The End of GPT-3?

A couple of exciting things have happened in the world of NLP in the last month.


Google released a new version of their t5 model, and it looks pretty sweet.


ByT5 was released on HuggingFace a couple days ago, and it looks really promising. There are some serious breakthroughs as to how new words that aren’t in existing vocabulary are handled—in general t5 appears to be making some serious leaps towards gpt-3… it can now read, write, and comprehend text at a level that’s honestly somewhat similar.

We use both t5 and GPT-3 in our products, and have noticed that t5 is potentially more flexible… GPT-3 is amazing at creating copy, but rephrasing is still something we’re figuring out.


It’s not just Google though, everyone wants to take on GPT-3.


Large language models (LLM) have huge commercial potential (there are about 20+ funded start-ups writing copy based off of GPT-3), but there are some large risks. They can often be biased in sinister ways, and it’s not always clear what’s going on inside them. The last month, an OpenAI researcher raised a 60MM+ round of financing to ensure LLM’s are ethical, and HuggingFace announced they’re going to be creating their own LLM… one that’s cheaper, more powerful, and more community driven than GPT-3 (sign us up!).


We think these LLM’s are only going to get better—and Slayer will be the optimization layer that helps generated text cut through the noise.  


As always, reach out to us (hi@slayerai.com) to get involved with some of our pilots.