kenmacd

Your base assumption here is that the 'crap' is actually 'junk'. Let's look at the easy one here, languages. Talk to someone that speaks multiple languages and they'll have examples of concepts in one language that are difficult to express in another. The multilingual person, or someone who just speaks a different language than you, will think differently[1].

Does the LLM take advantage of this? I don't know. It wouldn't surprise me if it did, and if it doesn't now I'd bet it will in the future. Either way though, throwing away those other languages could make the model dumber. As you allude to, there's a balance between intelligence and knowledge.

(in case you hadn't thought of it, those 'tools' can also be other LLMs with more specialized knowledge in a particular field. For example a 'translator' model)

Other 'facts' could also have more merit than it would first appear. Sure, one particular person's shoe size might not be needed, but if you were to filter out shoe sizes in general then the model might not be able to suggest how to find properly fitting footwear, or might not suggest that your back pain could be related to your shoes.

> That would be a worthwile endeavor and maybe even possible without boiling the oceans.

I think it's important to keep in mind that we're very early in the AI journey. Look at the power requirements of early computers versus the ones we use today. I'm all for keeping energy usage in mind, but I'd be careful with hyperbolic language as things are changing so quickly. Tasks that would have taken multiple GPUs can now run on my laptop CPU.

[1] https://www.edge.org/conversation/lera_boroditsky-how-does-o...

They run in to async, but do they run in to the internals of how that code is executed?

The chapter that example is from includes the disclaimer:

> In this section, we'll cover the underlying structure of how Futures and asynchronous tasks are scheduled. If you're only interested in learning how to write higher-level code that uses existing Future types and aren't interested in the details of how Future types work, you can skip ahead to the async/await chapter.

Would it though? Just running things 'async' is simple, but it's also simple in Rust. I wouldn't figure writing an Executor would be any easier in Go than the Rust example at https://rust-lang.github.io/async-book/02_execution/04_execu...

> For instance, calling out LLMs for what they are is appropriate to me,

Well perhaps you're unaware of the tone of your writing, or perhaps I am, but I'd suggest that your original comment sounds rather hostile towards the field. Following it up with a comment like 'for what they are' is again not a comment that fosters discussion or thought. It implies you know 'what they are' and anyone with another opinion is simply incorrect.

> One of the biggest criticism against all of this AI slop is its capacity to “fool” people

I considered not using that term because I thought we might have this understanding. I put 'the people generating them' in hoping to avoid it, and thankfully you noticed.

I see we still weren't quite talking about the same people, as I misunderstood your 'people' to be person that might be paying to use the flux.1 image generation models. If I'm understanding now you're focused on the ones that might take what's generated at face value because they haven't heard about hallucinations.

On that front I'd just mention that it's worth spending some time looking in to the research being done on the topic. It's interesting some of methods being looked in to in order to prevent it. (Also your 'glue pizza' isn't in the same category of hallucinations because it was simply summarizing websites pull from a search).

> A field that stole the whole of human creativity to sell it back to us for the empowering of the unethical sociopaths who control it. > > Why yes, I have considered my reasons; have you?

How do you steal creativity? I'm really not sure I understand what you mean here. I'm not even sure if you're using it in the "you wouldn't download a car" meaning, or that it's literally replacing creative work so stealing that... but then does that even make sense?

Either way, consider this, maybe the field will burn through a ton of billionaire money and all that will come out of it will be a bunch of open research papers and a bunch of open models. Would that really be so bad?

There's a lot that I believe to be misinformed in your post, but rather than address most of that I'll ask if you've considered your reasons that you're so hostile towards this field?

I doubt you refuse to work with anyone that's read a book that you refused to read. You'd probably also agree that while the topic of those books might be objectionable to you there's knowledge to be gained by reading them. So why is "AI read a bad book" a reason to write it off?

You also assert that it's fooling people. Have you tried the image generation AIs? Do you really believe these images are 'fooling' the people generating them?

If you're just looking for something that's similar in the auto-complete category, I'd suggest adding the Continue plugin/extension (https://docs.continue.dev/quickstart#jetbrains) and pointing it at Claude 3.5.

If you haven't tried it yet I'd also suggest trying https://aider.chat/. It's outside of the IDE but I find it's the right level of 'agent' for where we currently are with models. Even if you use it for nothing else it's nice to be able to run the `/test` command and have it use the test output to fix errors.

I have to agree with mid-kid here. GrapheneOS is great for security, but some of that ends up directly in conflict with privacy.

For example years ago I opened an issue around runtime-patching applications (https://github.com/GrapheneOS/os-issue-tracker/issues/284) and was told it didn't make sense in the OS.

Putting aside the validity of arguments on either side of that issue, runtime patching allows me to very quickly excise large swaths of anti-privacy code across every app on my phone.

If I was to switch to GrapheneOS I'd have to accept that I'd be giving up a fair amount of privacy for the added security.

> by providing it with 10 to 20 examples of a chat dialogue in the initial prompt context

Can you recommended examples (or a source of examples) that would have models act similar to the instruct models?

I'm sure you're right. I'm not sure how the 'the bad things with AI' shake out, but I'm trying to be optimistic. For example it's not like customer supports bots aren't already commonly used, and it's not like they can really get any worse.

And even if the academic side is a small amount of the funding, it's at least something. Some of the rest is being spent training models I can download and run forever more.

I guess mainly I'd rather focus on the interesting technical and scientific side of AI rather than worry about the investors.

> "Capital continues to pour into the AI sector with very little attention being paid to company fundamentals,"

Maybe I'm just overly optimistic, but on a wider society scale I see a lot worse things they could be pouring money in to (eg Ads). Even if they never make money on it, at least we end up with a lot of open knowledge through all the papers that have been published and even some open-source models that can do some neat things.

Bit of a strange article. It seems to be trying to claim AI is 'hype' while at the same time claiming it "has the potential to ruin what were stable [hand weaver] careers".

> as customers remained reticent to make the switch away from gas-powered cars.

I find this disingenuous. I think a lot of customer would gladly switch if there was anything electric and remotely affordable.

A while ago I hacked root on my LG tv, but eventually I gave up on the war against spying/ads and just unchecked all the agreements, unplugged the network from it, and hooked up a $20 Onn 4k tv box instead.

Now it does everything I want. It has an unlockable bootloader, root, can run custom roms, all the good things.

> I just haven't found the motivation yet.

I moved from Arch to NixOS. In case it helps these are my main motivations:

* In Arch, very rarely, I'd apply an update that would break things. Going back to a stable system was a pain. I ended up with some snapper setups and a list of commands to convert the latest snapshot to a r/w 'root' that worked well enough. With NixOS I just reboot to the previous generation on any issue.

* For development I'd often use third-party tools to manage multiple language versions. For example I used pyenv. With Nix it's much easier to just use python39, or nodejs-18_x, or whatever as Nix makes this really simple.

If you do decide to try it out I'll recommend using the github query 'path:*.nix ___' to copy off others, as the main complaints about Nix seem to center around the language.

Like wayland, podman was one of those things that took a lot of "I'll give it another shot? Nope, not ready" before I could switch.

Lately the only real incompatibility I run in to with podman is that the handling of `RUN --mount=type=secret` in a Containerfile/Dockerfile is a bit broken (https://github.com/containers/buildah/issues/5282).

You're also trading some amount of risk due to not having newer safety features. For example electronic stability control has been shown by study after study to reduce accidents. With your manual brakes there's simply no way to apply the brake to one wheel only to pull a vehicle out of a skid.

Example study: https://www.tandfonline.com/doi/full/10.1080/15389588.2014.9...

> In line with previous results, this study concludes that ESC reduces the risk for single-vehicle injury accidents by 31% when controlling for various confounding factors related to the driver, the car, and the accident surroundings.

> After 3 Pixels, ALL with issues that never should have been present, I decided Google can shove it's Pixel phones.

I can understand issues, but what I can't forgive is Google not taking responsibility for those issues.

My mother had a Pixel 3 that died an EDL-death at around 3 years old. She wasn't a heavy phone user, and this was obviously a defect, but yet Google would do nothing (Case ID [9-9224000032592] Google). They suggested paying $411 for a repair when second-hand phones were selling for $300.

My mom was planning on trading it in when the Pixel 6 was released, but it failed about a month before and as it wasn't working Google also wouldn't accept it towards an upgrade.

7-year software upgrades are great and all, but by not standing behind their hardware I really can't recommend buying them.

If you still use it, my best suggestion is to use a github search for 'path:.nix TERMs' when you run in to issues. It's likely someone already has that custom package, or an overlay to fix that missing compile option, or whatever else it is that you're looking for. I find reading other peoples solutions also helps with understanding and solving related problems.

I'm in no means claiming Nix is perfect, but coming from other distros to NixOS I really appreciate the ability to reboot to a new generation and then reboot back if I run in to issues. To do the same in Arch I had a bunch of 'snapper plus manual steps' to accomplish the same.

> You don't own the data, just a temporary right to use it.

Agreed, bonkers. You didn't 'own the data' when it was a VCR recording a broadcast TV show or movie, but now it's all "but they broke our ROT13-super-crypto, throw the book at 'em".

If you already have a copy of the original repo, add this as another remote with:

``` git remote add irfanhakim-as https://github.com/irfanhakim-as/yuzu-mainline git fetch irfanhakim-as 537296095ab24eddcb196b5ef98004f91de9c8c2 ```

Then if you want the commit to look like the latest one for the original repo edit the file `.git/packed-refs`. Mine now looks like:

``` # pack-refs with: peeled fully-peeled sorted 537296095ab24eddcb196b5ef98004f91de9c8c2 refs/remotes/origin/master ```

(Be careful with raw editing git files. There might be a more 'official' way to do it, but this worked for me)

> I know electricity production is not renewable everywhere in the world yet, but at least it’s on a path and possible

And somewhat counter-intuitively, even if you are going to consume that natural gas, it still works out better for the power plant to use it to generate electricity that's used to run a heat pump than to burn it directly for heat.

> here isn't any heat pump out there that will produce water hot enough (e.g. 185F)

For 185:

https://www.arcticheatpumps.com/high-temperature-heat-pump.h...

Or much more common, if you can deal with 176F, the SANCO2 ones will generate that down to -20F.

The hydronic temperatures you're talking about are only required if you have to stick with the existing radiators. They make radiators with little fans that work at lower temperatures, or larger panel radiators. There's lots of options for lower temperature forced hot water.

I'd much prefer companies actually follow NIST and stop using SMS for any type of authentication.

eSIM has the advantage of allowing me to switch to cheaper services without paying $10 and waiting for a physical card to arrive. Is it's security crap? Well so is the security of my mobile provider's kiosk minimum-wager workers.

> git his one linear history > Don't make me pick!

I'm not understanding what you mean... to me it doesn't.

For example you can browse by 'chapters' by adding `--first-parent` to your `git log`. You have commits, which are paragraphs, then you have merges (ie chapters), and you can view from either view, or many others.

I couldn't agree more. I've worked on multiple projects where PRs must be 'squash-merged' and I've never understood why people want to throw away history and make their bisects harder.

If you use proper merges you can always see a 'squash-merge' view of the repo. If you squash you can never see an individual commit view.

Also if you're required to tie changes to some ticket system—especially if you have to put that ticket info in the subject—then with merges you can put that ticket info just in the merge commit and the rest of the commits can have proper subjects with actual useful information in them.

It's not a new one, but also good to get the shingles vaccine if you're over 50. Even under 50 might be a good idea as cases—although still uncommon—are increasing in younger people[1]. (Unfortunately where I live they won't let you get it if you're under 50, even if you throw money at them).

From what I read it's thought that kids being vaccinated (a good thing) have reduced adult exposure to the virus, decreasing the immunity we received from being infected as children.

[1]: https://www.cbc.ca/news/canada/nova-scotia/shingles-under-ag...

> They’ll

What if I asked you to list all our source material that led you to use that particular contraction. Heuristics will not do, you must list each.

Can you do it? Do you believe AI should.

> I agree that it should be possible to implement

Those exact words appear in another forum post from 2006:

https://discourse.igniterealtime.org/t/cm-3beta-compression-...

Should you have quoted that as a source for your reply? What if we knew you'd read that post back in 2006, affecting your neurons, then should you?

It might not be too hard to imagine a simple case of a specific topic where you might have some more prominent sources, but even in those cases I believe if you think it through you'll find there was a ton of other sources that led to the weights that allowed you to 'know' the topic.

> I think generative AI should be able to provide links to similar source material in the training data

Except these aren't databases, so that's generally not possible, in the same way that it's not possible for your provide links to the source material it took to write your reply. How much learning led to the weights on your neurons that allowed you to generate that? Where did you learn about using italics and it's effect on how the words would be interpreted? Where did you learn the tone that would be appropriate in this particular forum?

> People should be able to opt out of having their content used for training

Okay... but then, if I write a book should I be able to opt out of you being allowed to read it? What conditions should I be able to put on who can read my work? Religion? Skin colour? People that aren't good at memorizing?

Hopefully the idea of putting limits on who can acquire knowledge sounds absurd to you. Why are those same limits okay if they're on 'what' rather than 'who'?

> AI companies are just trying to avoid lawsuits by keeping it secret

Which has created a barrier to further research. Instead of me and Joe being able to collaborate on research and papers using the same datasets, we now hide our training data lest the luddites come to smash the machines because learning is only okay if not done too well.

> TOTP + password manager is more convenient (and more online) than yubikey, in my opinion. Especially convenient with a browser plugin.

More convenient than yubikey webauthn, or just yubikey TOTP? Personally I find tapping my yubikey much easier than any 'enter the code' 2FA.

Also it should be noted that storing your TOTP secret in the same place as your password means you're one malware attack away from losing access to that account. While this risk may be acceptable for many accounts, it should at least be considered.

> due to security and version compatibility, I feel maybe it shouldn't exist anymore.

I've also thought similar. From a security standpoint though it seems like shared libraries are largely a win. Imagine you're running a bunch of services and a vulnerability is found in OpenSSL. With shared libraries you can update one package to be pretty sure they're fixed.