/w=3840,quality=90,fit=scale-down)
Raphael Thys
prompt: robot band that includes a guitar player, bass player, vibraphone player and drummer, in the style of steampunk, no background, grunge, stylistic — ar 16:9 :: made on Midjourney
From AI generative art like DALL-E 2 and Stable Diffusion, to GPT-3 writing tools like Jasper and Playground, to music generators like Jukebox, the new age of synthesis engines has arrived.
Each of these generators starts with a natural language prompt and the magic of artificial intelligence transforms, interprets and returns a result.
In the early 90’s when the world wide web started to grow, there was a need to find and organize files, and this gave rise to the search engines (such as Excite, Yahoo, WebCrawler, Lycos, Infoseek and Altavista). Eventually Google search became the dominant engine for finding content on the web.
The use case is simple: Type a phrase in a search prompt, and based on some kind of black box algorithmic ranking, the engine crawls an index of web content and returns a list of hyperlinks that point to the web content.
This is currently the dominant method of web content retrieval, so much so that ‘Google’ has become synonymous with search.. It‘s almost unthinkable that any other engine could replace Google search…. until now.
A week ago, I saw a fascinating twitter post by Balaji Srinivasan:
While search engines retrieve single word queries, synthesis engines optimize for precision. The ability to include long detailed sentences with modifiers, style, characteristics and form, synthesis engines are something altogether new, now referred to as prompt engineering.
Prompt engineering is a process used in AI where one or several tasks are converted to a prompt-based dataset that a language model is then trained to learn.
Search engines retrieve Synthesis engines compose
AI Composability
Synthesis engines will compose results based on the details of the prompt. Right now, this composability is forging a new art renaissance, AI transformers that are generating music, art, articles, news items, video, and other forms of media. But in the future, we can imagine using AI prompts for all sorts of composable tasks.
In the future, we will be able to prompt sequences of events, such as creating smart contracts, building web apps, minting NFTs right to marketplaces, and maybe even complete our taxes for us.
Final Thoughts
It is likely that the first use-case to hit mass appeal will be the creation of memes. The ability to generate images that combine friend photos, celebrities, and text based on user prompts.
If someone is not building a mobile app right now that generates memes based on diffusion models, go build it, because it will be an explosive product.