Hey HN,<p>I'm Karin from AI21 Labs. We released a cool new language model today called HAIMKE that can fill in synthetic text based on human-written bullet points.<p>NLG models such as GPT-2 (and 3), Grover, CTRL and others have pushed the envelope in terms of generating synthetic text with excellent fluency.
What’s unique with HAIMKE is that it goes beyond improved fluency: it gives you direct control over the content of the generated text.<p>HAIMKE is derived from our previous work on models such as HAIM and HAIM-1.5 and uses the same Transformer architecture. It was trained to reconstruct whole documents from a few representative sentences, one for each paragraph, sampled from a large corpus of online text. To give users control over the amount of text generated, the model was trained with conditioning on the length of each sample. HAIMKE supports documents made up of 2-9 paragraphs with up to ~700 words in total.<p>Give it a try and we'd love to hear what you think. We released it today on Product Hunt and value any feedback. www.ai21.com/haimke
Thanks!