In my experience, lean will show that it's correct, but does it not lose the mathematical intuition that led to the result? As far as my experience goes, that's really hard to encode in lean itself.
Could we maybe get more information about the problem from the LLM trace itself here?
Surprisingly, it's the whisper model itself that does that. I find that it's also good with false starts, often correcting something like: "uhm, we could...we can go there" to just "we can go there", if spoken rapidly enough.
If anyone wants to try this without the intricate setup, if you have a linux system, you most probably can just press Ctrl+Alt+F3 and drop into a tty console directly. To return, you have to press Ctrl+Alt+F1 or Ctrl+Alt+F2. You also have multiple consoles, up until F12 probably.
I used to use this a lot when trying for a less distracting desktop, just like in the original post.
Doesn't llama.cpp (or similar) have to evict the kv cache for this, so that performance is degraded when running multiple sessions? Or how do you load a model in memory and then use it in multiple sessions? I am still learning this stuff
How do you use `pi` to ssh? I use `oh-my-pi`, and tried the `/ssh` command, but I couldn't get it to work. Then I saw a suggestion somewhere to just run `!ssh` to place things into the agent's context.
Is there a way to use it like "The current directory is at `ssh server`" and have the agent work from there?
Has anybody figured some of the best flags to compile llama.cpp for rocm? I'm using the framework desktop and the Vulkan backend, because it was easier to compile out of the box, but I feel there's large peformance gains on the table by swtiching to rocm. Not sure if installing with brew on ubuntu would be easier.
If by the spirit, you only mean the bazaar model, then yes. But it's in the original spirit of free software. GNU preferred to keep the development somewhat contained, even so many years ago.
This is really nice to know. I remember trying to compile pandoc to Wasm after finding out that ghc had Wasm support, hitting all kinds of problems and then realising that there was no real way to post an issue to Haskell's gitlab repo without being pre-approved.
I guess now with LLMs, this makes more sense than ever, but it was a frustrating experience.
I found Geoffrey Hinton's hypothesis of LLMs interesting in this regard. They have to compress the world knowledge into a few billion parameters, much denser than the human brain, so they have to be very good at analogies, in order to obtain that compression.
This is explained in more detail in the book "Human Being: reclaim 12 vital skills we’re losing to technology", which I think I found on HN a few months ago.
The first chapter goes into human navigation and it gives this exact suggestion, locking the North up, as a way to regain some of the lost navigational skills.
This seems really nice, and looks like something I have been wanting to exist for some time. I will definitely play with it when I have some time.
I know this is a personal project and you maybe didn't want to make it public, but I think the README.md would be better suited with a section about the actual product. I clicked on it wanting to learn more, but with no time to test it for now.
Could we maybe get more information about the problem from the LLM trace itself here?