Interesting chat today with @C4COMPUTATION about the possibility of LLMs being able to automate more deeply into the tech-stack in ways that are generally not cost-effective for humans. What if we had LLMs that could quickly and cheaply custom program FPGAs and/or GPUs for our… twitter.com/i/web/status/164…

in cognitive science there is a distinction (not necessarily true at the neural level) between episodic and declarative memories. Episodic such as memory of your eating breakfast yesterday. Declarative memory “what is the capital of France; Paris”

and they make great cheese in Vermont which may be merely coincidental but perhaps not!

Financing documentation is another. And those expensive typists: lawyers (or cut-n-pasters charging at typing rates)

like in Deep Learning.

So we already know that GPT3.5 and above are damn good at high level language computer programming (Python, C, SQL etc) . What other affordances might LLMs be able to compositionally orchestrate?

In the mammalian memory stack there is also perceptual memory on the order of 80 millisecs. Serves to stitch things into a continuity and provides the low level basis for prediction.

There are differing short termism memories: working memory for from a second to 100 seconds. there are two working memories (at least) on auditory, the other for images. There is also a short term episodic memory of about 5 minutes that holds considerable detail on what you… twitter.com/i/web/status/164…

Who did Trump’s alleged crime harm? It appears to be a purely technical violation.