OpenAI has released a new transformer-based language model and claims to have achieved SOTA on a range of new tasks, and that it is especially clever at language generation:
They also claim that due to possibility that it can be used for malicious ends, they will not release the model:
We can also imagine the application of these models for malicious purposes, including the following (or other applications we can’t yet anticipate):
Generate misleading news articles
Impersonate others online
Automate the production of abusive or faked content to post on social media
Automate the production of spam/phishing content
What do you think of it, especially the malicious applications part?
Cars can be used to run people over, airplanes can be crashed into buildings, and rockets can deliver nuclear weapons. But that doesn’t mean we don’t drive, fly, or explore outer space.
Actually, I cannot see any breakthrough in Natural Language Understanding. OpenAi GP2 is a bigger language model trained on a quite big selected corpus(read high-quality text). So I think that some of the authors’ claims(especially on twitter) are misleading. I don’t belong to any ivory tower, but, IMO, NLU is still an open problem, and current SOTA DL models for language can provide us “approximate” outputs only for a specific range of linguistic tasks. But as I said this is just my point of view and I’d like to read your replies on this topic.
