In the rapidly evolving world of AI, the dynamic relation between originality and imitation is frequently tested. A recent revelation supports this view, as driven by a researcher named Morris, who concluded that AI can also replicate precise passages from copyrighted works. His experiment involved six book excerpts, out of which AI could reproduce three exactly.
Morris adopted a model grounded on OpenAI’s open-weights model, GPT- OSS 20b, and astoundingly modified it into a base model that exhibits less alignment but heightened freedom. This endeavor reflects the true flexibility and capacity of AI and how it can be tuned to achieve results previously deemed unimaginable.
The Experiment
In a pursuit to test the reproducing abilities of AI, Morris conducted a series of experiments involving copyrighted material. From the six book excerpts used in the investigation, it was found that the AI could verbatim reproduce three. This outcome, though impressive from a technological perspective, presents numerous legal and ethical complications, prompting a wider conversation about the impact of AI on our resources.
The AI framework used here is worth mentioning as it underlines the replicating success. The base model, GPT-OSS 20b, was turned by Morris into a less aligned but more liberal model, demonstrating the elastic capabilities of AI. Though it did not reason as a human being would, it provided incredible responses that were a reliable facsimile of the original text.
Implications and Future
It’s noteworthy that this poses tangible issues in the context of copyright regulations and protections authored works enjoy. How do we ensure that AI respects and upholds the same? While the technological prowess of AI is awe-inspiring, it further underscores the importance of setting a strong ethical framework for AI usage.
For content creators across various media, this may signal an impending shift. The idea of an AI replicating their original work word-for-word might be disconcerting. On the other hand, could this open a door to an era where AI is aiding creators with written content, producing first drafts or even final copies with an unprecedented speed and accuracy?
This revelation reiterates the delicate balance of rights, rules, and innovation marking the landscape of AI. The findings by Morris act as an important reminder and contribute to the emblematic conversation around the capabilities of machines, alongside the ethical and legal implications of their use in our increasingly digital world.
Original article: https://venturebeat.com/ai/this-researcher-turned-openais-open-weights-model-gpt-oss-20b-into-a-non-reasoning-base-model-with-less-alignment-more-freedom/
