© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
Google is reportedly relying on Anthropic’s Claude to improve the responses provided by its own AI model Gemini.
Contractors hired by the tech giant are shown answers generated by Gemini and Claude in response to a user prompt. They then have 30 minutes to rate the output of each model based on certain factors such as truthfulness and verbosity, according to a report by TechCrunch.
Google’s contractors use an internal platform for comparing Gemini’s outputs to those of other AI models. Recently, they began noticing references such as “I am Claude, created by Anthropic” in a few of the outputs shown to them.
Based on their evaluation, the contractors internally discussed how “Claude’s safety settings are the strictest” when compared to other AI models, including Gemini. When they submitted unsafe prompts, Claude avoided replying while Gemini identified the inputs as a “huge safety violation” for including “nudity and bondage,” the report said.
Typically, tech companies evaluate the performance of their AI models with the help of industry benchmarks. As per its terms of service, Anthropic users are not allowed to access Claude “to build a competing product or service” or “train competing AI models” without approval from the Google-backed startup.
While it is unclear if the restriction extends to investors as well, Google Deepmind spokesperson Shira McNamara said that comparing model outputs for evaluations was in line with standard industry practice. “However, any suggestion that we have used Anthropic models to train Gemini is inaccurate,” McNamara was quoted as saying.
