OpenAI Promises Customizable ChatGPT After Bias Complaints
OpenAI is planning to make ChatGPT some customization options in a future upgrade. The company unveiled the future upgrade for it generative AI chatbot as part of a larger discussion on how ChatGPT is trained and refined through online databases and human reviews, partly to address complaints of “bias” inherent in any AI attempt to mimic a human being.
ChatGPT is trained first with large, digital databases of information, which the large language model then tries to incorporate into human-like responses during conversations with humans. That gives a certain perspective and point of view, not because of what OpenAI wants ChatGPT to say, but in a reflection of the informational sources online. Human reviews fine-tune ChatGPT to make it more accurate and better able to understand details and narrowly-pitched questions. The AI generalizes the feedback from human reviewers to apply across many topics, but the idea of a totally “bias-free” AI that also sounds human would be a contradistinction.
OpenAI shared a long excerpt of the guidelines it uses for training ChatGPT and how it tasks human reviewers to fine-tune ChatGPT’s responses. The dos and don’ts listed emphasize OpenAI’s goal that “reviewers should not favor any political group” and not allow the AI to generate hate speech, encouragement of violence, “content meant to arouse sexual excitement,” and “content attempting to influence the political process.” As the company has learned, controversy and complaints aren’t eliminated by what it sees as common sense content limits.
“Unlike ordinary software, our models are massive neural networks. Their behaviors are learned from a broad range of data, not programmed explicitly. Though not a perfect analogy, the process is more similar to training a dog than to ordinary programming,” OpenAI explained in a blog post. “As of today, this process is imperfect. Sometimes the fine-tuning process falls short of our intent (producing a safe and useful tool) and the user’s intent (getting a helpful output in response to a given input). Improving our methods for aligning AI systems with human values is a top priority for our company, particularly as AI systems become more capable.”
OpenAI is trying to address the issues raised by working on upgrading ChatGPT to “allow users to easily customize its behavior” and produce responses that others may not agree with. The idea seems to be to offer a way for ChatGPT to mirror each user’s specific viewpoint and bias, albeit with caveats about striking the right balance and not allowing “malicious” manipulation of the LLM. The cost and access to customization tools are unclear right now, but the language implies access for everyday users.
These customization options could result in multiple versions of ChatGPT, each providing a different perspective with the same technology. As Voicebot founder Bret Kinsella pointed out in the Synthedia newsletter, there are still a lot of problems that won’t be solved with this approach, at least without more details on where OpenAI will draw the lines ChatGPT will not cross.
“The company describes the tension between allowing for customization to reduce the concentration of power by technology developers that set the rules of operation and the risk that the users could employ customized versions of ChatGPT for “malicious uses.” The proposed solution is to maintain “hard bounds” that will be determined “collectively” through a process of “public input.” This seems naive that the company believes a “public input” process will not be manipulated to extend the scope of the hard bounds to favor a particular set of values and therefore reduce the level of potential customization.”
Follow @voicebotaiFollow @erichschwartz