esjeon

> I make a lot of tests on how well LLMs get categorization and data extraction right or wrong for my Product Chart (https://www.productchart.com) project.

In fact, what you're doing there is building a local semantic database by automatically mining metadata using LLM. The searching part is entirely based on the metadata you gathered, so the GP's point 1 is still perfectly valid.

> In the near future, nobody will see passing a text through an LLM as a noteworthy amount of compute anymore.

Even with all that technological power, LLMs won't replace most simple-searching-over-index, as they are bad at adapting to ever changing datasets. They only can make it easier.

AlphaGo is an optimization over a closed problem. Theoretically, computers could have always beat human in such problems. It's just that, without proper optimization, humans will die before the computer finishes its computation. Here, AlphaGo cuts down the computation time by smartly choosing the branches with the highest likelihood.

Unlike the above, open problems can't be solve by computing (in combinatorial senses). Even humans can only try, and LLMs do spew out something that would most likely work, not something inherently correct.

I think the point is that it's practically impossible to correctly perform RLHF in open domains, so comparisons simply can't happen.

IIRC, there have been people doing similar things using something close to brute-force. Nothing of real significance has been found. A problem is that there are infinitely many physically and mathematically correct theorems that would add no practical value.

I think all of these can be summarized into three items

1. Search engine - Words like "teach" or "learn" used to be slapped on Google once upon a time. One real great thing about LLMs here is that they do save time. The internet these days is unbelievably crappy and choppy. It often takes more time to click through the first item in the Google result and read it than to simply ask an LLM and wait for its slowish answer.

2. Pattern matching and analysis - LLMs are probably the most advanced technology for recognizing well-known facts and patterns from text, but they do make quite some errors especially with numbers. I believe that a properly fine-tuned small LLMs would easily beat gigantic models for this purpose.

3. Interleaving knowledge - this is the biggest punch that LLMs have, and also the main source of all the over-hype (which does still exist). It can produce something valuable by synthesizing multiple facts, like writing complex answers and programs. But this is where hallucination happens most frequently, so it's critical that you review the output carefully.

Arch has always been 'beginner friendly' according to Arch users, so why can't Nix? /s

But, yeah, Nix can be really nasty to troubleshoot because of the lack of information. The Nix documentation is still PITA, and the much of usual Linux guides (which are everywhere on the internet) won't apply to Nix systems. You have to learn a lot.

If what you want is a simple system with a web browser and some games, this should mostly work, but Fedora Silverblue would provide much more integrated and solid DE for that purpose. Being image-based release-based distro, it does offer pretty decent UX. Nix is rolling release, and everything is scattered just like Arch.

> Educating Users

isn't dumb, because "users" are proven to be the weakest link in the whole security chain. Users must be aware of the workplace security, just like how they should be trained w/ the workplace safety.

Also, there's no security to deal with if the system is unusable by the users. The trade-off b/w usability and security is simply unsolvable, and education is a patch to that problem.

> your average 15 year old

What's the point of this restriction? It really just presupposes the limitation of LLM, so that any negative points would look moot.

EDIT: Also, I tried to discuss this very specific point w/ GPT, but it didn't really "get" it. 15-year old kids would be able to follow through.

I mean that's the point. Reminding people of possible dangers is one very important security mission.

> Guarantee that's going to void your support contract with MS if you have one (or worse).

The MS support ends in 2025. The whole point is to getting support beyond this point.

I’m pretty sure it’s not literally BitTorrent, but a distorted version of it to build private paid peer-to-peer networks.

I thought this was a right-of-publicity case where the "her" tweet basically misled people to believe the voice is of Scarlett Johansson, who's against the use of AI tech like any other Hollywood people?

I've been spending a lot of time on photography (though mostly w/ film cameras) during recent years, and only recently realized this is actually photographed. (It took long because I'm a long-time Linux user, so I barely seen the image.)

Many people say this would've been easier with VFX, but I disagree. The image has highly convincing details that would take a long time even for highly talented VFX artists to nail. Instead, with a camera, you can let the world do the work for you. The studio setup is also very simple (cardboard + acrylic panel + projector + fog) and easy to experiment with. I'm pretty sure photography was the right tool for the project.

... and on the other pathway, there's no fight because everyone's already dead. We fight because we're alive. It's just how life goes.

I might be overthinking, but this "store" thing would not work that well. I blame GPT here, because I'm quite certain that GPT itself can be used to analyze and clone GPT apps, perhaps fully automatically. Even those extra knowledge added by app developers would get extracted through jailbreak, which also can be automated. The store will end up being littered with auto-generated clones.

Yup, Logseq is simply too different. It basically wants to go farther than what Obsidian aims to achieve. That said, some features of Logseq are janky, as they are not refined enough. I think Obsidian hit the sweat spot that is easy to implement yet offers enough features.

Technically, yes. You could even do this using MS Word, and wikis, of course, can do all of these. The only difference is that this is practically the first local-first app that maintains links automatically.

It's not about oil, but about the geopolitics and the American strategy against China and Russia. It's a very complex topic, so it would be quicker for you to just read the fact sheet from the US government: https://www.state.gov/united-states-saudi-arabia-relationshi...

AFAIK, US cannot just "sit out", because the Israel-Palestine conflict can negatively impact the US-Saudi relationship. Even though Saudi shares some common interests with Israel (i.e. Iran), it simply can't look away from the sufferings of Muslims in Palestine, given its role as a major power in the Islamic world.

Also, if Israel takes the control of the Gaza Strip, a lot of refugees will spread in the region, creating more tension in the long-term. I deem this more sensitive than the religious strife, because it can leave concrete, direct, and explicit marks on the neighboring societies.

I think the best scenario for US is that Israel eventually stands down and falls back to the pre-war state - that is, no Israel control over the Palestinian territories. Israel won't have much choice here.

IANAL but it sounds more like SAI is not responsible for any outputs generated using their models and how those are used.

Wayland is good until you hit the corner cases that the dev decided to abandon without leaving any alternatives. We can always have extra protocol that can be optionally enabled, but good luck with standardizing that. It feels as if Wayland people are abusing their committee to keep Wayland as-is, instead of extending it. The protocol dashboard[1] doesn't look exactly good.

[1]: https://wayland.app/protocols/

Right. X11 is just old and widely used. Wayland is entrenched in its own bad decisions. It used to be one legacy and one new thing, but now it's one legacy and another dragging its own legacy.

I saw a person using this. The system had 4090, which can pull about 20-30 iter/sec. This roughly translates to 4 image/sec with 8 iter/image. This allows interactive AI drawing (thou a bit quirky). Once the desired image is reached, the user can re-run w/ 30-50 iterations to finalize the image. This is really cool.

I agree. I'm already sick of reading through political hit pieces, exaggeration, biased speculations and unfounded bold claims. This all just turned into a kind of TV sports, where you pick a side and fight.

> AGI in a very narrow domain

The definition of AGI always puzzles me, because "G" in AGI is general, and the word certainly don't play well w/ "narrow". AGI is a new buzzword I guess.

So, this is actually about: Why I Tend Not To Use Content Negotiation (to serve both HTML and JSON data from the same endpoints).

The author also suggests:

> The alternative is to ... splitting your APIs. This means providing different paths (or sub-domains, or whatever) for your JSON API and your hypermedia (HTML) API.

I believe the alternative has been the norm actually. For example, many front-end frameworks encode UI states in URL, and it's not so sustainable to keep the alignment b/w UI states and data APIs in the long term.

> For those downvoting me

From my own experience, short assertive comments tend to get downvoted on HN. Unlike reddit, votes here are less about agree/disagree.

There's a confusion here regarding "shipping". What I wanted to talk about was shipping a distro (integrated with DSP + profile) as a whole, not shipping software component. My original comment is clearly touching the integration b/w SW and HW, which common x86 distros normally don't enjoy, and shipping with DSP fully integrated (w/ zero friction UX) is very difficult for them. The situation may change if laptop vendors actively involve in the development of Linux distros, but, as of now, the bazaar model is the only answer, and it's painfully slow as noted by the release note here.

EDIT: Also, about recording IRS: both the mic and the room environment impact the response data. Using cheap mic in your room does introduce distortion.

However, I realized I was kinda making an assumption that IRS was being recorded for effects. If you're just flattening the response curve, they should work mostly fine. This can become an issue if it's used for effects (e.g. mimicking 2ch Dolby Atmos) because the distortion can be amplified by the environment.

Also, in the long term, using a standardized setup helps with maintaining the quality of IRS. You really want to be able to reproduce the setup, so that you can further improve the data. It's better than subjectively keep fiddling with EQ sliders.

Yup, and if we correlate accelerometer values to the strength of prompt, it's perfect. Kicking the machine while yelling at it will produce the best result.

> Anyone stuck in this mindset really is going to be hindered from making significant progress in developing utility from AI.

I think this specific line shouts out that this is a typical tribalism comment. Once people identify themselves as a part of something, they start to translate the value of that something as their own worth. It's a cheap trick that even young kids play, but can LLM do this? No.

Some might say multi-modal this, train on that-thing, but it already takes tens of thousands of the most advanced hardware and gigawatts of energy to push around numbers to reach where it is. TBH, I don't see it going anywhere, considering ROI on research will decrease as we dig deeper into the same paradigm.

What I want to say is that today's LLM is certainly not the last stop of AI technology, but a lot of advocates tend to consider it as the final form of intelligence. It's certainly a case of extrapolation, and I don't think LLM can do that.