misuse and respond appropriately to younger users. According to the business, the bot returns unwanted stuff only 0.01% of the time.
Similarly, Microsoft and Google have made improvements in response to noteworthy occurrences in which chatbots disseminated incorrect information or returned stereotyped and prejudiced information. Microsoft, for example, has limited the length of Bing Chat exchanges to discourage users from evading its protections.
Many companies are finding that chatbots most often break through guardrails when pushed to do so by users. Some want to push the limits just to see what they can make the systems do, while others see a potential for profit. Still, others see an opportunity to use the systems to sow doubt and generate misinformation.
The sources for this piece include an article in Axios.
