sharma-arjun

Not a core paper, but I found Formal Algorithms for Transformers [1] (a Google paper from 2022) to have a great pedagogical style.

[1] https://arxiv.org/abs/2207.09238

Skim is amazing, I've never had it crash on large documents or with keyword searches as Preview almost always does. Also has a great note-taking system for scientific / academic work. Not sure why it flies under the radar so much.

That's a fair critique, although I think you could mitigate that by decreasing the desired arclength - reducing Del S to 0.5 from 0.85 in your example mostly fixes the problem.

In the context of the game described in the post, the main variable we were concerned about was u_x, because we scaled difficulty by increasing it over time. This in fact results in flatter parabolas, so I actually had quite the opposite concern of what you outlined and might have over-optimized in that direction!

Thanks for taking the time to read and try out what's described in the post. It's my first time working on a technical blog, and I'm extremely grateful for any constructive feedback / criticism :)

Yeah, you put it better than I did - I'm not secretly raining an LLM lol.

I'm not really interested in changing my information diet to help me follow the news, but I feel like I'm missing out on the high-quality and original writing that I'm sure must exist _somewhere_ in much larger quantities than I can find through my current sources, and that there must be a way to easily aggregate them all so that I don't miss out on the sources that publish less frequently.

Are classical machine learning techniques that don't involve neural networks / DL not considered "learning-based systems" anymore? I would argue that even something as simple as linear regression can and should be considered a learning-based system, even ignoring more sophisticated algorithms such as SVMs or boosted tree regression models. And these were in use for quite a while before the ImageNet moment, albeit not with the same level of visibility.

The causes of homelessness are a little more complex than that, at least in the US:

https://homelessness.ucsf.edu/blog/vacancies-are-red-herring

I think the dumbfoundedness didn't arise from the inherent quality of the music, but his shock at the fact that five people in England didn't know who the Beatles were.

That said, I do think this movie missed out on a lot of potential, by not exploring what the actual significance of the Beatles to the world would have been: it never really asks what the world would be like had the Beatles never existed, which I think is a really interesting question, and instead develop a typical Hollywood plot.

This indeed seems to have been a heavy focus of their research team in the past year, eg. "Efficient Large Language Model Inference with Limited Memory" [1] and OpenELM [2]

[1] https://arxiv.org/pdf/2312.11514

[2] https://arxiv.org/pdf/2404.14619 (with 1.1B parameters, this appears to be their attempt at building a lightweight LLM)

The licences for open-source models are generally designed to restrict this kind of usage by big tech. See the "Additional commercial terms" of the Llama 2 licence, for example:

> If, on the Llama 2 version release date, the monthly active users of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.

There is probably a need for an alternative term to describe these kinds of models, because they're not exactly open-source in the traditional sense

I'm not the OP, but recently, YouTube has started adding an automatic summary below videos. Before that, they also started adding automatic chapter titles, which, in my experience, are surprisingly good for navigating slide-based talks but are otherwise fairly hit-or-miss

This is only true in the western world. Globally, education has been trending upwards, at least until the pandemic.

What is happening is a noticeable decline in the levels of trust in education and science. Generally, in the rich world, education used to be assumed to be a necessarily good allocation of resources, and whether you accessed it or not was largely a function of your wealth. Now, in some pockets of the world, this is no longer true.

I also think that space travel will only see a revival in public interest if it provides viable economic value or becomes a renewed front for competing nationalism, neither of which appear to be extremely likely in the short term. Up to that point, I think it'll continue to be a playground for billionaires.

Wyden is far removed from the part of the government which engages in surveillance. He's the same person who was questioning James Clapper in Congress about mass surveillance before the Snowden leaks [1].

[1] youtube.com/watch?v=QwiUVUJmGjs

Yes, but in the vast majority of cases, it's hard to tell just by reading a paper if there's been dishonesty somewhere in the pipeline.

Also, the LK-99 example is an exception, not the norm–the chances of receiving significant attention for a replication study are near zero in almost all other cases.

ChatGPT Plus was never intended to be OpenAI's main revenue stream. The real cash is in the GPT API, which they're currently making good money on, but which also directly competes with a GPT API offered by Azure.

OP might have a point: if OpenAI declines, devs might prefer the Azure API over the OpenAI API for factors like stability, quality, response time, better integration with existing Azure stack, etc.

One useful trick to avoid some of the pitfalls of ChatGPT for non-creative tasks is to respond to its output with something like:

_"Point out incorrect assumptions or statements in the above answer."_

I got the idea from Khan Academy's [implementation](https://youtu.be/A7REVn9gzgs?t=1208) of GPT-4, where they allow it to 'think' by generating an internal response first, before generating a final response. This apparently improved the veracity of its output by a significant margin.

If ChatGPT doesn't double down on its original answer, then I know definitely not to trust it.

This doesn't solve the underlying problem, because it doesn't ultimately correct the error in a reliable way, but it does help me avoid some of the more obviously garbage output.

Yeah, it's extremely common for people to be extremely prolific programmers in high school, without having much knowledge of foundational computer science or algorithm concepts.

Source: Am a high school programmer

Dropbox Paper should meet your requirements. I tried switching to it a few years ago, but the problem is that the other people I'm collaborating with always want to use Google Docs, and when I'm writing a document by myself, I'd much rather just write to a local file.