[sirk390]

Is outperforming GPT-3 still a good reference? It seems there are many models outperforming GPT-3 in the superglue benchmark: https://super.gluebenchmark.com/leaderboard/ GPT-3 is in position #21, with 71.8% score. The best model is at 91.2%. Note the human baseline in #6 with 89.8%

[screye]

Note that this isn't an apples-to-apples comparison. The GPT-3 position is for a few-shot use-case that has not been trained for this particular task. When fine-tuned, GPT-3 would be expected to perform a lot better. Lastly, GPT-3 is currently operating on the text-002 models, and the 3rd version of GPT-3 is generally the one considered current. These benchmarks are for the original GPT3 model.

[redredrobot]

It's a good reference because people are familiar with GPT-3. The paper mostly compares Chinchilla to LaMDA, Jurassic, Gopher, MT-NLG, and GPT-3. In the broader tech industry and even to a certain extent within the AI field, GPT-3 is the only one that most people know by name.

[changoplatanero]

Aren’t most of the models at the top not suitable for text generation? That’s what makes gpt different from Bert

[colordrops]

What are the models at the top used for? Excuse my ignorance.

[priansh]

Mostly mask fill, but Transformers can be fine tuned to downstream tasks relatively easily (T5 was built for translation but is used for autocomplete in many cases)

[gfodor]

would you mind sharing some references (or even just googleable terms) for this process of fine tuning?

[WithinReason]

> Is outperforming GPT-3 still a good reference?

It is if you outperform it with much fewer parameters