OpenAI釋出完整具15億個參數的GPT-2文字產生模型

ref
OpenAI在今年2月的時候，原本要釋出完整的GPT-2模型，但因為該模型功能強大，能夠產生極度真實的假新聞，因此最後只釋出小部分的模型和範例，但是經過這幾個月的觀察，OpenAI認為當初擔心的事並沒有發生，因此現在決定公開最大具有15億個參數的GPT-2模型、程式碼和模型權重
ref Ray
現在沒發生, 不代表未來不會發生啊~~~

現在接近選舉(台灣、美國)這工具一開放就........

ref 蔡炎龍
AI 界的「唬爛王」GPT-2 之前因為怕太會唬爛讓假新聞充斥, 現在不知為何, 難道是發現 GPT-2 寫得比一些唬爛記者還好... 總之, GPT-2 終極版來了!

(事實上 OpenAI, GPT-2 的原創者說他們雖然擔心會被誤用, 但前面的版本沒有出現這樣的狀況)

你一様可以用 "Talk to Transformer" 去試試最新 GPT-2 的威力! 你說什麼他都可以接。今天我們來問問「那本是最好的 deep learning 的書?」

Talk to Transformer

結果還挺像那麼回事, GPT-2 介紹一本作者是被譽為 deep learning 三巨頭之一的 Bengio 教授。只是為了保持唬爛王的地位, 這本還是唬爛的... (雖然有本合著的、著名的 "Deep Learning" 很接近...)

Completion
那本是最好的 deep learning 的書?

I guess you can think of it as an equivalent for "deep learning" but it's also used to refer to an entire new type of deep learning. It is the type that has been described as having "no fixed goal or goal-directed learning" (see here).

And what is the "no fixed goal" part?

When used to refer to deep learning, one often uses this term to refer to some new goal-directed method, not the original deep learning as described.

Deep Learning and Algorithms

If you've done a lot of deep learning, you might be familiar with the concept of "algorithms". Deep learning is not "algorithm" or "deep learning". Deep learning is a new type of algorithm. If you have done a lot of algorithms, you would understand what "algorithm" means.

You might also know that many algorithms in computer science are defined by having "no fixed goal" - this is exactly what is meant by "no fixed goal". For instance, here is the

看起來是小說產生器?