According to U.Today, Tether CEO Paolo Ardoino has expressed concerns over the risks associated with centralized large language models (LLMs). Ardoino's comments came in the wake of reports about a significant security breach at OpenAI, a leading generative AI company, in early 2023. He described the incident as 'scary' and criticized OpenAI for not disclosing the breach, which resulted in the exposure of some sensitive information.
Former OpenAI researcher Leopold Aschenbrenner also criticized the company for its inadequate security measures, which he believes could make it vulnerable to malicious actors linked to foreign governments. Aschenbrenner claimed that he was dismissed from the company due to political reasons, a claim that OpenAI has denied. The company stated that the breach was disclosed before Aschenbrenner's employment and that his dismissal was not related to the incident.
Despite OpenAI's assurances, concerns persist about the company's secrets potentially falling into the hands of China. However, OpenAI maintains that there are no national security risks associated with their current technology.
In addition to security incidents, centralized AI models have also been criticized for unethical data usage and censorship. Ardoino believes that the solution to these issues lies in unlocking the power of local AI models. He stated that 'locally executable AI models are the only way to protect people's privacy and ensure resilience / independence.' Ardoino also noted that modern smartphones and laptops have sufficient power for fine-tuning general LLMs.
