Hmm, you said the new guy was consulting with the senior devs all the time..

What about the time lost by the devs having to babysit him?

Also consider that the guy is unlikely to improve mich using this technique. About 10 years down the line when It's his time to be the senior and babysit another AI kiddy might not be able to

I think there's a hidden tech and knowledge debt associated with the AI rush but it will only become visible years down the line, when It's likely too late for a whole generation of devs

None of your two examples are bugs though. It's just features that Firefox doesn't have. (Sounds like it at least)

What Bugs did you actually encounter?

Circumsition without a medical requirement is mutilation. You take a perfectly healthy creature and decide that you know better and start cutting parts off. That's not the same thing as "culture". No-circumsition is the default value and some "cultures" and religions simply decided to do something extra, which means the burden of proof is on them to show that It's necessary, which is not Something you should leave at "Well, we just like it this way" because that's how you get oppression, slavery and many other bad ideas like blood sacrifices. We're smart enough to learn and evolve beyond our mistakes and beliefs and we should celebrate that rather than defending "the old ways" under the guise of "culture". At the very least people should have the right to their bodily autonomy and they should be old enough and presented with sufficient data to decide on their own what they want to do with their bodies

You tried Open WebUI? It's a good self hosted solution

Tl;dr; People include that in their prompts because they have no idea what they are doing.

Long version:
It doesn't do anything other than filter statemets of people on drugs where the word "hallucinate" apprared in the context, probably filters out some drug usage and names along that, and also sonce AI now eats itself, it probably also filters some previously bad answers that people in the dataset called halucinations. LLMs are connectionists, they operate on keywords and concepts, but they can't do anything that's not part of the dataset. So since the concept of "being wrong when you don't know any better" is impossible to really "learn" it means they can never "learn not to halucinate"... Researchers might include it in training prompts to do more reasoning tokens first, but that has limited effect since the llm already prompted and convinced itself with It's past reasoning tokens.

The point behind the claim is that failures tend to cascade. So a RAID setup, even RAID 1 setup offer less protection than a secondary raid, or a separate disk you connect and disconnect just for backups, or the best solution: offsite, as you mentioned.

That's why we often say that RAID provides redundancy, but is not a backup. For example if your RAID driver fails or burns might kill both drives... Or if your bedroom caught fire, you might lose the entire raid and all drives, but not the disk you keep in your livingroom... Etc...

So there's a method to the madness

People will eventually come to the same conclusion and stop buying those, which means they will stop being made. At least I hope so :D

Phew, that was close. I almost used react!

Thanks for the heads up!

@kiki Last one I looked closer at was the source for moosefs since I was setting up my own distributed file server and it had some issues so I took part of it apart to check how it looks, fixed and reported a bug I found and am still kinda looking at optimizing it for my setup specifically. That was 4-5 Weeks ago? Still on/off looking at it

Also looking at the Catima app for card storage, that Im just starting

I also took apart and recompiled ollama a couple months ago since I had some optimization concerns and disliked they left debug logs on then. Took that opportunity to also check it doesn't connnect to anything weird then.

Those are the most recent and biggest ones. I don't remember many of the small projects and libs that I just skim to see of they are what Im looking for :D

I think this is also on topic: I don't do it all the time, but I also sometimes just pull the projects and search for some basic stuff like "http", or language dependant classes loke URL or Requests when I know the project is supposed to be offline... Stuff like that. You don't need to read everything all the time, because usually the Numbers of releases, the active issues and the history of the project can tell you enough about safety. And then if It's a smaller starting project then a rudimentary keyword check can tell you all the important stuff.

It's not 100% safe, but it's easy to do if in doubt.

I host FOSS stuff on my own server at my own home, and in some cases I modify it to do some extra stuff I need from it.

I can't do that with software hosted by companies, and I do end up reading the code when I modify it.

So I suppose this sort of software is a really good fit for me, my skillset and my time allocation. It's obviously not for everyone, but It's good It's an option, I personally appreciate it a ton.

I usually just wonder how many people already figured it out, but never put it anywhere, so the solution sits in someones abandoned projects folder already. Just like it will now sit in mine!

Why are we using an autocomplete to interview for positions? Are they all insane? Using a language model to select people is like using a facial recognition model to select people. Sure, the fact the object in the picture has a face is a good indicator they are a smart human and not an apple, but you can't be sure of false positives nor false negatives with any degree of certainty. Similarly, just because someone can type doesn't mean they are smart.

We're loosing touch with reality -_-

What, you're not supposed to use AI to generate logos or text. Just generate the Pics/Backgrounds and do the rest in photoshop... Take Screens of the App and all that...

It's like a 5 minute job at this point, what is this person even paid for?!

What do you mean? MKL is being used for AI, just not by default because CUDA is king still and MKL barely makes a difference on most consumer chips

Whats wrong with soap? It's a bit dated and verbose but it works fine. I mean all of xml is pretty verbose anyway

My only positive support story was when I bought new Jabra wireless headphones and they were broken. Wrote to support, they asked a few questions and then told me they will send me new ones and I can keep the old ones.

Really easy, well communicated, no obstruction support...

I even mamaged to fix the first ones myself (it had a damaged PCB, fixed via a jumper wire, worked ever since) so I essentially got one free pair. But this service alone means I will consider buying again from them once mine break

I use the most powerful computer known to man, the human mind ._.

Stuff like this is why the tech and package managers for frontend are all memes. The node folder, the randomly broken deps, the new version every week, the new tech-stack every month...

Id argue parenting is not a job though, so it doesn't belong on the list. It's not a paid nor assigned position. I mean baby-sitting is a job, but parenting? It's very 21st century to think of it as a job or a chore to do...

I'm pretty sure the idea behind "readable code" is

a) adhering to common language patterns and naming schemes

b) Proper variable and function naming

c) Avoiding god functions that do everything all the time

d) comment whatever cannot be easily understood otherwise, mostly useful for optimization magic and parallelism patterns where the flow is no longer easily understandable

I don't think these are "subjective"... Clearly you can name a function something else than I would, but as long as you name it correctly, and it's short enough so that it's easy to understand, it's going to be readable by me even if your choice of words is difference.

ditto with language patterns. If everyone is using camelCase, don't randomly write with snake case. My eyes will be searching for specific patterns that are common for the language world wide, don't be *that one person* who likes his way better because that's how he does it in TyPEsCrIPt or something.

and so on

Yep, these days SW RAID is deffinitely Worth it.

Are babies good or bad?

how would a person by "inherently" good or bad? At what age does one become a "person" so you can judge them by this criteria?

Due to this, my answer has to be no. I think nature may play a part, there are plenty of personality disorders to choose from that can compel someone to do something "bad". But then can you really blame them if it's caused by a disorder?

But I more incline that people are skewed by nurture. Bad childhood, Bad friends, Bad colleagues or just Bad experiences with other people may make a villain. There's a notion of the "cycle of abuse" for example. So was such a person "inherently bad" when it was done to them? I don't think so. You are obviously responsible for your action, and how you deal with events that happen to you, as such someone that turns a villain my be blamed for weak psyche, but I don't think I can call them inherently bad, more like "bad right now, in this moment"

And then there's subjectivity like Lens mentioned

@jestdotty You're saying what we're all thinking!

Being a garbage man is an important job. Also teachers and medical staff.

Your job is easier than the job or a nurse, yet they are not paid nowhere near as much as software engineer or a game developer on average.

So I don't care about how easy the job is, It's not a relevant metric, the real metric is how much value you bring in. And now we can start talking about how not all CEOs actually bring better value and definitely not better than your overworked medical staff.

Please, pay the people that matter, not the people that we assume must matter

@cuddlyogre for Local LLMs I do recommend DeepSeek R1, at low sizes like 7B it can get good enough speed and quality in my experience. And it has a good selection of sizes that you can experiment with. I used it to help me write a couple of game design docs. It's never perfect, but you can get close by refining your initial prompt instead of chatting, that just polutes the context imo, I prefer editing and re-generating the initial prompt until I get roughly what I want and then touch it up manualy.

Still, Something like Obsidian is gonna give you the most control and you likely want to use it even if you also use an llm

Regarding speeds, if you're in normal consumer HW and need to squeeze a bit more, you can try looking at Llamafile. It's an optimized way to run a single llm from a single executable and it can give you some extra tokens/s.

That's just my 2 cents :D

I actually set up my own mail just last month. It's awesome cause I can get as many throw away emails as I want and no website has my domain blacklisted so they can't fight back... Good for those pesky Times when Websites block or don't allow +suffixes on standard gmail

That is actually really cool. I like resource pooling. I may donate some at some point.

The model list is currently empty though, broken or just offline?

The only issue security wise is that the host can read the conpletion Traffic. Especially since ollama is open source you can hack your own client to even log all the messages to disk if you want. I built my own ollama client before, It's a well put together project so anyone can do it. :D but for things you don't care about It's always a good idea

@jestdotty Europe is bigger then the US though. Not by much, but it is ._.

@Demolishun they can, but as with all things AI, It's not perfect and It's an endless game of cat and Mouse.

My hope is people will value human art more than before, we just need to give it time. All the AI slop looks the same after a while, AI can't make new things, it can only mash old ones. So truly unique pieces will be valuable.

Where AI will dominate imo is Marketing. Worst art job anyway