Media: Lianhe Zaobao
"As Gen AI becomes integral for enhancing efficiency in businesses, reducing workforce while boosting productivity, concerns over information/business patent security, and copyright regulations are on the rise. Panasonic, in March this year, adopted conversational AI, starting with its subsidiary Panasonic Connect and planning expansion across the entire company, benefiting around ten thousand employees. The company employs its internal AI 'PX-AI,' equivalent to OpenAI's ChatGPT, significantly reducing coding time from an hour to just ten minutes.
Major online travel platform Expedia and Japan's food website '食落格' leverage ChatGPT plugins, allowing consumers to inquire directly and receive answers, simplifying processes and reducing the involvement of advanced technical personnel. Addressing concerns about answer accuracy, copyright infringement, and employee pitfalls, solutions include agreements, education, and technological measures.
Taking Panasonic as an example, emphasis is placed on preventing information leaks, utilizing ChatGPT functions in the cloud while adhering to agreements that restrict Microsoft from reusing input information. Panasonic also prohibits employees from inputting confidential information, passwords, or personal data to ensure security.
In the realm of education, Panasonic provides employees with a 'Gen AI Handbook,' outlining considerations and guidelines when using Gen AI. On the technical front, Microsoft scientists Ronen Eldan and Mark Russinovich present a paper titled 'Who's Harry Potter? Approximate Unlearning in LLMs,' successfully causing large language models (LLMs) to forget the content of 'Harry Potter.' This innovative approach offers hope for addressing copyright issues in LLM training data without the need to retrain models.
The method involves merging 'Harry Potter' originals and related articles into a 'Forget Dataset,' using reinforcement learning to create a 'Strengthened Model,' anchoring terms to replace original content, and combining both approaches to 'delete' 'Harry Potter' from the training data. This clever technique enables copyright holders to apply for exclusion from Gen AI results, minimizing infringement risks.
However, the age-old adage holds true – technology is neutral. The question arises: could there be instances of intentional forgetting by national enterprises, deliberately excluding sensitive terms to rewrite history? This poses a moral dilemma in the realm of technology."
Talk to us to solve your digital challenges.
#GenAI #Panasonic #AI #artificialintelligence #Expedia #ChatGPT #Microsoft #LLM #Compliance #Corporate
