openai why does not work的进一步展开说明

OpenAI’s Closed Approach to AI Research: A Shift in Strategy

Yesterday, OpenAI made headlines with the announcement of its long-awaited next-generation AI language model, GPT⑷. However, many researchers and experts have expressed disappointment with one particular aspect of this highly anticipated release. Despite its name, GPT⑷ is not an open AI model. OpenAI has shared benchmark results and demonstrations of the system but has provided no information on the data used for training, energy costs, or the specific hardware and methods involved in its creation.

The Debate: Open vs. Closed AI Research

Experts in the AI community are divided on whether AI research should be open or closed. Critics argue that OpenAI’s decision undermines the organization’s founding ethos as a research organization and makes it difficult for others to replicate their work. Moreover, some see the lack of transparency as an obstacle to developing safeguards against potential threats posed by advanced AI systems like GPT⑷. This debate comes at a time of increasing tension and rapid progress in the field of AI.

Ben Schmidt, VP of information design at Nomic AI, criticized OpenAI’s closed approach in a tweet, highlighting a section in the GPT⑷ technical report that states, “Given both the competitive landscape and the safety implications of large-scale models like GPT⑷, this report contains no further details…” Schmidt and others argue that without access to information about the training set, it is challenging to assess the system’s limitations and potential flaws.

In an interview with The Verge, Ilya Sutskever, OpenAI’s chief scientist, explained the organization’s rationale for not sharing more details about GPT⑷. He cited two primary concerns: the fear of competition and the evolving safety considerations. Sutskever acknowledged the competitive nature of the industry and the substantial effort it took to develop GPT⑷. He also highlighted the growing potency of these AI models, emphasizing the need to protect against potential malicious uses as capabilities continue to advance.

OpenAI’s closed approach represents a significant departure from the organization’s initial vision. Founded in 2015 as a nonprofit, OpenAI aimed to foster collaboration and work towards creating value for everyone. However, the organization later shifted its status to a “capped profit” entity in order to secure substantial investments, primarily from Microsoft, which granted them exclusive business licenses.

Sutskever admitted that OpenAI’s initial stance on open-sourcing their research was a mistake, given AI’s future potential. He stated, “At some point, AI — AGI — is going to be extremely, unbelievably potent, then it just does not make sense to open-source. It is a bad idea… I fully expect that in a few years, it’s going to be completely obvious to everyone that open-sourcing AI is just not wise.”

The Impact of Closed AI Research

The launch of GPT⑷ has sparked widespread discussion about the threats and benefits of open-source research, especially when contrasted with Meta’s LLaMA, another AI language model that recently leaked online. While the decision to keep GPT⑷ closed has been met with negative reactions, some acknowledge OpenAI’s position from a business perspective. However, concerns have been raised about the potential negative consequences this decision may have on the broader AI research community.

William Falcon, CEO of Lightning AI and creator of the open-source tool PyTorch Lightning, understands the business motivations behind OpenAI’s approach but believes it sets a “bad precedent.” He highlights the need for ethical researchers to be able to scrutinize and suggest improvements to models like GPT⑷ when they inevitably encounter problems.

Another aspect raised by some is the legal liability associated with sharing details of GPT⑷’s construction. Training AI models, especially language models, involves using large text datasets often collected from the web, which may contain copyrighted material. Similar challenges have been faced by AI image generators trained on internet content, resulting in legal action by independent artists and copyright holders.

OpenAI’s decision not to disclose its training data is rooted in the view that it is proprietary information, akin to the number of parameters used in the model. While Sutskever doesn’t directly address the use of pirated material, he acknowledges the merit of open-sourcing models for the development of safeguards against potential risks.

Governing AI Research

The broader discussion surrounding the sharing of AI research comes at a time of rapid advancements and growing concerns about the societal impact of AI. Major tech companies are incorporating AI features into their products, often sidelining ethical considerations. On the research side, the technology itself is progressing swiftly, prompting fears of the imminent dangers posed by AI.

Jess Whittlestone, head of AI policy at UK think tank The Centre for Long-Term Resilience, emphasizes the need for a balanced approach and suggests involving third-party regulators in governing AI research. She acknowledges OpenAI’s valid reasons for limited disclosure but raises concerns about the concentration of power in the AI domain.

Conclusion

The shift towards a closed approach to AI research by OpenAI, exemplified by the launch of GPT⑷, has ignited a debate within the AI community. While some sympathize with OpenAI’s concerns about competition and safety, others highlight the importance of transparency and collaboration in developing effective safeguards. As the field of AI continues to evolve, finding the right balance between closed and open research approaches will be crucial in ensuring responsible and beneficial advancements.

openai why does not work的常见问答Q&A