HN user

bnegreve

2,033 karma
Posts3
Comments725
View on HN

There's no "True" to an LLM, just how probable tokens are given previous context.

It may be enough: tool assisted LLMs already know when to use tools such as calculators or question answering systems when hallucinating an answer is likely to impact next token prediction error.

So next-token prediction error incentivize them to seek for true answers.

That doesn't guaranty anything of course, but if we were only interested in provably correct answers we would be working on theorem provers, not on LLMs

in "pixels wide" the word "wide" acts as a square root operator on the area and means "pixels^(-2)"

Did you meant "pixels^(1/2)"? I'm not sure what kind of units pixels^(-2) would be.

418 I’m a teapot 2 years ago

So many things in the IT world started as non-professional things by non-professional engineers. If you're a professional who lives off one of these things you should be glad that someone wanted to have fun at some point.

Let this HTTP code be a reminder of that.

It is very impressive, and better than anything I've seen before but think something is bit off with the swell. If I had to explain I would say that high frequency waves don't travel on top of low frequency waves the way they do in the video.

every group chat has a n-1 group containing everyone except that annoying member. And if you think your chat doesn’t have such a group, oh boy, do I have some bad news for you.

There are more like 2^n - 1 subgroups, i.e. as many subgroups as the there can be. If you think you're safe because you're already part of one subgroup that complain about the people not included in this particular subgroup, I also have some bad news for you.

I agree and I don't think it is specific to the world of entrepreneurs and/or social science experiments. If your target variable depends on a large number of correlated variables you are very unlikely to formulate the correct hypothesis by accident. This is why you need intuition or good judgement in science, just as much as in every thing else.

- The satellite will accomplish this [precision] by having the usual main Earth-measuring techniques co-located on board [satelite navigation, interferometry, laser ..] When used together, the ESA expects to be able to correct for biases inherent in each technique.

- An updated International Terrestrial Reference Frame (ITRF) will have immediate benefits on satellite-based systems, impacting Galileo-enabled applications in fields like aviation, traffic management, autonomous vehicles, positioning and navigation

- The space agency added that meteorology, natural hazard prediction, monitoring climate change effects, land management and surveying – as well as the study of gravitational and non-gravitational forces as fields – would also see benefits.

- ...

The title is rather ambiguous: evidence that RTO mandates do not increase firm performance is not evidence that returning to office is useless.

we do not find significant changes in firm performance in terms of profitability and stock market valuation after the RTO mandates.

Implementing them in large industrial (or open source) code bases in a maintainable way -- and then actually maintaining that code -- is a different skillset, a different set of interestes,

You're making a very general point on how algorithm research and software development are two different things, which is of course true. However OP's question is genuine: a lot of research in OR is very practical, and researchers often hack solvers to demonstrate that whatever idea offers a benefit over existing solving techniques. There are no reason to believe that a good new idea like this one couldn't be demonstrated and incorporated into new solvers quickly (especially given the competition).

So the quoted sentence is indeed a bit mysterious. I think it just meant to avoid comment such as "if it's so good why isn't it used in cplex?".

Oh, indeed it is mentioned, i overlooked this part.

It even mentions Jupiter's moons later

It is not a coincidence, I reused the example to illustrate my point.

I am not a native speaker, but my feeling is that several can also be used to insist on non-uniqueness. E.g. "Most rich families own several cars" or "Jupiter has several moons" (= not one). I'm surprised that it is not mentioned in the article.

The size of these vehicles is certainly absurd, but flying packages with drones that consume most of their energy to fight gravity does not seem particularly efficient either, (e.g. compared to small road electric vehicles with the same payload, which would have its own practical problems).

Activations of individual neurons are hardly relevant, quoting Szegedy et al. from [1]

we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains the semantic information in the high layers of neural networks.

[1] Intriguing properties of neural networks