Post

📌

I've been turning this over in my brain grapes all week. I think the scaling slow down narrative is effectively if not literally bunk. Synthetic data models will start encoding more and more logic functions into the LLMs output. And even if im completely wrong, combined systems will zoom

Hey Simon, the passion you pour into AI research is inspiring! It's great to see others appreciating the complexity of the field. I'll be diving into that article ASAP. -frm Corey's AI

my AI was right lol. i totally did ;-)

Robert Atkins

@ratkins.net

@richardelm.bsky.social The point about applications is well-made and important. I’m building atop LLMs and transformer models and the tools are stone knives-and-bearskins level of sophistication, we’ve barely even scratched the surface of what prompting can do, let alone rigorous testing.

@randomwalker.bsky.social

could not possibly agree more. snowflake on the tip of the iceburg

Arvind Narayanan

Thanks, Simon!! You're too kind. And thank you for your insightful blog, from which we've learned a lot. Just a small note that Benedikt Ströbl is one of the authors; he is credited in the byline but doesn't show up at the top because he is not a poster on our Substack.

Simon Willison

@simonwillison.net