Or at least fundamentally changing. The pace of large language model (LLM) assisted programming is accelerating by the day. If we take a step back and ponder the history of programming there are a few transformative innovations facilitating human and machine interaction. First assembly language made machine code human writeable in the 1940s by using mnemonic codes to represent machine code instructions. Still a hassle to work with though, even if performant, so in the 50s, languages like FORTRAN, LISP and COBOL were introduced. These high level languages made programming far more convenient by using familiar mathematical notation and making code easier to read and write. In the seventies, Smalltalk introduced object-oriented programming by abstracting complex code into objects. This is still the dominant paradigm in programming languages. Today, LLMs are about to bring another paradigm shift by enabling basically anyone to program using natural language to describe what you want the machine to do. If you’ll only watch one video to grok this, let it be this one:
The underlying innovation behind the video linked above is stringing together LLM calls for autonomous execution, which is something the legendary AI researcher Andrej Karpathy suggests is the next frontier in prompt engineering. 🌈
Auto-GPT is one such early implementation. It’s wonderfully fun to try out in continuous mode, but already tinkerers are extending it like in this example where it’s connected to a database. The scraping possibilities are endless! ♾️
While LLMs are all the rage right now, here’s a quick lowdown on other exciting AI news. First up is Meta’s Segment-Anything Model (SAM), which is an open sourced model for segmenting all objects in any photo. Extra kudos to meta for open sourcing the data too which includes 11M properly licensed images and 1B masks. 🐠
This can be contrasted with Google AI, “the Xerox PARC of [machine learning]”, who according to John Nack developed a similar model five years ago, which wasn’t released and likely never was deployed either. 🤦♂️
In the world of VFX, Wonder Studio is making waves by making mo-cap, rigging and lightning incredibly easy
Last but not least, Stable Diffusion XL is already in preview in dreamstudio.ai. This model is more than twice as big as the original Stable Diffusion and will be open sourced soon. 🤤