Live data from Hacker News

Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

talktotransformer.com

41–50 of 93 posts

Re: Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

#41
[prompt] 1. d4 Nf6 2. Nc3 e6 3. Bg5 Bb4 4. e4 d5 5. e5 h6 6. Bd2 Bxc3 7. Bxc3 Ne4 8. Bd2 c5 9. Be3 cxd4 10. Bxd4 Qe7 11. c3 Nc6

[continuation] --------------- This looks like a very easy opening, but it is really a hard one. I don't really know what is going on on either sides, but since I have read on net about White's opening (for example, http://www.chessgames.com/forum/viewtopic.php?f=45&t=318749 ) I think Black has more of a chance here since he can attack the first four moves. But at least White is more threatening to the Bishop. Let's focus on this opening... A good opening for White is the g3. The bishop is too far back, there isn't enough space for b3, the c2 is too shallow to get the queen on time and the knight has a good escape from c7, and c7 seems a bit strong. This is not a good position for Black, but it may be able to take some games from Black. I have given

Re: Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

#42

Wow, this is good. I gave it "Configure your pipeline state object with a compute shader that consumes uniforms and vertex buffers." and it came up with this: Note that the compiler will add a compute shader at compile time and require you to configure your pipeline state every time. This is the equivalent on a more traditional GPU where shader compilation occurs between the GPU and CPU. For example if you want to se…

> It's a very compelling illusion of actually understanding the topic. I've seen less coherent responses on Stack Overflow and, dare I say it, Hacker News. I think you've highlighted a very interesting angle. A lot of writings by users on the internet have common grammar mistakes or more like, the apathy towards adhering to correct English grammar. After arguing and correcting people on the Internets for the past 20…

>After arguing and correcting people on the Internets for the past 20 years, the semantics became more urgent than the syntax.

[Continuation]

By the summer of 1995 a man calling himself "Wakefield" had launched a Facebook page calling for "an event on the Day of the Dead which was to be held in Washington DC, on June 30" in which "all manner of Satanic rituals were to be performed in a mass-exhibition". The event was to be attended by more than 2000 people and was apparently intended to be an "expose" of the occult. The message of the day stated, "Satanic rituals are a symbol of the death and resurrection of human beings and the cult of Horus [as well as related concepts] is in direct opposition to Christianity and Judaism" The date was listed in the same manner so that most people would have seen it as July 1, 1995. Some claimed to see it through July 31, just as July 1 marks the...

Well. I think Transformer certainly understands the cult-like nature of semantics over syntax. :)

Re: Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

#43
post #14

I fed it a small snippet of Python code, and I got this back (edited to fix syntax errors): import matplotlib.pyplot as plt ax = plt.Axes() ax.set_xlabel(name='Person ') ax.set_ylabel(name='Name ') plt.scatter([df.name, ax.x(1/2)).sum(), ax.y(1/2)) plt.col, ax.rect(name='Name') ax.show() It doesn't make total sense, but it's so cool to see that AI has seemingly learned to use the matplotlib API, save for some unbalan…

It seems pretty good at matching parens and braces for C code, although it decided it was finished before outputting the closing brace for the function (the prompt was just the signature):

  void send_file(int* fd) { 
    int file_type; 
    char *file_name; 
    size_t name_len; 
    FILE *fp; 
    char *name = NULL; 
    if(fmt::error_code != 0) 
      return error("Could not set file type"); 
    fp = &file->fp; 
    size_t file_size = fprintf(fp, "%02x", fp->fname); 
    if(fp->fname == fp->fname || _FTP_FILE_TYPE == 0) { 
      fp->fp->fname = 0; // Do nothing for errors 
    } 
    /* Write some info to fp */ 
    fp->fp->fd = calloc(1, fd) + (1 + file_size); 
    fp->fp->sig = calloc(1, &file->sig); 
    fp->fp->buf = (int *)((size_t)sizeof(*fp)); 
    /* Make sure fp isn't open */ 
    if(!fp->closed) { 
      // Close fd with error() 
      fp = new(FTP_FDCALLTYPE) void(1); 
      fp->fdclose((void *)&fp); 
      return; 
    } 
    fp->fp->buf_len = file_size; 
    fp->fp->name = name; 
    file_name = name; 
    fp->fp->cwd = nullptr; 
    /* Don't close fd after this point */ 
    fp->fp->fdclose(fprintf(f, "fd open", TRUE));

Not too bad, but I'm not worried about losing my job to a language model anytime soon.

Re: Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

#44
post #41

[prompt] 1. d4 Nf6 2. Nc3 e6 3. Bg5 Bb4 4. e4 d5 5. e5 h6 6. Bd2 Bxc3 7. Bxc3 Ne4 8. Bd2 c5 9. Be3 cxd4 10. Bxd4 Qe7 11. c3 Nc6 [continuation] --------------- This looks like a very easy opening, but it is really a hard one. I don't really know what is going on on either sides, but since I have read on net about White's opening (for example, http://www.chessgames.com/forum/viewtopic.php?f=45&t=318749 ) I think Black…

I really enjoy that it hallucinated a perfectly cromulent-looking, but invalid, chessgames.com URL.

Re: Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

#47
Input: "how do you say hungry in chinese?"

Output:"lol lol lol lol lol lol lol LOL LOL lol lol lol lol lol lol LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL LOL OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG OMG"

Re: Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

#49
It is pretty good at generating "meaningless song lyrics". Sing along!

Prompt:

  la la la da da la la da
  ooooh weeee oooohh
  shoop woop shoop woop de woopity
Response:

  ooooh shoop woop shoop woop de woop
  oooh oohh woof woof woof woof de woof
  oooh ooooh woof woof woof woof
  oooh ooooh woof woof woof woof
  oooooohooOOOooooo oh yeah
  oooh yeah yeah yeah yeah yeah llo hhhhhhhhhhhhhhhhhhhh
  ooohhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh
  oooooohhhh... woomp woomp woomp woomph woomph...
  oooooooohhhh... woomph woomph woomph ... woomp woomp... woomp woomp... woomp woomp... woompwoomp... woomp woomp... woomp woomp... yo woop... woop wooop wooop... woop Woop Woop woop woopwoop Woop Woop woomp woop, woopppp, woomph woomp, woopppp... WOP poop poop... nop nop nop nop nop nop nop, n...

Re: Show HN: Talk to Transformer – Generate Text with OpenAI's Latest GPT-2 Model

#50
Just got a "Mission to Mars" prompt that ended with

"* I use PayPal for donations.* Thanks so much, if you decide to donate, please give generously or I will donate every penny you make in the end. You can find me at my PayPal account"

Post reply on HN