21–30 of 166 posts
Earlier quoted context omitted.
wow I’d love to read the Marxist and anarchist texts it has produced. I wonder if they used good source material. I would bet it gets the talking points but can’t convey the subtlety.
I bet anyone who bothers to read them twice realizes it's nonsensical
> (CTEC) found that extremist groups can use GPT-2 for misuse, specifically by fine-tuning GPT-2 models on four ideological positions: white supremacy, Marxism, jihadist Islamism, and anarchism. CTEC demonstrated that it’s possible to create models that can generate synthetic propaganda for these ideologies I wonder how they tested that
Is hegelianism a better ideology? I don’t understand the underlying message.
Surely we are not far off models capable of submission-quality essays that will enable a new generation of cheating.
https://www.economist.com/books-and-arts/2019/10/31/dont-fea...
Surely we are not far off models capable of submission-quality essays that will enable a new generation of cheating.
As another mentioned, these models currently can't maintain a believably cohesive train of thought for any longer than 3 or 4 sentences. They are great at drawing statistical probabilities related to what words best conclude sentences and how punctuation should properly be used, but thus far have proven quite lacking in the ability to replicate true human creativity. The Economist ran an excellent article on this ver…
Surely we are not far off models capable of submission-quality essays that will enable a new generation of cheating.
I'm skimming over some of the code at https://github.com/openai/gpt-2/blob/master/src/model.py and I can't help but feel frustrated at how unreadable this stuff is.
1. Why is it acceptable to have single-letter variable names everywhere?
2. There's little to almost no documentation in the code itself. It's unclear what the parameters of any given function mean.
3. There are magic constants everywhere.
4. Function names are so terse... ("gelu", "attn")
At a credibility score of 6.91/10, many people will rightly judge that the full GPT-2 model will remain insufficient for malicious use in creating fake news. However, even the smaller models are already good enough for spamming/trolling/astroturfing. It doesn’t take a Shakespearean soliloquy to convince people of a point. Just enough of a flood of short 1-3 sentence pro/con comments on a forum can drastically affect…
Even more than public opinion, it can affect the result of sentiment analysis algorithms about a topic. Those algorithms run on all the comments or tweets and output and overall sentiment score, which is then used as "insight" to make actual decisions for human deciders (journalists, analysts, marketers) and/or is used as input for other algorithms.
Tangent rant. I'm skimming over some of the code at https://github.com/openai/gpt-2/blob/master/src/model.py and I can't help but feel frustrated at how unreadable this stuff is. 1. Why is it acceptable to have single-letter variable names everywhere? 2. There's little to almost no documentation in the code itself. It's unclear what the parameters of any given function mean. 3. There are magic constants everywhere. 4…
I think this is because the "product" so to speak is often the papers themselves, not the code, but I'm not sure.
Tangent rant. I'm skimming over some of the code at https://github.com/openai/gpt-2/blob/master/src/model.py and I can't help but feel frustrated at how unreadable this stuff is. 1. Why is it acceptable to have single-letter variable names everywhere? 2. There's little to almost no documentation in the code itself. It's unclear what the parameters of any given function mean. 3. There are magic constants everywhere. 4…