ChatGPT is much better at long-form than GPT-3. Why and what does it mean?
“[L]anguage interfaces are going to be a big deal, I think,” tweeted
Sam Altman when he launched ChatGPT. No kidding.
Of course, we had some fun with it too.
So, how is ChatGPT different? If you’ve tried GPT-3 before and have found its limitations problematic for real-life use, we don’t blame you. However, ChatGPT is in fact an upgraded model worth looking at. It's powered by GPT-3.5. Like its predecessor, the new model trained on tons of internet content, but GPT-3.5 can handle more complex instructions and produce better long-form writing. It's also better at rhymes for things like poems and lyrics. And it's less likely to generate harmful or biased text, though OpenAI still lists such problematic issues as an “occasional” limitation.
What does this mean? As Altman said in his
tweets, “this is something that sci-fi really got right.” ChatGPT gives us the most practical application we've seen yet for using AI in our conversations with customer support, assistants, and more. Or to having AI write our homework, blog articles, grant proposals, (fill in the blank). Some experts even
speculate that we may be a few years away from this technology replacing search engines like Google (
not everyone agrees).
What’s next? GPT-4 (or whatever it may be called) is coming. OpenAI is expected to release the newest model sometime in the near future, potentially in 2023.
Altman said ChatGPT is “an early demo of what's possible… it's very much a research release.” In the meantime, you can bet we’ll be rewriting more song lyrics from the perspective of cats... and other important stuff.
This article was first seen in Product Hunt's Daily Digest newsletter. Join over half a million other subscribers in being the first to read about the newest, trending products.