helltone

How do you think of uv vs pixi?

Follow up: has anyone suggested the projects collaborate or eventually converge?

I thought glass was a solid?

I'm building in robotics. Setting up a new 3d camera today. I found that the 10m active USB C cable that I bought transfers power in both directions, but only transfers data in one direction, it turns out to be some weird video USB variant. Next I needed to plug a gripper into a modbus controller. That uses an M8 8-pole 20cm cable. The controller manufacturer recently decided to switch from male to female connector, so now the cable needs to be male-male. After searching online for hours, I believe that is impossible to find as everyone only sells male-female cables.

I'm continuously surprised by how difficult it is to plug things together and how non-descriptive cable "standards" are about the actual capabilities of cables and connectors.

Is there anything like a linter to force you to stay within some subset of c++? I like the language, but it is hard to avoid language constructs that are outdated or enforce a single (or a few) ways of doing things. A c++ subset could be nice.

In the UK, probably nuclear.

The goal of automation is not necessarily reduce personnel, it could be increased reliability, reduced risk to life, reducing waste from scraps, faster lead time, etc

Do you have some references for the pick and place and other reconfiguration things you mentioned. I've been out of this space for a while but last I checked these were still incredibly challenging things to get right.

No amount of fine-tuning can prevent models from doing anything. All it can do is reduce the likelihood of exploits happening, while also increasing the surprise factor when they inevitably do. This is a fundamental limitation.

Right I think I see it.

This is insanely cool.

But then there are performance tradeoffs in reusing intermediates vs recomputing that I think you can't represent.

Some of these may affect numerical stability btw. See eg https://herbie.uwplse.org/

There is so much potential in this project.

I have a background in program analysis, but I'm less familiar with the kind of kernels you are optimising.

- Can you give some more insight on why 12 ops suffice for representing your input program?

- With such a small number of ops, isn't your search space full of repeat patterns? I understand the will to have no predefined heuristics, but it seems that learning some heuristics/patterns would massively help reduce the space.

This heavily depends on share classes and preferences. Surely the new investor wants better terms. The issue isn't so much dilution as a preference but added risk of never even getting a payout at all.

This resonates with me a lot. Can I contact you? My email is in my profile.

Every time I see these headlines, the tech seems to be at least 10 years away from product.

- demos done in a lab controlled environment without the crazy things that happen in a real world.

- no humans nearby so none of the safety features that would be needed should this thing work alongside/near humans.

- no regards for economics, expensive vision models, expensive hardware, no consideration for maintenance and repair costs

This is very cool. I'm wondering if some of the templates and switch statements would be nicer if there was an intermediate representation and a compiler-like architecture.

I'm also curious about how this compares to something like Jax.

Also curious about how this compares to zml.

No? Or maybe I'm missing something. If the goal is to be able to bound the computation of f, you can:

1) compute f with interval arithmetic

2) compute f normally and f' with interval arithmetic

3) compute f rounding towards zero, compute f' from f rounded towards infinity, and round f' up (if f positive) or round f' down (if f negative).

In all 3 cases you can use what you computed to figure out bounds on f, (1) is direct, the other two need extra work.

I think perhaps this could be done in other ways that don't require interval arithmetic for autodiff, only that the gradient is conservatively computed, in other words carrying the numerical error from f into f'

How does the chart generation work under the hood? It's quite magical. Also how did you build the spreadsheet interface it's very cool.

I'm building a tool to make ml on tabular data (forecasting, imputation, etc) easier and more accessible. The goal is to go from zero to a basic working model in minutes, even if the initial model is not perfect, and then iteratively improve the model step by step, while continuously evaluating each step with metrics and comparisons to the previous model. So it's less ml foundation research, and more trying to package it in a user friendly way with a nice workflow, but if that's interesting feel free to reach out (email in profile).

Backtracking idea is interesting, could maybe diffusion help? At some point it turns into sat solving.

Can I reach out to you directly to hear more? My email is on my profile.

How do you modify ASTs?

Program equivalence is undecidable, in general, but also in practice (in my experience) most interesting cases quickly escalate to require an unreasonable amount of compute. Personally, I think it is easier to produce correct-by-construction decompilation by applying sequences of known-correct transformations, rather than trying to reconstruct correctness a posteriori. So perhaps the LLM could produce such sequence of transforms rather than outputting the final decompiled program only.

Off topic but how is MCTS usually implemented efficiently? It has a branching structure that doesn't seem parallelizable (GPU).

It seems GNNs operate on a fixed topology. What if I want to approximate some transformation of the topology of the graph? For example learning how to layout a graph, or converting program abstract syntax trees to data flow graphs.

I'm curious and it's not obvious to me: what changed in terms of weight initialisation?

None of the example tasks come with example solutions. Would you mind sharing what the solutions found look like?

In addition to what is mentioned by the other comment, also threading and io.

Off topic: I think there's some opportunities for making bayesian inference technology more accessible, and I'd love to chat with other people in this space. Email in my profile.

std::start_life_time_as and std::launder honestly feel like a massive hack

Possibly off topic, but I was wondering: what are the most comprehensive data structure benchmarks out there?