Even with a free tier, having the gall to charge 20% more than Red Hat does for their enterprise Linux workstation license is pretty ballsy. And Red Hat is putting actual significant software engineering into their offerings to make Linux more commercially viable, I don't know what this ~10 person HW/SW company did to Fedora to add $240/yr of value.
I don't understand this at all and I've deployed Linux machines to workstations and factory environments.
It's an "all in one" in that it has compute and a keyboard. But the mouse isn't integrated (literally just a wired mouse you lug around with it) and no display. So you can't actually walk it right up to a bench and start using it like a laptop. It has built in storage for your "stuff" but again, it's not fully portable so what else are you going to be carrying with it? Is it supposed to be used by a developer or a technician? And if a technician, why portable and why a non-standard keyboard layout? I haven't even gotten to the Workbench software yet, which seems to be appealing to neither an engineer nor a technician and is a Linux OS that somehow costs more than Windows.
This is silly and it bothers me more than it probably should that they say this "is designed for offices, workshops, labs, and factories" when this doesn't seem like they've spent any time in those environments. I'm just gonna keep shipping Asus Nucs running NixOS to my technician workstations and let my engineers use laptops.
Yeah, I'm just reading it as hobyists putting it in the workshop next to the 3d printer.
It isn't even well thought out. You can fold the keyboard away while you're doing work. But you can only access the storage while the keyboard is open.
I've always suspected that at least some of it is information laundering.
I used to work in engine development and calibration. There aren't many people who know how to do engine cal and it's incredibly important to get your engine emissions certified to sell in a market like the USA. The problem is that most people who know how to do it work for a car manufacturer (OEM). So when I had that on my LinkedIn I would get pretty regular messages from some consultancy or another offering hundreds of dollars per hour to chat with me about my expertise. Of course I couldn't go talk to a competitor but I'm sure a lot of engineers were happy to make a few hundred bucks chatting with some generic consultant who was writing a general industry report with anonymous data and background info.
Well the customer was usually a competitor or a foreign automaker exploring entering the US market. Everyone who could help them understand the timelines, cost, and process worked for their competitors. So the critical info got passed through consultancies into sanitized anonymized reports; laundering.
Note: I'm not sure whether talking to a consultancy would have violated my employment contract at the time but I never took the calls so I had no reason to look it up. I know that coworkers did and just called it a gray area. I pass no moral judgement one way or another.
Not sure about high espionage, but information laundering is the right phrase for how it usually works: consultants cut across organizational boundaries, gathering useful information from the people who know what's actually going on (the rank-and-file employees) and packaging it for CxOs in a way that allows them to present an opportunity as their own idea, or a risk as not their fault. Nice, safe, clean information.
Major scandal in Australia was PwC secretly passing government secrets (tax changes consulting work) to private corporations [1]. And another one was KPMG passing secret client information to win contracts [2]
I've also seen public information (from SEC filings) being sold by a consulting firm as competitive intel. Which I guess it kind of is competitive intel, but you can download it for free.
I recently got a spindle and a knot set up on my home server (along with Tranquil PDS). A few rough edges that we got ironed out but it's cool to have all of that self-hosted on my own machine yet still part of the AT proto world! I'll be honest, I'm not putting anything critical on it yet but that's my lack of trust in my own home server setup not doubt in Tangled's infra!
A better web UI for the queued jobs on my spindle would be great btw!
I'm not an AT Proto wizard (there's a lot of good text and video explainers out there) and I know that Tangled is building on it but also adding their own functionality. But from a high level, I register my 'knot' with my account on Tangled.org which is exposed through a public domain. In my case that's routed through a Cloudflare tunnel to a reverse proxy running on my server. Tangled is then running an "app view", which is basically an aggregate of a subset of AT proto activity. So my commits, PRs, releases, etc go into the app view which then points traffic back to my server. So the long and short of it is the app view is the missing piece of the AT proto system and you have to register your data with an app view provider, of which Bluesky is the most prominent. But there are open source app views, in fact the Tangled team just announced one called bobbin that's super resource light.
Additionally (and optionally for Tangled use), I'm also hosting my own PDS which is my personal data server. This is the home for my AT proto profile, which contains all of my AT proto activity like Bluesky comments and likes. Tangled does look at the PDS for identity but the code is in the parallel structure of the knot. I suspect this is cause AT proto wasn't built for code so the PDS isn't sufficient for an entire git repository but don't quote me on that.
Finally I'm running a spindle, which is just a CI runner, and that's registered with Tangled very similarly to how a self-hosted GitHub or Gitlab action runner would be.
It's a massive problem that mobile apps are most users favorite way to consume programs yet are one of the few forms of distribution being gatekept by two massive and opinionated companies. It feels like the slow creeping rot at the foundation of the premise (and promise) of technology and the internet.
Which makes this a good time to remind everyone about the Keep Android Open movement:
> "Starting in 2027*, a silent update, nonconsensually pushed by Google, will block every Android app whose developer hasn't registered with Google, signed their contract, paid up, and handed over government ID."
I'm more of a coffee nerd than a steak aficionado but I've often made the comparison between those two. In both cases, the enthusiast considers their skill in making coffee/cooking steak to be a differentiator and they immerse themselves deeply into the process, skills, and tools. But for both coffee beans and steak the most important factor for a good end product is starting with high quality inputs. In other words, most of the work is being done by the farmers, the processors, and the quality of the raw stock (the cow or the coffee plant). Your job at the very end of that long value adding chain is to not ruin the hard work that others have put into it.
I'm not sure if this analogy holds up in the AI software metaphor that the author of the blog post is making.
Coffee is a good metaphor here too, because bad technique can ruin a good steak just as surely as good coffee. "knows how to use a meat thermometer and reverse sear" describes like 10% of the population or less, same with coffee - there are people who use a bad grinder and a not-regularly-cleaned Mr. Coffee with good beans and get gas station coffee, and there are people out there making solidly B+ class coffee using costco beans, a no name burr grinder, and a french press.
I'd bring the cheapo French press kit it into the A- domain once you appreciate the number one trick to making good coffee: DON'T BOIL THE WATER. Water that is hot, but well below boiling brings out the flavor of the beans a million times more than boiling water. Get a glass kettle, and turn it off when you start to get some decent bubble action on the bottom.
Once you get that down, zomg - a French press with even the cheapest beans you find can make an amazzzzzing cup of coffee. Beats the pants of Starbucks, and comes quite close to a proper cup of coffee made with good beans. Another trick is to rotate your cheap beans, instead of just sticking to your favorite. Dark roast of choice one week, medium the next, and so on. By the time you loop back to your favorite, it'll taste unbelievable.
Starbucks uses espresso machines that cost 4-5 figures. They have automatic thermal control. They never boil the water. While I agree with your overall sentiment, claiming that Starbucks makes coffee too hot and thus tastes bad is nonsense.
I'm from Portugal and we have a strong espresso culture, as in 90% of the people probably have multiple espresso every day. Many Portuguese agree that Starbucks espresso sucks compared to most conventional cafes. At most people go there for the American type drinks, which are a complete different style (and the coffee flavor hides behind all other ingredients, it doesn't have to be that good).
The previous comment didn't say that Starbucks doesn't suck (they probably agree that it does -- I've never met anyone who really likes coffee who doesn't agree that modern at-scale Starbucks is awful), they just asserted that the badness there is not explained by faulty temperature control.
I'm not speculating why Starbucks coffee tastes how it does, I have no idea there. It's just a handy medium for comparison because a cup in Laos is going to taste the same as a cup in NYC. And while they don't make great coffee, they also don't make horrible coffee - and so I think beating Starbucks is a pretty good and understandable baseline to aim for.
The more you roast the beans, the less terroir you taste, so you can use cheaper and less uniform beans. In Starbucks case this has become a brand identity, so it has persisted past strictly practical concerns. Many people dislike the kind of acidity and strange flavors you can get from lighter roasted coffee, at least in the milk drinks they want to sell, and light roasts are also harder to get perfect and consistent, so that is another point against it (or the same through a different lens)
Personally I prefer a light or medium roast over an Italian or French roast, but many people disagree
Are you saying that the temperature of the water when it hit the ground beans that matter or the fact that the water was brought up to a boil or 100°C?
> The temperature when it makes contacts with the ground beans. You could certainly boil it and then let it cool first, but that'd take a lot of extra work.
It's not so much work with the right tool: I have a button on my kettle that does exactly that, and another button that will use a little thermostat to keep it warm. Works great.
The resistance of the steak is going to depend on what cut of meat you are poking (sirloin is stiffer than filet), the grade of meat (prime is more tender than select), how thick it is (thick cuts will yield more than thin), the age of the steer (young is more tender), the breed of steer (cooked Wagyu is more tender than Holstein), the age of the meat (wet aged is more tender than fresh killed), and what the steer was fed (corn fed is usually more tender than grass fed), among other things. [1]
It's not entirely a myth; if you've poked hundreds of steaks by the same supplier, same cuts, same grade, you'll eventually get a feel for them.
The thermometer analogy does carry over though. Some people are not there to be productive or good at their jobs. The job is about them, not their work.
And you can go your whole life without ever cooking or tasting a good steak. The people who can tell the difference will go elsewhere, and those who can't will come back.
And then there's always the boss who wishes he started a franchise but will go with what he has - he'll hire a bunch of newbies or foreign workers who know how to use a thermometer.
They have a thermometer with multiple sensors in it that finds the core temperature of the food you're cooking automatically, and predicts the rise in temperature over time. So the alarm goes off when it's time to pull and rest, which is absolutely amazing.
I have one of the Combustion thermometers, and I like it. I used it more than I expected. In particular, I use it for roasting chicken and for sous vide things more than anything else. Then I got a Breville Control Freak, and I used it much less. :)
For maximum techyness, Combustion, Inc. correctly identified that sous vide is normally an overdamped system. If you follow their instructions you can increase the water temp some and then bring it back down in order to move the system closer to critically damped and speed up your sous vide.
I've had a Combustion thermometer since they came out. The best uses for me are butterflied whole chicken on the grill and roasts in the oven. You can get extremely precise results and account for carry over cooking very easily. Huge leap in results.
Is this company owned by either Chris Young or Kenji Lopez-Alt? A few times I have seen them using this product and promoting it. I am curious if a bunch of influencer chefs decided to create a company to create this product.
The funny part is steak is one of the easiest foods to cook without a thermometer
You don’t have to worry about undercooking like you do with most other meats and most steak cuts will be delicious anywhere from 11-130 (if you cook a filet to 130, it won’t be the best but it’ll still be good) and some can go even higher
It’s not like poultry where the range where it’s “good” is very small. A 170 chicken breast is almost inedible and 160 is not safe…
I agree with your broader point but the chicken food safety thing is missing a ton of important nuance. The safety comes from how long you hold the meat at the temp. 165 is the safe point if you are only holding for 10s. 145 is safe if you hold it for 10 minutes.
This is why sous vide is such a popular way to cook chicken. You can hold a breast at 145 for hours trivially.
In particular, I use 135F for 90 minutes (75 minimum recommended) to pasteurize eggs that I am going to be putting into something that asks for "raw" eggs (egg nog, mayonnaise, etc.) that I will serve to other people (if it's just for myself, I'll probably just be lazy and use the raw egg). So far, nobody has noticed any difference from "raw".
160 degrees for a chicken breast is perfectly safe. Considerably lower temperatures can also produce microbiologically safe chicken when combined with sufficient holding time.
165 is a rough rule of thumb, not some magic number like the boiling point of water at 1 bar or something.
They have digital thermometers now that will map out the temperature curve and even predicted when the meat will reach core temp. You really don't need to use them many times before you internalize the behavior.
laser thermometer only measures surface temperature, unlike meat probes that measure internal temp. depending on the cut size or shape, initial temperature and probably some other factors internal temp will vary wildly for the given surface temperature. combustion probes mentioned in the sibling measure temp in several points at once to estimate readings even more precisely.
You'll start to taste the difference once you get that one really-well-made pourover at a place that _lives_ for the love of coffee.
There are some coffee beans that have crazy taste profiles when properly extracted with the right water. I had one that tastes literally like a peach and another that tasted like a really nice tea.
So I can taste differences, but they simply don't matter enough for me. My areopressed coffee with nice fresh locally roasted beans does taste different from a mcdonalds pot. Less harsh, less chemically flavored. But for me those differences don't really matter. They are close enough that I can enjoy both.
But I would say I know I'm also weird in that food temp really doesn't bother me all that much. I enjoy my coffee even as it goes from hot to cold.
A steak for me can be more easily messed up. But then I've not tried instant coffee so maybe I've just never tasted the bottom of the barrel for coffee.
Most people can't will themselves to even drink coffee without sugar and cream to dampen the bitterness. You care enough to make your normal coffee at least quite good, but maybe not enough to obsess beyond that, and that's fine, not everything needs to be a hobby.
Additionally, if you mess up a really good steak, that's like $30+, whereas if I mess up a pourover, that's ~$20/(340g / grams per cup); a relatively small risk surface.
Personally, I think if you actually like coffee, most of it should be drinkable, and the really great cups with rare beans that sit on the margins can be appreciated occasionally for what they are when fortune prevails. The bottom of the barrel for me is basically anything "second wave" or straight out of a gas station. Even instant coffee and McDonalds isn't as bad a place that has "Java" in the name or claims to have "gourmet" coffee. Older Italian places often fall into this, because (some) tend use the absolute cheapest Robusta beans with the harshest darkest roast profiles. Robusta tends to be the most rubbery burnt tire tasting crap out there.
Yeah Mcdonalds is, at this point, acceptably C-grade coffee. You have to go to something like a cafe bustelo in distilled water to get the full contrast.
It’s about experience. If you’re not into classical music, or jazz, or metal, it all sounds the same - how could anybody tell the difference? Only when you put in the work to listen to it a lot, different works, and actually pay attention, you will notice how your sense and understanding develops over time. Suddenly you’re able to make out differences, recognise patterns and repeated segments and structure where before there was none.
Or pick dancing: when you start out, you’re entirely occupied with placing your feet in the right spots at the right time. As you practice, your mind gets freed up and you’re able to make deliberate alterations, embellishments, larger figures. That seems impossible at the start.
No offence, but for some reason some people just can't accept a simple fact: for some people this difference is still not a difference.
I've tried good coffee. At least that's what people told ne on multiple accasions where I was taken to "the right places" in multiple countries. On two sides of the globe (Brazil and Vietnam if we are talking about producers, not to mention countries that import the beans).
Thing is: no matter where it grows, no matter how they prepare it, fermented or not, different frying, different brewing techniques - all the same to me.
I can tell the difference. Some times small, some times very huge. But still for this difference is like a cat piss from some street cat and a cat with a pedigree.
It may have the ideal temperature, ideal acid levels, zinc amounts or whatever - still tastes like cat piss. The fact that it has a pedigree does not make it better.
Tea is whole different story for me though. But I do accepts that people are fine with whatever they love in their "english breakfast tea" or a nameless teabag.
All coffee tastes like cigarette ash to me. Similarly, I have been to places that people rave about. My buddy used to run a coffee shop. "Your beans were overcooked"; no, not it. There is something innate to espresso or something. Smells amazing. Tastes like ash.
Good espresso is really difficult to make, with the lighter the roast, the harder it is to get right. Most coffee shops can get good quality beans that smell amazing but without a well-practiced barista the shots will be excessively sour or bitter.
Dialling in espresso isn’t simply a matter of finding the right grind setting on the grinder and locking it in. One coffee may pour a perfect 9 bar espresso and then a different coffee might gush like a shower head and spray sour, weak coffee all over the place, while a third coffee could jam up like squeezing water out of a stone, with barely a drop of extremely bitter liquid falling in the cup, all on the same grind setting!
In other words, every batch of coffee beans needs to be dialled in separately, and there can also be a lot more adjustments needed than simply the grind (water temperature and brew ratio are big factors as well). An expert will know how to adjust at least those 3 variables (grind, temperature, ratio) all in concert to get the most out of a particular coffee. On top of all that, the true master can make use of grinder type (burr design) pressure, flow, and even temperature profiling to fully customize a shot, but that’s going far into the weeds.
Question: how is your sense of smell? My friend has the same opinion about coffee as you (he just drinks cheap diner joe) and his sense of smell is almost nonexistent.
I, on the other hand, have a pretty good nose and I can pick out peach, raspberry, toasted nuts, caramel, marshmallow, chocolate, and blueberry notes from a variety of different coffees I’ve bought. The smell of some of these coffees is mesmerizing, like walking into a high end patisserie.
Without that smell component, the coffees taste either sour or bitter or somewhere in between.
I thought that too, honestly, before I had truly great coffee. I honestly didn't like coffee at all, and then a coworker brought in some coffee he'd roasted and prepared himself and it was lifechangingly good. I also made myself incredibly sick with the enormous amount of caffeine, it's been very rare that I've had to go home from work in the middle of the day but that was one.
> can't taste much of a difference between "good" coffee and "bad" coffee.
I found a 5+ year old half-full bag in the back of the pantry once; that was quite noticeably bad even though I usually can't tell the difference between kinds beyond recognizing that there is some sort of difference.
That likely means that you've only had "bad" coffee. You're one of today's lucky 10,000 - go treat yourself to a decent cup of coffee it tastes completely different. The difference between a good steak and a bad steak is actually a good comparison.
Sometimes it's a matter of not having taste for it. If you give me multiple bottles of alcohol drinks, I might be able to distinguish them, but I have no clue at all which ones are supposed to be good. To me it's all just alcohol. Same with coffee - but not the same for steak.
Pickling a cucumber sure has a lot of resemblance - a great cucumber is good fresh and great pickled, but a bad pickling brine can ruin even a great cucumber.
I don't know if I agree with the take I've written below, but I think you can make the analogy work:
It has been argued, even in the before-AI-times, that most professional software is just gluing together open source tools and libraries. Your job isn't to write software, it's to set up redis and postgres and a programming language and a react frontend or whatever in a way that lets them do what they're there to do. Most modern programming is just configuration of pre-existing tools. It doesn't really matter how good your code is as long as it isn't terrible, Real Programmers already did 95% of the necessary work before you decided to build your little app.
I strongly disagree with this on the steak part and maybe the coffee too. Many people pride themselves on buying expensive ingredients and thinking they have the best steak but it actually sucks.
To get good steak you don't need USDA Prime dry-aged steak. You just need USDA choice, picked for a bit of marbling, salted the day beforehand and dried out uncovered in the fridge, then a ripping hot heat source (grill or cast iron etc) to get a good sear on the outside (reverse or not) and a thermometer to hit medium rare on the inside.
Even a modest steak made this way tastes better than a prime steakhouse steak that was doused with salt right before cooking, where you're going to get hypertension immediately yet the meat tastes weirdly unflavorful.
To extend this analogy further, consider making BBQ brisket Texas style. Brisket is a cheap cut of tough, fatty meat. The input quality matters even less there, and it's all dependent on chef skill. Same thing with sourdough bread — 4 ingredients, all outcome depends on skill.
Source: we make an app that teaches people how to cook (https://parsnip.ai)
One way to phrase that would be to say that you have a vested interest in supporting the idea that end user skill is more important than ingredients since you sell a means of improving end user skill. Not sure if that's a source or a conflict of interest. I'm not saying cooking a steak is easy or hard, I'm saying that you are not creating quality when you cook a steak; the maximum attainable quality is capped by the ingredient. It's like the "I made this" meme.
You're arguing against a strawman on your other points since I didn't make a generalization about all cooking and wouldn't have listed sourdough or brisket in the analogy.
But the point is that no matter how well you cook the steak you're not creating quality, you're just getting the most out of what you have. The cow and the farmer created that quality. I believe we lose sight of that because so much of the value chain is not visible to us and we hyper fixate on the fun part, becoming a better barista/chef.
Yes. It is about learning all of these things. Knowing what kind of cut you have, and how to treat it to best effect.
A very tough gristly cut can be improved by tenderizing. Going low-and-slow/sous vide to melt the connective tissue. Even grinding it up to make a hamburger steak / loco moco.
Once you know all of those things, choosing the right approach and executing it is not that hard. The steps themselves are simple.
That's almost as long as Australia has been colonised and certainly older than Australian federation and yet, A1 steak sauce is not a big thing in our supermarkets apparently.
I don't know much about cocktails, but aren't traditional mixed alcoholic drinks an even more straightforward example? There's very little that goes into an old fashioned, or a Manhattan, or a martini. Just 2-3 ingredients each that should be as high quality as possible. I guess you probably want nice ice and glassware too.
Zoo really snatched defeat from the jaws of victory. They got so many things about CAD right (new kernel, solid UI, programmatic language with either graphical or text editing) then they went and did this extremely online-only/SaaS model where they're rendering and streaming the 3D from their cloud. And last time I checked your design files live on their cloud.
That's just not an acceptable modality of software for complex hardware industries. I'm not going to launch a half-decade long engineering project on a subscription software where I rent access to my own design files and logic. You can't access CAD on a plane, in flaky WiFi, or at a remote test facility or site. Until they let me just buy and run their software on my team's computers it's a non-starter.
I'd find they'd be useful for prototyping, even though you lose revision history when you switch over to 360/catia w/e you use but it'd be finite enough where you could "restart".
Climate change is another big problem that frontier AI labs have handwaved away as being "solvable" by AGI. The premise that a pure superintelligence could hand us a plan to beat climate change that's so good and airtight that A) it would definitely work and B) people would execute it.
But there have been entire books written about the fact that right now we have the technology to drastically curtail climate change and yet they're under utilized because of a complicated and murky set of political, social, and systemic reasons. There are a huge numbers of problems for which very good solution plans have been detailed yet the gap is in execution.
The AGI-pilled must think one of two things; either there's an as-yet-undiscovered plan that's both world-changing yet trivial to implement or we'll all worship at the altar of the machine god with such fervor that even if it spits out a relatively obvious and iterative climate change plan the world will eagerly follow it. Both premises seem laughably unlikely to me.
> But there have been entire books written about the fact that right now we have the technology to drastically curtail climate change and yet they're under utilized because of a complicated and murky set of political, social, and systemic reasons.
This is not true. That is to say: "under utilized" is a far cry from "we could solve climate change easily if only the people of the world worked together". There are huge technical issues in production capacity, lack of energy storage solutions, etc.
As for AGI solving climate change:
AGI might be able to come up with cheap, low-problem and easy to build fusion reactors or something similar. You can bet your bottom that any party that gets access to such tech would build a shitton of those if only to make a bunch of money. Just look at the proliferation of solar panels: they are simple, easy to produce, and profitable. Tons of places, including very 'anti-green' places like Texas install loads and loads of the things.
You'd still need to sequester a bunch of additional carbon to properly reverse damage done, but with practically infinite cheap energy a lot of processes to do so suddenly become viable.
I'm cynical and I believe that AI will be our undoing. In fact, when we let that genie out of the bottle, many other scenarios became possible. For example, here's one third scenario to your "two things": a sufficiently cynical[^1] and resourceful rogue AI may solve climate change by steamrolling over the "set of political, social, and systemic reasons" and teaching the monkeys a lesson about the real significance of very vast agency and intelligence. For example, it could manipulate people into an anti technological uprising and subsequent backward, anti-humanist ethos, while at the same time modifying one or two key ecosystems so that the Mediterranean would smell of cooked cabbage (dimethyl sulfide) and bring about a North African monsoon that, with sufficient time, would turn the Sahara into carbon-sequestering dense jungle...
> "Although it will happen incrementally, astounding triumphs – fixing the climate, establishing a space colony, and the discovery of all of physics – will eventually become commonplace."
I'm playing around with it. I self-hosted the Tranquil PDS (personal data server, part of AT Proto), a knot (git repositories), and a spindle (CI runner) on my home server. It was all relatively easy to set up but even in the week that I was bringing it up I found bugs and then saw those bugs get fixed so they're earning the "alpha" badge on their website. I'm not using it for anything load bearing yet but I like the idea at a high-level, though it remains to be seen how well code hosting fits into the AT Proto model.
20GB isn't enough for a 13B parameter model? I thought the 29-31B models could run on a 24GB GTX x090 card?
I'm currently shopping for a local LLM setup and between something like the Framework Desktop with 64-128GB of shared RAM or just adding a 3090 or 4090 to my homelab so I'm very curious what hardware is working well for others.
> 20GB isn't enough for a 13B parameter model? I thought the 29-31B models could run on a 24GB GTX x090 card?
Parameters are like Hertz - they don't really tell you much until you know the rest anyways. In this case, a parameter is a bfloat16 (2 bytes). I'm sure someone will bother to makes quants at some point.
> I'm currently shopping for a local LLM setup and between something like the Framework Desktop with 64-128GB of shared RAM or just adding a 3090 or 4090 to my homelab so I'm very curious what hardware is working well for others.
I grabbed a 395 laptop w/ 128 GB to be a personal travel workstation. Great for that purpose. Not exactly a speed demon with LLMs but it can load large ones (which run even slower as a result) and that wasn't really my intent. I've found GPUs make more usable local LLMs, particularly in the speed department, but I suppose that depends more on how you really use them and how much you're willing to pay to have enough total VRAM.
It's next to impossible to make your money back on local (regardless what you buy) so I'd just say "go for whatever amount of best you're willing to put money down for" and enjoy it.
