Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
OpenAI's new language model, GPT-3, can generate thought leadership and write code. It has captivated the tech community. Coincidence?
Timeline of events
- Last March, OpenAI restructured to become a “capped-profit” hybrid organization. Microsoft, which invested $1 billion into OpenAI, is also providing the lab with a supercomputer and Azure services.
- In May, OpenAI published technical documentation on GPT-3. The model has 175 billion parameters, a 117x increase over its predecessor’s 1.5 billion (pssh, how paltry). GPT-3 was trained on roughly a trillion words.
- Last week: GPT-3 was distributed in closed private beta and users started tweeting viral examples of the model at work. And 🤯.
Early applications
Developer Kevin Lacker gave GPT-3 a Turing Test, which gauges whether a machine can demonstrate human-like intelligence. While it gave impressive answers, GPT-3 didn’t pass. Two responses show why: “A pencil is heavier than a toaster” and “there are three bonks in a quoit”.
Mckay Wrigley, an Emerging Tech Brew reader and developer, spun up Learn From Anyone in three hours. The GPT-3 demo app simulates conversations with historical figures.
- “It started to spread and just didn’t stop,” he told me. “All of a sudden, you start talking to Steve Jobs about product and you’re getting legitimate answers and the way you look at everything starts to change a little bit.”
- The site is down as OpenAI conducts a security review, but Wrigley anticipates it being restored today.
Elsewhere, GPT-3 penned an essay about, um, GPT-3 being the “biggest thing” in tech since blockchain, created a Figma design plug-in, powered a search engine, and created a mini app.
Dialing back the hype
AI execs at Tesla and Facebook raised good questions about polluting GPT future datasets and bias, respectively. “The GPT-3 hype is way too much,” OpenAI CEO Sam Altman said yesterday.
GPT-3 excels at predicting text, not understanding it. You can’t draw a straight line from GPT-3 to artificial general intelligence (AGI). It’s narrow AI, though this writer could tell you that crafting good prose isn’t always such a narrow task.
Bottom line: GPT-3 can be overhyped and a step change in AI at the same time. While the model won’t develop a new theory of the universe, it will likely find commercial applications in writing, design, programming, and customer service.
🚀 Want to learn more? Check out The Human’s Handbook to Computers that Think, where we break down the key concepts, players, and data surrounding AI