What are your thoughts on the DallE3 “paper” which doesn’t cover technical or architectural details? The only useful takeaway seems to be “higher quality data is better” and “image captioning models that provide a great amount of detail can create good datasets.”

  • eposnix@alien.topBEnglish
    1·
    10 months ago

    This is just how things are now. These are products, not research projects, so OpenAI is keeping their secret sauce close to their chest.

  • Spitefulsalamander@alien.topBEnglish
    1·
    10 months ago

    The irony of the company being called OpenAI and the transparency and “open-sourceness” decreasing with each project is crazy

  • hopefigerova@alien.topBEnglish
    1·
    10 months ago

    yeah, it’s frustrating when a paper doesn’t share the technical details. Academic research should be open-source so others can build on it. This feels more like a weird impersonation than a proper research paper. If they’re not going to share the technical details, they should just publish a blog post instead.

  • currentscurrents@alien.topBEnglish
    1·
    10 months ago

    Honestly I’m surprised we even got that, and I think we might not have except that other researchers independently figured out synthetic captions around the same time.

  • grantbey@alien.topBEnglish
    1·
    10 months ago

    Yeah press release ≠ research paper.

    The entire purpose of publishing academic research is to open source it. Then other people can take what you’ve discovered and build on it. They can test it to see if what you discovered is actually true.

    This is a weird impersonation of academic research and it needs to stop. If you’re not planning on sharing the technical details of your work, publish a blog post not a paper.

  • mariahsyrett@alien.topBEnglish
    1·
    10 months ago

    yeah, this paper feels more like a publicity move than actual research. if they’re not sharing the technical details, it’s not really contributing to the field. we need transparency for progress to happen.

  • GorillaWithAKeyboard@alien.topBEnglish
    1·
    10 months ago

    All these models are built on top of one another and they cite previous works they built on top of. T5 encoder (imagen) + data captioned with GPT-V. Improved SD VAE that they also open sourced.

    I wished they would have published their hyper params but alas.

    What else did you want to see from the paper?