Kerfuffle

Is anyone using these small models for anything? I feel like an LLM snob but I don't feel motivation to even look at anything less than 70-40B when it's possible to use those models.

Do you value your time?

Sure, generally speaking yes. Who wouldn't say that?

Here's your problem though: the process of logic where you look at someone's actions and conclude that they wouldn't have performed the action unless it was valuable to them only works if:

1. Everyone is a perfectly rational agent that only chooses to perform actions that they believe will benefit them.
2. Everyone is capable of perfectly predicting whether an action will benefit them overall.

Obviously neither of those things are true. Even if #1 was true, your approach still would run into problems because you wouldn't know if a particular action was a misprediction (that actually didn't end up providing value to that person) or whether it truly was beneficial. Since humans are both often quite irrational and pretty bad at predicting effects to boot, well... back to the drawing board.

Giving something your attention not only implies it has value to you

This is what we're talking about: value to me specifically.

If something has no benefit to me, I can see that doing it is a personal flaw but I still can't resist the compulsion (could be anything: time wasting reddits, crystal meth, alcohol) then it seems really weird to conclude that I "value" it. I'm pretty sure most people don't use the word "value" like that: it's for something that has a benefit to the person. In this case, the "value" is negative if anything.

Don’t kid yourself. Reddit was valuable to pretty much all of you or else you wouldn’t continuously shitpost about it.

It doesn't really work that way. Lots of people visited subreddits like cringetopia, whatcouldgowrong, confidentlyincorrect, etc. Basically a compulsion to kill time by gawking at dumb people doing dumb stuff. That doesn't mean those subreddits or that use of time had actual value, even to that specific person though.

That seems like they left debugging code enabled/accessible.

No, this is actually a completely different type of problem. LLMs also aren't code, and they aren't manually configured/set up/written by humans. In fact, we kind of don't really know what's going on internally when performing inference with an LLM.

The actual software side of it is more like a video player that "plays" the LLM.

By "attack" they mean "jailbreak". It's also nothing like a buffer overflow.

The article is interesting though and the approach to generating these jailbreak prompts is creative. It looks a bit similar to the unspeakable tokens thing: https://www.vice.com/en/article/epzyva/ai-chatgpt-tokens-words-break-reddit

The air fryer definitely should work (make sure the basket is too small for beans to slip through). I haven't used that method myself so I can't really give you advice, and I'd guess it's reasonably noisy so it might be a bit difficult to hear when the cracks occur (especially the second one which is fairly quiet).

If the grill can get hot enough, that should also work fine. You can even use a frying pan and stove top, so really anything that gets the beans to the correct temperature and ideally can stir them for even roasting is going to work — though the times and such might vary. I've only used the oven so far.

You should be able to web search for more specific instructions for any of those methods. If you can't find what you need, let me know and I'll try to search as well.

It’s awesome if you want to do your own as a hobby, but my time doing that every week is going to cost more than buying a pre-roasted bag.

If you think that way then why are you preparing meals every day to eat instead of just buying them pre-made? Do you consider preparing those meals your hobby?

I could just as well have created a post here aimed at people who had never made their own food like "You should know it's actually really easy to prepare some simple meals like stir fries or pasta". Who would say "My hobby is knowing how to cook pasta and throw some sauce on it to make a meal" though? Or say to the person "It's awesome if you want to cook pasta and heat up sauce to serve on it as a hobby, but I'm not into it myself". Guess it seems weird to me because you're talking about it like it's a defining characteristic that a person would be pretty invested in rather than an approach to dealing with something that could have some advantages. (Arguing with random people on the internet though? I think that's something we can safely call one of my hobbies.)

Also, yeah I’m in Canada. So I’m sure there are online options

100%. If you wanted them, you'd have no problem.

time and equipment are huge factors.

I mean, the equipment is an oven, a cookie sheet and a heat resistant bowl or two. Nothing exotic.

The time is about what it takes to make a simple pasta dish or stir fry. It seems weird to me stressing about the time when it's about the same as preparing a simple meal and only has to be done once every week or so.

Obviously not trying to pressure you to roast your own coffee if you don't want to, it's 100% up to you. I guess I feel like you have to be greatly overestimating the time and effort to say something like that.

I remember the inventor of the Behmor machine saying that the smoke suppression was required by regulation of some sort…

They say technically correct is the best kind of correct!

A coffee roasting machine has to be able to handle basically whatever people do with it: dark roasts, light roasts, small enclosed spaces, whatever.

Like I said, do whatever you like, but I advise against roasting coffee indoors without an exhaust fan that dumps the exhaust outside, at minimum.

I just haven't seen enough smoke to be concerned. Maybe my experience is abnormal, I don't know. It's not even like the oven was full of super dense smoke or anything, but keep in mind the only smoke that's really going to come out is what drifts out naturally for the < 10 minutes the beans are roasting and the 10 seconds the oven door is open.

Now if you're roasting well past the second crack then certainly that might change things but I usually prefer pretty light roasts. I took my current batch out right at the beginning of the first crack. There was virtually no smoke at all. Of course, you too should do what you like: though I'd respectfully suggest that maybe you shouldn't be super confident about the effects if you haven't actually tried this approach yourself.

The problem with roasting your own coffee is I have a life in a city in the Western hemisphere and nowhere around me sells raw coffee beans.

Does "Western hemisphere" mean the US? If so, it's no problem getting them. Just for example, this is what I have currently: https://www.amazon.com/gp/product/B01GGMF8RA/ (not a referral link or anything)

https://www.freshroastedcoffee.com/ also sells green coffee beans. Those are ones I've bought from previously without issues. If you live in a different country then it might be harder but it's almost certain that there will be online stores in your country that sell green coffee beans. You might even be able to find local stuff.

Because coffee beans have a flat side, this method will produce an uneven roast.

Interestingly, when I was researching this I saw some people say they actually referred that for the variety or making the coffee taste more interesting (broader range of flavors getting combined). Not sure I'm enough of a gourmet to really tell the difference, or have a problem with it even if I could. I'd guess it's about the same as just mixing two different roast levels unless you do something that actually hurts the quality like burning some beans.

The way I was taught was to use a pan and to stir consistently.

I might try it eventually. I'm super lazy, so just tossing them into the oven and leaving them there is a lot less work than standing at the stove stirring the beans.