Nesco

I am pretty sure uncouplers will make a come back. Just something a little more targeted and safer than good ol’ DNP

Yeah, trying to create one right now to try to do some E Coli + Desnoyer’s style flower experiments.

I really underestimated the cost

Zuck wouldn’t have leaked it on 4chan of all the places

To people using org mode, how does it help you more than Markdown? Genuinely curious because I tried at some point and it felt too heavy.

Maybe because I am a vim user instead of eMacs?

[flagged]

I have been working on a symbolic solution to ARC-AGI: https://github.com/nesco/DPS

Shapes are converted into trees of paths, and those trees are compressed using a symbolic lookup table

A LLM can then use those symbols to reason about shapes, and a tree edit distance can be used to group them in categories

It’s the French spelling. Saying it’s only to indicate the disappearance of an “s” is reductive

You can make the difference between a single wavelength color and a composite color which looks the same, by looking at objects nearby.

If they are of one of the composite colors, they should appear in their natural hue

Else they will just appear darker

No, in fact you did!

I knew a good friend who lived in Okinawa, and never locked his front door even when going abroad

You could grow the nominal wealth of your household by removing the lock

Then why do you not have a full open-order policy? It will definitely grow the nominal size of the economy

Technical Product Manager / Product Engineer | Paris | Remote OK | Relocation OK

  Location: Paris, France
  Remote: Flexible
  Willing to relocate: Case-by-case
  Technologies: Product Management, AI/ML (LLMs, NLP, Pytorch), Python, Javascript, React, FastAPI, SQL, MongoDB, AWS, Google Cloud.  
  Résumé/CV: https://drive.google.com/file/d/1PjAo-PtmdDnmPqd5esLQx54sK01InWDC/view?usp=share_link
  LinkedIn: https://linkedin.com/in/emmanuel-federbusch
  Github: https://github.com/nesco
  Email: emmanuel at federbusch dot fr

Hi HN, I am a Technical Product Manager / Engineer looking for new opportunities.

Most recently, I served as CTO for a startup organizing pharmacological data for LLMs, selected for CES 2025.

Previously, as a Product Manager at Allphins, I helped grow ARR from $125K to $2M by extending platform capabilities for reinsurers. And before that, as a ML Engineer @ IBM, I helped BNP Paribas deploy OCR internally to automatize operational processes using OCR, saving significant manual effort.

Feel free to contact me, even if it's just to connect!

From my own experience on a codebase using a lot of custom algorithms on trees and sometimes graphs.

There were qualitatively leaps in my day-to-day usage:

Claude Sonnet 3.5 and ChatGPT O1 were good for writing slop and debugging simple bugs

Grok Thinking and Sonnet 3.7 were good to catch mildly complicated bugs and write functions with basic logic. They still made mistake

But recently, Gemini 2.5 pro has been scary good. I liked to made fun of the feel-the-AGI crowd but for the first time a model made me raise an eyebrow

It can one shot unusual function with complicated logic and subtle edge cases

Hey rossng, at which email address could you be contacted?

There is no one truly in charge, it's a distributed standard

Well, from my very limited comprehension of diffusion models, they apply to fixed length structure, mostly from a continuous space. Maybe a way to make them work with tree structures could be found - that's no trivial task

LLMs are autoregressive models. However, the notion of order in ASTs might be nonexistent, especially for parallel branches of computation/control flow. You could attempt to untangle each branch into N sequences, but this would erase control-flow information.

Even when there is an objective ordering of the children of every node, you still have four traversal options: {preorder, postorder} × {BF, DF}.

Note: For children lacking an objective ordering, you might apply generic rules to define a traversal order, but you’d end up with as many depth-first traversals as there are possible orders—essentially a crude heuristic. If you want the evaluation order to be dynamic at each step (e.g., using RL), the complexity grows geometrically worse. That’s been my experience tinkering with a custom AST DSL for ARC-AGI.

Unfortunately it’s the reason indeed

At the very first glance, it's like ARC 1 with some structures serving as contextual data, and more complicated symmetries / topological transformations.

Now, I wonder what surprises are to be found in the full dataset.

The focus on solving cost efficiently discrete tasks might actually lead us towards deep learning systems that could be used reliably in production, and not just give a whoa effect or need to be constantly supervised

Most of the usual components listed should be standard at some point. The main issue is still that the Web was thought for documents, not apps. The current mess is partially due to OS vendors not agreeing on a standard, cross platform GUI API, so it was done via the Web

Modern LRMs do have some tiny degrees of intelligence

Goodbye HashiCorp...

Here is the comparison:

Claude Electron app vs Claude Safari PWA

Pros:

* Can be opened with a customizable shortcut

* Takes less RAM (400 MiB < 600 MiB)

Cons:

* Needs to be downloaded

* Weight 150 MiB

* Slower to start

* Loose language localisation

* Loose Safari's capabilities:

  - Printing the content of the current page

  - Speech2Text

  - Accessibility

  - ...

On macOS, why have an electron app when you can add the webpage to the dock as a PWA?

I know a guy who took 1000ug. He told me he had side effects / visual glitches for about an entire week

Inside a black hole, the center is in your future, it’s a point in time where geodesics end

Time doesn’t seem to end in the universe at large…

The universe has no center, a black hole has one. The limits of the visible universe is an horizon on your frame of reference

In nerdspeak, the geometries are not the same, one is isotropic the other anisotropic

This mass to event horizon radius relationship is a property of a Schwarzschild spacetime geometry, globally the universe has a FLRW spacetime geometry

Reminder that sleep apnea is mostly caused by something hindering the normal airflow. It might require a jaw/orthognathic surgery or a nasal obstruction surgery, depending on the cause being a jaw or nasal dysmorphia + muscular rehabilitation

The ARC-AGI challenge / Arc Prize. It helps me enjoy programming even after becoming PM. I got a bit rusty on structural induction for trees, it’s now all fresh again. After playing with it a bit I am now convinced a A* on the right kind of AST structure will be sufficient to make a significant breakthrough in the scores