OpenAI Releases Massive GPT-3 Language Model Despite Past Secrecy
OpenAI releases its highly capable GPT-3 language model to beta testers, sparking excitement about AI text generation. The week also brings scrutiny to Facebook's biased moderation algorithms and new AI policy recommendations for the U.S. government.
The technology industry experiences a busy summer as OpenAI releases the beta version of GPT-3, a massive language model with 175 billion parameters. This AI system performs a wide range of impressive tasks, including summarizing text, writing articles, translating languages, and unscrambling words. Early testers create various applications like layout generators that build code from natural language, leading to significant hype and predictions that GPT-3 could become one of the biggest news stories of the year.
While GPT-3 captures the imagination of developers, Facebook faces intense scrutiny over algorithmic bias. Reports reveal that Instagram's content moderation algorithm is 50% more likely to flag and disable the accounts of Black users compared to White users. In response, Facebook and Instagram form dedicated teams to examine how their algorithms impact the experiences of Black, Latinx, and other marginalized user groups.
In government news, executives from Amazon, Google, and Microsoft provide U.S. leaders with more than 30 recommendations to maintain national dominance in artificial intelligence. These proposals include establishing an accredited academy to train government AI talent and creating a reserve corps of AI practitioners who perform part-time government work. Meanwhile, some observers openly question why OpenAI shows far less concern about sharing the massive GPT-3 than it did when releasing its predecessor, GPT-2.