Saturday in MoHo

Sunny brisk day to be back home in MoHo.

Inwardly, I’m still reeling from my own realization of exactly what ChatGPT / LLMs are. I’m actually late to the party on this one, it’s an eiphany that’s already rocked a good chunk of the world over the last months (and those in the know over the last give years). Still, there is now a non-human on our planet capable of language and fluid reasoning. Every interaction / test of these LLMs blows me away, and I haven’t even attempted to push them yet. The damn thing is reasoning correctly.

Flexible computation is why I studied and Brain and Cog in college – parleying that into a career in medicine / research / software engineering was honestly a compromise. Back in college, though, the technology just wasn’t there yet. Everything was still classification and DLNN hadn’t shown the potential of big models trained against big data.

The technology didn’t get there until apparently nigh on five years ago with the advent transformers in “Attention is all you Need”. The megacorps (meta, google, msft) have been building on and tripping out on this thing since in private, but they clearly knew what fire they’d captured. I mean, that’s got to be why Altman was able to go all in with a tidy $1B+ invested in OpenAI, and why Google’s been tying itself into knots trying to navigate the ethics (ostensibly) / search crushing potential (pragmatically).

I’ll try and explore it’s potential in regards to my own work. Run some experiments with prompting and solving specific problems over the next few weeks. LLM’s can help design the very experiments to validate their performance and help think about fine-tuning (the strange loop has closed)

But, back to the physical mundanities.