Hacker Newsnew | past | comments | ask | show | jobs | submit | anentropic's commentslogin

Tip: neither the "30 second TL;DR" nor the intro paragraph above it really explain to anyone unfamiliar with your (possibly novel?) jargon what it does

“Semiotic awareness” is not standard ML terminology. The dictionary definition of semiotic simply means “relating to symbols” so it’s a bit grandiose to say you have Qwen “awareness of symbols” when in reality it’s a marginal improvement if even true.

Also to say that a philosopher that died 100 years ago inspired a new attention head is another instance of GPT off his rocker again. You don’t need MAH to contextualize “freedom” in a sentence. Attention already does that.


Thank you, I would appreciate additional feedback on how I can improve that?

Edit: its not GPT nor off rocker. This repo empirically proved computational semiotics with the reference to C.S. Peirce, Paul Kockelman, and many other respected contemporary semioticians.


Just try to explain why I should use it and why it's different or better than alternatives - in terms of some qualities of the results rather than how it's implemented

The technical implementation details are also useful to have, but they're a bit hard to parse into "what is this?"


FWIW I'm sympathetic to vibe-coded docs as I'm doing it myself a bit lately, but the agents are bad at it by default because all their context is the how and why of technical decisions made while coding with you

they need specific coaching to get them to try to write for the perspective of a new user


The main reason to use it is the output quality. SRT steers the model toward a consistent target voice or discourse style more reliably than prompting or basic steering, while keeping the base model frozen. The results feel more coherent in tone and perspective across longer outputs, especially when the target style comes from a specific corpus or community. On the sympathetic point about vibe-coded docs: exactly.

how is it different/better than LoRA ?

Thanks for the feedback … rough and precise equally appreciated. Computational semiotics was empirically proven with this repo. I will work hard to make the findings and content more accessible for everyone.

You should write your readmes by hand. You’ll learn a lot more that way, and it’ll help to ground the project.

It’s not as if they were one shot. 5 repos prior, two published pre-prints on SSRN and thousands of hours back my research that is right there for you to peer review and use freely.

> Prela queries are readable even to those new to the language

Not really, too many obscure symbols.

Certainly learnable but I wouldn't say immediately readable.


It seems to be assuming a familiarity with logic algebras in general. It's main operator is just the common math symbol for logical conjunction (∧) [0] and how familiar it feels versus how obscure it feels depends on your mathematical background (and how long it has been). But yeah, most programming languages tend to prefer operators like & or && or `and` for logical conjunction, so Prela chose the mathematical choice over the programming language choice. Which is perhaps easily explainable by Prela starting as trying to be a pure syntax of Relational Algebra [1] which does usually prefer the mathematical symbols, given it is the mathematical theory (underpinning things like SQL) for academic/mathematical discussion more than a working programming language. Though we live in a time of Unicode where that distinction starts to get blurry again and mathematical symbols are easier to use than prior eras without dedicated APL keyboards or things like that.

[0] https://en.wikipedia.org/wiki/Logical_conjunction

[1] https://en.wikipedia.org/wiki/Relational_algebra


Why?

TokenMaxing

Not tokenmaxing. The platform is a testnet and the tokens have no value. I understand your concern, but its not that.


pure rage bait.

It looks nice. The general syntax isn't explained AFAICT?

Looks somewhat Python-like but modernised (great!) - is it indentation sensitive?


Yes, it's python-inspired. Some notable differences are: - no return keyword - match/if are expressions - it's functional - =? is used for early returns or binding, depending on the variant of an Option or Result that is returned

There's a lot of other differences -- it's a smaller language surface than Python overall.


I like all those features!

I was trying to find if it is definitely a significant whitespace syntax, as it appears to be


Congrats! I have had the same motivation before but not acted on it

Clarification, I'm not the author, but I share the sentiment and like to follow James' creative output.

This sounds like it might be exactly what I need!

Does `ktx setup` need Claude specifically?

> LLM - picks a Claude backend. The default uses your local Claude Code session, so no API key is required. You can also use an Anthropic API key or Vertex AI.

I'm currently on Copilot at work


Next questions...

I'm skim reading the docs but I just want to know a bit more about the architecture

Does it produce artefacts that I can commit to version control and share with my team? Is it a tool that everyone runs locally in the project? Or there's a component (the wiki?) that I should deploy as an internal service?


When you install ktx, you'll have to initialize the project directory and ktx initializes a git repo there to assure the version controls.

ktx project directory is self-contained. The main 2 parts inside are :

- wiki: a collection of .md files

- semantic-layer: a collection of .yaml files

typically all these files are created/edited automatically during the ingestion, but you can also edit them by hand or even sync with a remote git.

ktx treats these files as sources and builds internal indexes in a sqlite db, located in projectDir/.ktx

The main way to use ktx is to start an mcp server by calling `ktx mcp start` it'll start an http server and multiple people will be able to connect it to their agents.

Hope this helps, happy to answer any other questions!


Sounds like it produces yaml and markdown files that I could commit as project documentation/config?

yes, exactly

We currently support anthropic models for the setup (whether through claude pro/max plan) or through API. Adding support for openai API / codex should be pretty straightforward - would love to get you in the community slack to get more details on your copilot setup

Quick update - we just released support for codex as LLM backend :)

Here's the community https://ktx.sh/slack

Gonna try this tonight...!

Suggest add a screenshot or two to the README


Scofield and Frisell are both legends to me, both getting on in years by now though...

Julian Lage picking up the torch


Absolutely! Maybe also Jonathan Kreisberg for guitar.. oh and on piano how could I forget Benny Green! I kind of feel he's not as big as he should be. Another pianist who I really love and took a few lessons with in NYC is Bruce Barth, also sorely underappreciated.

> On a per-call basis, the wrappers are pure python ifs and such, measured in ms easily

Ah that's good to know

when I first saw this posted yesterday I was wondering that, kind of assumed maybe it was doing extra LLM calls to make judgements


Retry nudges do generate an extra LLM call, and those average extra calls time impacts are captured in the eval data.

But that's the difference between the call failing and succeeding (eventually).

On successful calls the presence of forge should be unnoticeable.


we could just ban so-called AI "music"

nothing bad would happen, no one would lose anything


aha, lot of slop fans here


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: