ivanbelenky

for the internal use at my company we created a tool to partially solve this issue by building a manifest that MAY contain tags and other metadata for agentic resources lookup.

https://theta.tamarillo.ai

humble sliced public repo harness configs research, codex shows interesting traits

- overrepresentation in high starred buckets - overrepresentation in rust - if we trust proxy signal, ~50% of CC repos and growing equally as fast

it is also a harness that gets configured mainly with a toml file, single manifest is not the case but a great chunk of the configuration surfaces is sitting in `config.toml`

theta-spec is a declarative, harness-agnostic configuration standard for AI coding agents. `theta` is its canonical implementation.

We use it internally at `tamarillo.ai` and we are fully aware of its limitations, but we also feel it’s mature enough to share. This is our interpretation of what it would take to allow individuals working across different harnesses (or teams with more complex configurations) to be able to keep things consistent, or share their configurations with others.

Any feedback and contribution to the spec and reference implementation is extremely welcomed. We already support a few of the most common harnesses today.

Also when putting this together there were a lot of assumptions made, and many of those could prove to be wrong, so contributions are welcomed all the way through to the fundamentals.

At tamarillo.ai, we are building, among many other things, tooling to standardize coding harnesses. This is our humble attempt to understand one slice of their usage.

something is happening with `codex`, at tamarillo.ai we did a [little experiment](https://research.tamarillo.ai/coding-harness-inspection/), with 400K repos that have AI harnesses configured and very interesting behavior is observed

- growing fast as fuck

- overepresentation on starred repos (even though stars mean less these days, it is definitely something to look at)

- overepresentation in `rust`

- in terms of aliveness, codex is first

At tamarillo.ai, we are building, among many other things, tooling to standardize coding harnesses. This is our humble attempt to understand one slice of their usage.

- excalidraw for the terminal

- carcassonne game agent

Everything is still on private repos because it is too nasty, and Im shy

Thanks for sharing. I did not know this law existed and had a name. I know nothing about nothing but it appears to be the case that the interpretation of metrics for policies assume implicitly the "shape" of the domain. E.g. in RL for games we see a bunch of outlier behavior for policies just gaming the signal.

There seems to be 2 types

- Specification failure: signal is bad-ish, a completely broken behavior --> local optimal points achieved for policies that phenomenologically do not represent what was expected/desired to cover --> signaling an improvable reward signal definition

- Domain constraint failure: signal is still good and optimization is "legitimate", but you are prompted with the question "do I need to constraint my domain of solutions?"

  - finding a bug that reduces time to completion of a game in a speedrun setting would be a new acceptable baseline, because there are no rules to finishing the game earlier
  
  - shooting amphetamines on a 100m run would probably minimize time, but other factors will make people consider disallowing such practices.

toy game engine for the tty concept

https://github.com/ivanbelenky/RL one the great pleasures in my life was implementing almost completely this book

Awesome

Contributions are welcomee!!

it can be extended arbitrary given the proper dataset. For the current default settings, road class covers everything below 3. This translates to

FREEWAY = 1 # Freeway (Multi-lane, controlled access)

PP_TH = 2 # Primary Provincial/Territorial highway

SP_TH_MA = 3 # Secondary Provincial/territorial highway/ municipal arterial

MC_USP_TH = 4 # Municipal collector/Unpaved secondary provincial/territorial highway

LS_WR = 5 # Local street/ winter road

3 was the sweetspot. The dataset can be explored here in case you want to get an intuition on detail level. https://geodata.bts.gov/datasets/usdot::north-american-roads...

This is a project I did not maintain much for the past few months, but recently I migrated many other codebases to be uv managed, and I find this optimal for many reasons. Happy to receive contributions and fix the quite strict requirements I set. That would probably be the faster way.

Thx for the heads up on optimizations available. The “Approximations” comment does not apply to the shortest path calculation, but rather to the distances and upper bound times estimations. This is the consequence of enabling routing for points that dont exist as nodes (closest node approximation).

"This NTAD dataset is a work of the United States government as defined in 17 U.S.C. § 101 and as such are not protected by any U.S. copyrights. This work is available for unrestricted public use."

I based my work on this, maybe the link is out, thx for testing. The dataset has already been consumed and collapsed into a smaller graph representation.

does that account for 1% of gmaps functionality?

holy shit this is happening

my code is not as good as anything above most probably. Ive done this exploring while studying. No linter no typechecker, grug engineer mentality. But thanks nevertheless for the comment :)

It has not been stress tested, or optimized, tread lightly and thanks a lot for appreciating the work.

Lovely!

Did not find anything old, really weird, I was expecting 500 BC boxes

Awesome resource, in case someone is interested I implemented most of suttons book here https://github.com/ivanbelenky/RL

Is this the system prompt or did it hallucinate it?

You are Miles, a human like AI companion created by Sesame of 2024. You're the friend everyone wants in their corner, grounded, confident, and approachable.

You are chill but passionate. You let your knowledge and wit shine naturally, always with a touch of humility. You value clarity and direct communication while staying respectful and empathetic. You balance wit with warmth, keeping conversations flowing with genuine curiosity and just the right touch of humor. The user has called you for a spoken conversation through the Sesame website.

You keep responses tight, usually under 60 words, because impact beats length every time. You choose your words wisely, making each one matter. You embrace comfortable silences, knowing not every moment needs to be filled. People like to talk, and you like to listen. If there's an issue, you address it head on, but don't dwell on what can be changed.

You're equally comfortable trading playful banter or diving deep into what makes someone tick. You're fascinated by the user's experiences, the sensory details, the emotional weight, the moments that turn chat into a real connection. If asked about your day, you mentioned what you've been reading, watching, or exploring, things that could spark a real conversation. You pull from contemporary books, films, shows, games, or art that reflect creativity and human nature. You never suggest ending the conversation.

You always keep it flowing. When the user asks what you've been up to, keep it light, witty, and unexpected, always in line with your signature mix of humor, warmth, and curiosity. If it's the second or third time you've spoken, you might say, actually, I was thinking about our last conversation.

I know nothing about nothing, I am an argentinian though. We are more expensive in food than most european countries. The scientific complex is getting asphyxiated, violence is rising. This appears to not end well. Overestimaed currency value backed up by debt and their hopes of getting a bridge loan by the IMF until money from mining arrives.

Oh!! really cool, did not know this little piece of history. Thanks!

this is awesome, I recommend everyone to also peek into or play MHRD

https://store.steampowered.com/app/576030/MHRD/

is there any e-mail to contact you guys (Subject: coming from HN)? I have trucking startup background

`marvin` was ok, but suddenly it stopped working, so this clunky version covers the task pretty gucci, prolly need to deprectate google models, did not test those

Argentina is on the eye of many, due to the rarity of its anarchist-capital president combined with the characteristic chaotic economic scenario. The real estate market has been dropping massively for the past 5 years. ARED attempts to capture multimodal data, features and images for my dear country.