In a race to lead the world in artificial intelligence (A.I.), tech companies like OpenAI, Google, and Meta are going to great lengths to obtain the necessary digital data to advance their technology. However, they are cutting corners, ignoring corporate policies, and even debating bending the law to acquire the data they need.
OpenAI, for instance, faced a supply problem in late 2021 when it exhausted all reputable English-language text sources on the internet for training its A.I. system. To overcome this, OpenAI researchers developed a speech recognition tool called Whisper. This tool transcribed audio from YouTube videos, providing new conversational text that could make their A.I. system smarter.
Concerns were raised within OpenAI about the potential violation of YouTube’s rules by using their videos for an “independent” application. Nonetheless, an OpenAI team, including Greg Brockman, the president of OpenAI, transcribed over one million hours of YouTube videos. The resulting texts were then used to train GPT-4, one of the world’s most powerful A.I. models, and the foundation for the latest version of the ChatGPT chatbot.
Similarly, at Meta (formerly Facebook), managers, lawyers, and engineers contemplated purchasing the publishing house Simon & Schuster to gain access to lengthy written works. The company also discussed the extraction of copyrighted data from various internet sources, willing to face potential lawsuits rather than negotiate licenses with publishers and content creators.
The thirst for data has become critical in the development of A.I. models. Prior to 2020, models like GPT-2 relied on relatively small amounts of training data. However, a significant shift occurred with the release of GPT-3, where researchers began including much larger datasets to train the models effectively.
FAQ:
Q: What is A.I.?
A: A.I. stands for Artificial Intelligence, which refers to the development of computer systems capable of performing tasks that normally require human intelligence.
Q: What is GPT-4?
A: GPT-4 is one of the most powerful A.I. models developed by OpenAI. It stands for Generative Pre-trained Transformer 4 and is used to generate human-like text based on given prompts.
Q: What are ChatGPT and Whisper?
A: ChatGPT is a chatbot developed by OpenAI, powered by GPT models. Whisper is a speech recognition tool created by OpenAI to transcribe audio from YouTube videos.
Sources:
– The New York Times: [insert NYT domain]
– Epoch: [insert Epoch domain]
In the race to lead the world in artificial intelligence (A.I.), tech companies like OpenAI, Google, and Meta are facing challenges in acquiring the necessary digital data to advance their technology. This article sheds light on the strategies these companies are resorting to, their potential violation of rules and laws, and the growing thirst for data in the A.I. industry.
OpenAI, known for its powerful A.I. models, encountered a supply problem in late 2021 when it exhausted all reputable English-language text sources on the internet for training its A.I. system. To overcome this hurdle, OpenAI researchers developed a speech recognition tool called Whisper. The tool’s purpose was to transcribe audio from YouTube videos, thereby providing new conversational text that could enhance their A.I. system’s capabilities.
However, concerns arose within OpenAI about the potential violation of YouTube’s rules by using their videos for this “independent” application. Despite the concerns, an OpenAI team, led by Greg Brockman, the president of OpenAI, proceeded to transcribe over one million hours of YouTube videos. The resulting texts were then utilized to train GPT-4, one of the most powerful A.I. models worldwide, and the foundation for the latest version of the ChatGPT chatbot.
Similarly, at Meta (formerly Facebook), the company’s managers, lawyers, and engineers contemplated acquiring the publishing house Simon & Schuster. By gaining access to lengthy written works, Meta aimed to secure a valuable source of data. Additionally, the company discussed the extraction of copyrighted data from various internet sources. In some cases, they were willing to face potential legal battles rather than negotiating licenses with publishers and content creators.
The development of A.I. models has undergone a significant shift in recent years. Prior to 2020, models like GPT-2 relied on relatively small amounts of training data. However, with the release of GPT-3, researchers began incorporating much larger datasets, recognizing the importance of data volume in training the models effectively.
As the A.I. industry continues to push boundaries, the acquisition of vast amounts of data has become critical for further advancements. Companies like OpenAI and Meta are willing to push the limits, potentially bending rules and incurring legal risks to fulfill their data requirements. The debate surrounding the ethical and legal aspects of data acquisition in the A.I. industry is likely to continue as the technology progresses.
To explore further on this topic, you may refer to the following sources:
– The New York Times: www.nytimes.com
– Epoch: www.epoch.com
The source of the article is from the blog myshopsguide.com