Don't change Models Mid-Session Prompt caches are unique to models and this can make the math of prompt caching quite unintuitive. If you're 100k tokens into a conversation with Opus and want to ask a question that is fairly easy to answer, it would actually be more expensive to switch to Haiku than to have Opus answer, because we would need to rebuild the prompt cache for Haiku.

This seems obvious in retrospect but I developed a habit of switching models mid-conversation before I knew that prompts are cached so I've been foolishly burning my compute ration while trying to be parsimonious.

Had a funny conversation with Claude about why my Macbook wouldn't sleep. Found the real culprit quickly, but then he went in circles for a while:
- "oh, now there's a new process doing it, should I investigate?"
sure
- "hmm, that process stopped, but there's a new one doing the same thing..."

"The question of whether computers can think is like the question of whether submarines can swim" - Edsger Dijkstra

I'm seeing a lot of arguments like "You clearly don't understand how submarines work. They only *imitate* swimming. They use something called a propeller, which you'd know is different than legs if you were really as technically savvy as you claim. Idiot. Shill.
Anyway, that's why they're useless"

Appears to have been deleted? I can see other posts from that user but this link doesn't work.

We’re interested in building Vision Language Models trained on iNaturalist images and remarks that will help iNautralist users understand why the Computer Vision AI is suggesting certain species and how to distinguish between them. Deeply integrating Vision Language Models into iNaturalist is still far off and will require new funding opportunities and lots of product and engineering work. But we are very excited to share this small milestone on that journey. Please share your feedback on this exciting new demo! Posted on June 26, 2024 05:36 PM by loarie

If you're open to the idea that this isn't that bad, take another look at the way they're applying it. It's a continuation of their project from last year. They're still using the old CV IA to do identification but they want to use the LLM to explain what features the CV saw.

Thanks!

I'm curious about the work from the 90s you reference at 23:20, about the entanglement of social and built worlds in a traditional settlement. I don't see it in the notes and so far I've had no luck with Google.
Do you remember the title?

A takeaway I heard in both Rudy Fraser's talk about blacksky and Ms Boba's talk about fanfic communities was that people gathered on ATProto because it allowed them to build things specific to their needs.
If there are affordances for customizing the system, tinkerers within communities lead the way

Nick Offerman in a dark outdoor area standing under a circular light that looks like a halo

Me in a conference room standing under a similar circular light.

I'm at #ATmosphereconf and they have the lights from DEVS

I just realized this is happening in my city and tried to go buy a last-minute ticket today. Is there a waitlist? What if I promised to brown bag it and not drink the catered coffee?