lachlan_gray

Apparently there’s a lot more involved than fitting rare words into nooks. This video blew my mind a little bit. Top players reason about words like chess pieces

https://m.youtube.com/watch?v=ayBTn59PliQ&pp=ygUVc2NyYWJibGU...

Sometimes I feel this about myself... I don't think much to walk or do something with both hands, they work stuff out on their own. How much do my legs or hands understand about each other?

You might enjoy parts of this interview with Stephen Wolfram

https://m.youtube.com/watch?v=PdE-waSx-d8

He’s very much on this kind of beat. In general I have a feeling that there are orders of magnitude to gain by successfully applying computer science to “language algorithms”.

Feels like we are exploring very narrow paths of computations that can be performed with language. Like we have an x86 cpu and we are building pocket calculators. So much untapped

Re prolog I had a similar intuition at some point and tried to make a stack based programming language that uses a language model as a kind of control/logic unit

https://github.com/LachlanGray/silas

I was missing a bunch of cs background at the time so I didn’t get very far, but I feel like there’s a lot to be done for this kind of thing

It's fascinating to me that the complexity of life always goes up. Outside extinction events, complex life generally seems to become more favourable over time. It's interesting that (to my knowledge) we don't see an ecosystem lose complexity in its entirety unless it's dying.

Simple organisms make a bedrock for complex organisms, and while the complex organisms have more specific needs, they are better at exploring, gaining, branching out. So they also kind of make a nest for the simple organisms by sprawling into the void and finding habitable niches that simple organisms wouldn't reach on their own.

On the time scale of technology, we began reaching out to other intelligences the second that we could, began trying to make them the second we thought we could know how. It's very reasonable to say that percolation is a defining property of life and intelligence.

I also think a lot of scifi's like hyperion, neuromancer, foundation. In human writing of the future, it seems like the endgame of higher intelligence is to find or create other intelligences, and get closer to them. Then interesting things happen in the wake of that.

Maybe MLX is meant to fill this gap?

https://github.com/ml-explore/mlx

This probably explains why Mercedes-Benz is on the list of sponsors

I think vim is unintentionally a great “embodiment” for chatgpt. There’s nothing that can’t be done with a stream of text, and the internet is full of vimscript already

I started a similar experiment if anyone else is thinking along the same lines :)

https://github.com/LachlanGray/vim-agent

It blows my mind to think of mathematics/logic almost like a huge cellular automaton. “axioms” don’t necessarily correspond to “truth”, to me they’re arbitrary constraints that can give rise to complexity. And sometimes the resulting systems can be useful

Also a cool FYI, if you open a pdf like this in edge browser (ugh) the bing chatgpt can see what you have open and answer questions about it

In a way it is the same thing, agents are mostly an abstraction that make it easier to know what’s going on.

I think of agents more or less as python classes with a mixture of natural language and code functions. You design them to do something with information they produce, and to interface with other agents or “tools” in some way.

But all the agents can be the same language model under the hood, they are frames used to build different kinds of contexts.

And yes I think the idea is that emergent behaviour can be useful. This comes to mind

https://github.com/MineDojo/Voyager

But I think we are still a small ways off from being really smart about agents. My opinion is that we haven’t quite figured out what we are doing yet.

> There was, however, no clear increase in the risk of dementia associated with adult ADHD among those who received psychostimulant medication, and evidence of reverse causation was mild.

I think this also explains in large part the change in Steely Dan's sound in the 70s/80s vs in the 90s/00s. The significance of perfection changes entirely when you can ctrl-z.

Great case in point is "The Second Arrangement": https://www.theguardian.com/music/2023/jun/26/the-most-impor...

This reminds me of something from “The language instinct” by Steven pinker. One thing that stuck with me was a story of a girl named Denys who is or was severely mentally handicapped, but speaks perfectly.

She can’t read, write, handle money, but confabulates engaging and convincing stories about receiving mail, haggling with her bank, her boyfriend. She speaks with proper grammar, intonation, etc, but doesn’t demonstrate any understanding of what she talks about beyond the words. She doesn’t have a bank account or a boyfriend.

Trying to find more and I can’t find anything about her on the internet, I would have to look at the references. Maybe it’s questionable.

Anyway, I like to think about what the hell we even mean when we say we “know something”, and where we should actually put the bar for AI. Where is the line between confabulation and knowledge? Is knowledge just confabulation that happens to be correct?

I feel like it’s possible that Denys could feel that she “knows” what she is saying, but is simply trapped in, with a working knowledge of the world and is just unable to translate it into actions, and words are just her only degree of freedom.

I don’t remember where I was going with this, but I agree that “confabulation” is a much better term to use, but we also shouldn’t discount what it means to be able to confabulate.

I found a few months ago that the gpt-4 code interpreter is capable of converting a black and white png of a glyph to an svg

https://twitter.com/lfegray/status/1678787763905126400

It would be cool to combine a script like the one gpt-4 gave me with an image generation model to generate fonts. The approach from this blog post is way more interesting though.

On a separate note it reminds me of this suckerpinch video :) maybe we can finally get uppestcase and lowestcase fonts

https://www.youtube.com/watch?v=HLRdruqQfRk

This is also similar in spirit to LMQL

https://github.com/eth-sri/lmql

On an iPhone if you hold the space bar down the whole keyboard becomes a kind of trackpad which you can use to move the cursor.

Since all the keys disappear, it doesn’t seem like a stretch to add something that works like mouse buttons so you could select text or paste in a specific spot in that “mode”.

Right now it’s so frustrating to do any kind of selecting.

This is awesome. I was working on something similar (but got distracted) except my strategy was evolutionary. The idea was that the LLM had to figure out what it’s optimizing for itself.

I find that as soon as an LLM has an explicit definition of what it’s doing, the exploration/exploitation ratio swerves into exploitation and it gets stuck

I haven't read the paper, so maybe completely irrelevant, but isn't there an analytical solution for a system of N coupled oscillators?

You could maybe switch to a block cursor when you’re in visual mode with

let &t_SI = "\e[6 q" and

let &t_EI = "\e[2 q"

in some auto commands to sort of address this

It looks like grid cells!

https://en.wikipedia.org/wiki/Grid_cell

If you plot a head map of a neuron in the hidden layer on a 2D chart where one axis is $a$ and the other is $b$, I think you might get a triangular lattice. If it's doing what I think it is, then looking at another hidden neuron would give a different lattice with another orientation + scale.

Also you could make a base 67 adding machine by chaining these together.

I also can't help the gut feeling that the relationship between W_in-proj's neurons compared to the relationship between W_out-proj's neurons looks like the same mapping as the one between the semitone circle and the circle of fifths

https://upload.wikimedia.org/wikipedia/commons/thumb/6/6f/Pi...

This would combine really nicely with virtualedit=true in vim

Not that it is necessarily of value, but has anyone got a LLM to run on bare metal?

Something I’m wondering lately is if you are generating tokens fast enough, is restricting the logits actually worth it computationally? If tokens are cheap enough it might be more efficient to validate/discard them as they come rather than place constraints on how they come out. I don’t know how this one works, but the sampling or renormalizing scheme would cost something too right?

This is why I’m convinced vim is great for the ADHD brain. You don’t need to remember where something is or what it looks like, you just have to know the name of what you want to do, and you tell the editor to do it!

I found a heap of their magazines from the 60’s and 70’s in the recycling at my university, and boy are they a good read. I think one reason they used to be so good (speculating) was that the internet didn’t exist. It was a medium where you could regularly learn out about distant things like peculiar mountain goats or an arctic tribe and get a firsthand report of it. Now there’s no shortage of interesting things like that to read, so maybe it’s not a viable strategy for a print magazine.

Here’s a project attempting to do just this!

https://github.com/yacineMTB/talk

I’m amazed that this idea never crossed by mind or my twitter feed. I am usually a dissenter of notebooks, but I could see this being a great use case.

Especially for debugging LM programs. Usually I make logs and have to manually set breakpoints in a program to see where the language algorithm or agent is going wrong, but it would be much simpler to break the algorithm down into cells where I can see where a tangent is arising, and iterate over some different prompts or control flow to get a feel for what can happen at a specific juncture.

Raycast does this really well, among other things

Maybe we will need “human farms” in response to bot farms. Maybe “professional human” will be a viable career option if human created data accrues extreme economic value.

I love Letters of Note. Each post features reflection on letters that were sent/written at some point in time, often but not always by notable people. They are whimsical, short, and have introduced me to many interesting facts and rabbit holes.

https://news.lettersofnote.com/

P.s. My favourite one so far, where among other things, I learned about the infamous and hilarious “please advise” CBS memo surrounding Miles Davis : https://news.lettersofnote.com/p/this-is-my-favorite-memo-ev...