nsagent

As an NLP researcher, I've come across no convincing research indicating LLMs are "the possible digital twin of human intelligence."

Claims certainly get made, but they are subsequently shot down by future research. This has happened for discussions of emergence [1], reasoning [2], and other human attributes people have ascribed to LLMs.

[1]: https://neurips.cc/virtual/2023/poster/72117

[2]: https://m-cacm.acm.org/blogs/blog-cacm/276268-can-llms-reall...

Microsoft has a very long history of fighting piracy and they are one of OpenAI's main funders. It's definitely hypocrisy.

Former Microsoft CEO Steve Ballmer: "Piracy reduction can be a source of Windows revenue growth." [1]

Current Microsoft CEO Satya Nadella: "Well, we’ve always had freemium. Sometimes our freemium was called piracy." [2]

[1]: https://www.zdnet.com/article/only-consumers-pressure-can-cu...

[2]: https://www.cnbc.com/2014/10/20/cnbc-exclusive-cnbc-transcri...

See https://github.com/minimaxir/hacker-news-undocumented#downvo...

I've got to say, Apple definitely lost me on this one. This feels bad enough that despite having an iPhone since the iPhone 3G, I'll likely jump ship when I need to upgrade unless they make a U-turn on this move.

EDIT: Wow, getting downvoted for saying Apple is losing me as a customer with this move is surprising. I don't understand how people on Hacker News of all places want conformity of thought. I switched from using Microsoft products to Apple when Vista came out. Microsoft lost my trust by that point and Macs were a viable alternative. Despite having a Macbook, iPhone, and an Apple Watch it seems like Apple is starting to lose me with this move.

I wonder if this will open the floodgate for more apps to offer free trials through the App Store, with an in-app paywall linking to an external payment processor with lower fees. Basically, Apple gets stuck with the obligation to maintain the App Store, while taking a considerably smaller chunk of the overall pie. Might make them rethink the value proposition of a closed ecosystem if they can't recoup enough of the cost for running the store.

Decided to rewatch this as I was finishing up my reviews for the next NLP conference. I can't help but think we didn't know how good it was back in 2017 when Ali Rahimi gave this talk.

Please don't be antagonistic. I went back for a PhD after working over a decade in industry, where I had been both a lead overseeing a team of 25 contributors and an engineering manager.

Channeling dang:

> Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith.

I wonder if this is what happened to me. I'm finishing up my PhD and despite being pretty set on going into academia, I applied to some industry jobs back in July that I was pretty uniquely qualified for based on my background and research topic (which is pretty niche). Never heard back. I have a nice postdoc lined up, so it's fine that I never heard back, but I was pretty surprised.

I recently checked and one of those job postings expired and was replaced with an equivalent posting in December. The other job posting (from a different company) is still open.

I've certainly thought about testing LLMs on Connections and I'm glad someone has. It might be possible to increase their performance, but LLMs as-is are not suited for the task.

The problem is that Connections is ultimately a search problem that requires more than simply grouping similar words. There are lots of combinations to assess. I bet if you enumerate, score, then rank all possible groupings, an LLM would perform much better.

In the late 90s, I lived in rural Appalachia and had terrible pings with lots of packet loss on my 56k modem. Found a local ISDN provider and my parents paid to have the ISDN line installed into my bedroom.

I was in heaven! Now I could play Action Quake 2 and Quake 3 Arena without glitching all the time! Those were good times :)

And they are both new accounts.

Sounds like we are in a similar boat :-/

I'm in a similar boat. Not a postdoc yet, but will be starting a postdoc in July (assuming I graduate in May, which I'm on track for).

I've always been interested in the intersection of AI and interactive storytelling. Worked in the game industry for a while, then came back for a PhD when ML really started taking off.

With the current frenzied climate in NLP research, I just feel demotivated. Mainly because I think my research outlook is very different from the mainstream, so I feel my work gets undervalued or ignored entirely.

I spent over a year and a half on my last published research project [1], and it's been largely ignored. Despite having strong reviews after rebuttal, the paper was relegated to the Findings of EMNLP, likely because my research was video game related.

I'm usually the kind of person that focuses on things because I care about them, rather than because others do, but the reality is that hiring decisions in academia (or even industry) require that others value your work. If I truly thought I could do research I cared about and get paid a living, without having to worry about whether others accepted it, I think I'd be much more motivated.

[1]: https://pl.aiwright.dev

My site is a static site hosted on sourcehut, which is having an outage. If it's still down, try https://web.archive.org/web/20240110040908/https://pl.aiwrig...

The article does not downplay resilience. It merely says that resilience has both internal and external components. That is, you can only get so far while just focusing on your personal ability to overcome adversity, but you'll do even better if you combine that with external support.

To be honest, I'm not sure how that can be misconstrued on Hacker News, when people frequently focus on the power of groups working together (whether that's open source, corporations, or other organizations). It's clear that some of the most lauded organizations are ones that often support their members, especially in times of hardship.

This comes from someone who is very much a proponent of perseverance. My first tattoo enshrines this Japanese proverb on my forearm where I can always see it:

  七転び八起き
  Fall seven times, stand up eight

Sorry, just saw your response. I indeed use ProtonVPN.

Thanks for the wonderful resource! Really appreciate your other CS resources as well.

Haha, two hours later and no engagement. Maybe this is just my "Old man yells at cloud" moment.

This code is from a larger research codebase I have that hasn't been released. I just recently decided to break this chunk off, as I haven't seen something similar out there which retrofits existing Pytorch modules to make them directly loadable from a config file.

It seems you might be arguing a that it's untenable to account for individual actions. While a precise measurement is certainly infeasible, there are numerous philosophical frameworks that look to aggregate morality on a scale that can be compared, e.g. Utilitarianism and Consequentialism.

Seems like you could do the same with rqlite [1], since SQLite supports JSON.

[1]: https://rqlite.io

If you are correct, let's ensure the browser duopoly is broken up first, rather than forcing Apple to open the market for Google to corner it. Then it will truly be a monopoly and all consumers lose.

I don't know if I'm in the minority or the majority, but I can say my experience definitely doesn't match the view you're invoking.

When I procrastinate due to my perfectionism, I pretty much never do something enjoyable instead. Rather, I frequently lament that I don't have enough time for fun activities.

When I'm procrastinating in a WFH setup I could easily play video games, watch a show/movie, read a book instead, but don't. It's almost as like a punishment, rather than a reward.

I understand the sentiment, and often feel this way too, but I think it's reductive to assume the answer is to let large companies/investors to fail.

Ideally, if we ensure adequate diversity in the market, then having companies default makes sense -- then one company going under won't trigger a string of defaults that can cripple a sector of the economy.

But when we let businesses concentrate effort or corner markets, then there can be catastrophic consequences to the economy when these large entities go belly up. Efforts should be put in place to mitigate these issues.

Ultimately, I think the pragmatic approach is to assess defaulting on debt on a case-by-case basis. I'd prefer if a panel of economists across the ideological spectrum came to a consensus for each case based on the latest understanding of the market, but I know it's more likely going to be politicians choosing the argument that best fits their worldview.

I also occasionally try Brave search when a DDG search fails. Sometimes Brave finds what I want, but I frequently get Captcha (and now proof of work) challenges that are quite annoying. I don't get this with any other search provider (though StartPage would frequently do this a while back). I hope this is just a phase, because I would likely use Brave Search more if not for this issue.

You can also use fzf with ripgrep to great effect:

[1]: https://github.com/junegunn/fzf/blob/master/ADVANCED.md#usin...

Not just every launch, but also if you leave it long enough, it'll pop up again during the same session. This was what made me finally uninstall it.

And while Firefox has lots of flags and ways to tweak things like this, it's not straightforward and requires time to interrupt my workflow to address. Definitely not worth it for me on macOS. I might change my tune if I switch to Linux, but at that point I'll likely look into something like LibreWolf.

Orion, which is based on Safari [1], supports uBlock Origin [2].

[1]: https://help.kagi.com/orion/why-orion/orion-vs-safari.html

[2]: https://help.kagi.com/orion/browser-extensions/ublock-origin...

Legitimately curious. Considering Firefox Quantum was released in 2017, is it even possible that my M1 is running any other version of Firefox? Doesn't seem likely considering the copy on the Mozilla site [1] for Firefox Quantum:

> Firefox Quantum was a revolution in Firefox development. In 2017, we created a new, lightning fast browser that constantly improves. Firefox Quantum is the Firefox Browser.

[1]: https://www.mozilla.org/en-US/firefox/browsers/quantum/

As someone in the Apple ecosystem, Firefox is not a very compelling alternative. I've certainly tried: Firefox Focus was my main iOS browser for a few years (I know it's Webkit-based) and I've tried switching on desktop as well.

On desktop, various Firefox updates would change my settings subtly and nag me about features I had no interest in. On top of that, it was simply a slower browser, both with DOM manipulation and Javascript.

For example, when I made my latest research project (https://pl.aiwright.dev) Firefox would routinely struggle with the large complex graphs (over 100k nodes, though only a few hundred nodes attached to the DOM at any given time). Safari handles it with no issues whatsoever. I'm sure there are likely workarounds to speed things up on Firefox, but I don't have the time or energy for something like that since it's a research project, not an end-user facing product.

Maybe you can make the argument that switching to Firefox will incentivize improvements, but so far that hasn't been my experience. I've now moved away from Firefox to things like Orion (which is still quite buggy, but useful enough). Safari seems to be getting better over time, with things like containers now and extensions like Userscripts [2].

[1]: https://pl.aiwright.dev

[2]: https://github.com/quoid/userscripts

I tried Gentoo Prefix for a while on macOS, so I could use the same package manager across OSes, but always compiling from source got a bit tiresome on a laptop, so I went with MacPorts instead.