graemefawcett

6 ms·

1.
by graemefawcett 5mo ago
The other cases are the original request is still in flight or never occurred. The former case was explained by the prior comment, one request is processed, the other is returned by 409. The system cares little for which is which and neit
2.
by graemefawcett 6mo ago
Stateless router to router with lossy scratchpad is a step up, still not going to ask it to check my Lisp. That's what linters are for
3.
by graemefawcett 6mo ago
RAG is like when you want someone to know something they're not quite getting so you yell a bit louder. For a workflow that's mainly search based, it's useful to keep things grounded. Less useful in other contexts, unless yo
4.
by graemefawcett 6mo ago
It's not just an issue of tokenization, it's almost a category error. Lisp, accounting and the number of r's in strawberry are all operations that require state. Balancing ((your)((lisp)(parens))) requires a stack, count r&
5.
by graemefawcett 7mo ago
Exactly. They've implemented a VM inside a transformer, turned an O(1) memory access call into O(n), optimized it down to O(log n) and wrote a post about how smart they are. It's a nice bit of engineering, if you don't subsc
6.
by graemefawcett 7mo ago
Basically just madlibs - the models generate intermediate tokens that help predict a better answer based on training (RLHF & otherwise). They tend to look like "reasoning" because those tokens correlated with accepted answers
7.
by graemefawcett 7mo ago
I believe that's I and the paper are both saying as well. The LLM is pure routing, the constraints currently are located elsewhere in the system. In this case, both the constraints and the motivation to perform the work are located i
8.
by graemefawcett 7mo ago
This is exactly how problem solving works, regardless of the substrate of cognition. Start with "all your questions contained in randomness" -> the unconstrained solution space. The game is whether or not you can inject enough
9.
by graemefawcett 7mo ago
Connecting them is easy, one is the math of the exchange and one of the state machine. A better question might be why no one is paying more attention to Barandes at Harvard. He's been publishing the answer to that question for a while
10.
by graemefawcett 9mo ago
I think the biggest benefit is bandwidth more so than efficiency. This gives you multiple streams to mux which and a means to control their mixing. The biggest innovation I think may have been accidental. The doubly stochastic matrix impl
11.
by graemefawcett 9mo ago
Are you aware that there are certain members of your very own species that are as intelligent as you or I, who lack those qualia. Non standard cognitive architectures are already coherent. Even if they were, why do you think qualia cannot
12.
by graemefawcett 9mo ago
I took that idea just as far as I could and landed here https://zenodo.org/records/18181233 It parses the AST out of it and then has a play We're using it as an agentic control plane for a fortune 500s developer p
13.
SyneState: Machine Synesthesia with Manifold-Constrained Hyper-Connections(zenodo.org)
2 points by graemefawcett 9mo ago|1 comments
14.
by graemefawcett 9mo ago
We propose SyneState: a communication prosthetic built on DeepSeek's Manifold-Constrained Hyper-Connections (mHC) architecture. By parameterizing cross-channel mixing as a doubly-stochastic matrix constrained to the Birkhoff polytope,
15.
by graemefawcett 10mo ago
Besides, if everyone could paint the Sisten Chapel, then we'd have works equivalent to the Sistene Chapell everywhere. Why is that a problem? That to me sounds like the opposite of a problem. Used effectively, these tools are elevators
16.
by graemefawcett 10mo ago
Is art then just the outcome? The artifact that was produced? What's your criteria then for who is allowed to produce art? If allowing everyone to create it lessens its value such that it becomes worthless, there must be a cutoff. If
17.
by graemefawcett 10mo ago
I've been doing it for months, it's lovely https://tech.lgbt/@graeme/115749759729642908 It's a stack based on finishing the job Jupyter started. Fences as functions, callable and composable. Same shape
18.
by graemefawcett 10mo ago
I had a sickie yesterday with nothing better to do than attempt to prove isomorphism across domains Like drawing lines on a mids menu but for grown ups when the price is right isn't on
19.
DNS as a Filesystem: A Practical Study in Applied Category Theory(loss.dev)
1 points by graemefawcett 10mo ago|3 comments
20.
by graemefawcett 10mo ago
Two systems are isomorphic when they admit the same morphisms—when the set of valid transformations applicable to one equals those applicable to the other. If Hom(X, DNS) ≅ Hom(X, Filesystem) for all relevant X, then DNS ≅ Filesystem. Where
21.
Abusing DNS TXT Records as a Planet‑Scale Key‑Value Store for Fun (and Geese)(graemefawcett.ca)
1 points by graemefawcett 10mo ago|0 comments
22.
by graemefawcett 10mo ago
This is the right attitude. GenAI isn't a thinking machine, as much it might pretend to be. It's a theatre kid that's really motivated to help you and memorized the Internet. Work with them. Let them fill in your ideas with
23.
by graemefawcett 10mo ago
I posted the Rspec version of this earlier this week on Obie's keynote on Ruby + TDD + AI. I've been working on a project to turn markdown into a computational substrate, sort of Skills+. It embeds the computation in the act of r
24.
by graemefawcett 10mo ago
The nice thing about the transformer architecture is that they can cross these domains, to an extent. I have a very spatial way of reasoning through problems and using an LLM, especially an agentic one like Claude Code with access to my lo
25.
by graemefawcett 10mo ago
They're "repeat after me" machines, not "think for me" machines. For the former task, they're brilliant but everyone seems to have fallen for the branding and forgotten the technology behind it. Given an input
26.
by graemefawcett 10mo ago
Sorry, I had to get my kids off to school (there's two of them) and daddy tends to need some help on the best of days, but the abstraction is already designed for a polyglot runtime and my entire current platform is based on Ruby (she&
27.
by graemefawcett 10mo ago
Apologies, errant drop of water on an already cracked screen caused an early posting. The live markdown provides a substrate that causes the tests to run when the user guide or reference document is read. We use these reference documents i
28.
by graemefawcett 10mo ago
We do something similar with automation we're calling Looking Glass Development (there's a theme). https://tech.lgbt/@graeme/115608630019342209 We marry the user guide or reference manual to the unit tests in
29.
Loss.md(tech.lgbt)
1 points by graemefawcett 10mo ago|0 comments
30.
by graemefawcett 11mo ago
I agree with what Mario says overall and I can be honest, I don't really use MCP I don't think - at least not what it's intended for (some sort of plugin system for extensbile capabilities). I use it for an orchestration lay

More ›