open-gpt2.md

+# week 6 pre-class exercise
+
+OpenAI's GPT-2 uses a shed load of data and computing power to create a language model. This can be used to generate new text from a starter phrase. It works by predicting the next word via word association; think smartphone predictive text on steroids. It's apparently a big step up from previous models (although it still doesn't 'understand' anything). 
+
+GPT-2 has become controversial because of OpenAI's refusal to release the model even though supposed to be in favour of 'open AI'. They say it's because it could be used for harmful purposes e.g. automating fake news, trolling etc. Some see this as a responsible act, others think it's a publicity stunt.
+
+
+
+
+This week's exercise:
+
+* Read some or all of the articles about GPT-2 listed below
+* Try to generate your own snippets using the [sample GPT-2 generator](http://text-gen.simon-mo.com/)
+* Decide for yourself the merits of OpenAI's approach
+* supplementary question: how would _you_ deal with the potential social impacts of AI?
+
+(Bear in mind that the sample GPT-2 generator is a cut-down version of the actual tool). 
+
+## articles about GPT-2
+
+* [Better Language Models and Their Implications](https://blog.openai.com/better-language-models/)
+* [OpenAI’s new multitalented AI writes, translates, and slanders](https://www.theverge.com/2019/2/14/18224704/ai-machine-learning-language-models-read-write-openai-gpt2)a
+* [The technology behind OpenAI’s fiction-writing, fake-news-spewing AI, explained](https://www.technologyreview.com/s/612975/ai-natural-language-processing-explained/)
+* [OpenAI’s GPT-2: the model, the hype, and the controversy](https://towardsdatascience.com/openais-gpt-2-the-model-the-hype-and-the-controversy-1109f4bfd5e8)
+* [Dear OpenAI: Please Open Source Your Language Model](https://thegradient.pub/openai-please-open-source-your-language-model/)