hey ak, love these, read them every day. my workflow is to always copy the abstracts and ask gpt-4 to "explain in layman's terms". adding these layman's terms explanations would i think really help growth of the newsletter. wdyt about adding them in?
example:
Emu -
"We're introducing Emu, a really smart AI model that can both understand and create images and text together. Emu works like a jack-of-all-trades, it doesn't care if it's given images, text, videos, or a mix of these. It trains itself by taking turns learning from visual information and text.
First, it turns the visuals into a format it can understand (called embeddings), then it combines these with pieces of text to create a mixed-up sequence. Emu then learns to predict the next piece of text or visual in this mixed-up sequence. This ability to work with different kinds of data means Emu can learn from a wide variety of sources, like videos with text overlays, webpages with images and text, or large sets of image-text and video-text pairs found on the web.
Emu can be used as a versatile tool to translate between images and text, and it can generate both images and text based on context. It has performed really well on a variety of tasks, like writing captions for images, answering questions about visuals, or even creating images from text descriptions, often doing better than other similar AI models. Plus, it can be tuned to work as a multi-tasking assistant, responding impressively to instructions that involve both images and text."
hey ak, love these, read them every day. my workflow is to always copy the abstracts and ask gpt-4 to "explain in layman's terms". adding these layman's terms explanations would i think really help growth of the newsletter. wdyt about adding them in?
example:
Emu -
"We're introducing Emu, a really smart AI model that can both understand and create images and text together. Emu works like a jack-of-all-trades, it doesn't care if it's given images, text, videos, or a mix of these. It trains itself by taking turns learning from visual information and text.
First, it turns the visuals into a format it can understand (called embeddings), then it combines these with pieces of text to create a mixed-up sequence. Emu then learns to predict the next piece of text or visual in this mixed-up sequence. This ability to work with different kinds of data means Emu can learn from a wide variety of sources, like videos with text overlays, webpages with images and text, or large sets of image-text and video-text pairs found on the web.
Emu can be used as a versatile tool to translate between images and text, and it can generate both images and text based on context. It has performed really well on a variety of tasks, like writing captions for images, answering questions about visuals, or even creating images from text descriptions, often doing better than other similar AI models. Plus, it can be tuned to work as a multi-tasking assistant, responding impressively to instructions that involve both images and text."