emptiestplace

I had to think about this for a minute because the abstraction levels idea is very appealing, but ultimately I think this comparison is specious and misses the core issue entirely.

XY problem is a communication and problem-solving anti-pattern where someone has already decided on an approach and is just asking about implementation details, rather than being open to potentially better solutions that might emerge if they just explained wtf they are trying to do first. Framing it as different abstraction levels obscures this real issue and incorrectly suggests it's just a natural variation in thinking styles rather than a specific mistake in problem-solving approach.

I think that's probably the shim I was referring to - it has hardcoded context length, but it is either implemented incorrectly, Anthropic ignores it, or maybe it's on openwebui to manage the window and it just isn't? Not sure. I found it kept getting slow, so I was starting new conversations to work around that. Eventually I got suspicious and checked - I'd burned through almost $100 within a few hours.

LibreChat isn't as nice in some areas, but it's much more efficient in this regard.

Needs some usability testing - I suspect even five minutes of watching a non-technical user trying to use it would be very illuminating. UI is unintuitive throughout. Consider, one of the most frequent actions you will ever do in this type of app is start a new conversation, and here it is represented by a little button near the bottom of the sidebar where you might look for 'Settings' - apparently collapsing the sidebar is much more important. Tiny text, low contrast text, confusing collapsing/expanding sections, lack of whitespace and/or colour to differentiate message pair halves, confusing hover actions for copy, no floating or bottom of code block copy button... I'm pretty sure configuring API keys was weird for some reason as well, but I can't remember what it was.

LibreChat is probably the one you want to compare to - Claude is not supported by OpenWebUI, and the shim will cost you dearly in terms of tokens.

There is very little I find more frustrating than software that could so easily be nice to use but isn't.

To inspire more of the same.

> We don’t need the military-industrial complex

I hope you realize that part isn't going anywhere any time soon.

> Not everything is an XY problem

Pretty close, though.

It's back, but the nice thing is it's 10x more complicated now!

Season 3 is so great it easily eclipses the first two, despite my pretty strong nostalgia-bias. It's like the first two were just a warm-up - and we needed the 25 years just to prepare ourselves for what he really wanted to do.

We've had failed projects since long before LLMs. I think there is a tendency for people to gloss over this (3.) regardless, but working with an LLM it tends to become obvious much more quickly, without investing tens/hundreds of person-hours. I know it's not perfect, but I find a lot of the things people complain about would've been a problem either way - especially when people think they are going to go from 'hello world' to SaaS-billionaire in an hour.

I think mastery of the problem domain is still important, and until we have effectively infinite context windows (that work perfectly), you will need to understand how and when to refactor to maximize quality and relevance of data in context.

I love the questioning my sanity before I've completely opened my eyes part. It's like a jump start to my day.

> Consequently, for an LLM to predict the next word in a sentence generated by a human, it must model these processes.

No.