Chicken and egg, companies stop paying, say no devs, devs not interested as no companies paying for it.
Tell me that COBOL to Java porting pays $150K p/a and I'll be happy to take the job, at least for a year or 3.
You work that as a contractor easy if you can find them. The problem is they're rarely hiring one offs to do this because these companies don't have engineering cultures, which is why they contract this shit to system integrators of the world like IBM, then they outsource the work to shoddy consultancy shops to maximize their profits doing basically zero work. If everything goes exceptionally bad (which it did at least in the case I was involved in), then they hire competent contractors to come in and save them from frighteningly large law suits. That's the new dev cycle of these legacy modernizations.. ahh
It’s more than that. If you spend a couple or more years learning to maintain those systems, where else will you work? Will you find a co-founder or investor for a COBOL based system when you go “entrepreneur”? Maybe, but it better be a damn strong niche.
Not moving forward with the software industry is a weird kind of conceit that separates these companies from the mainstream by far more than money.
It’s not the language, it’s the business rules and history.
Why is this input file loaded and rechecked 3 times? Because 30 years ago a file load failed, breaking end of quarter reports. This was the fix: if we can read that file three times and it doesn’t change then we know it’s good
It still comes down to paying enough money for someone to do it. Pay me twice what I make now and I'll switch to coding absurd legacy all-caps COBOL systems in a heart beat. Probably less then that too.
Pretty much exactly. There are wizards at Project Zero doing reverse engineering of systems at the hardware microcode level. Our own kens shows up every few weeks with transistor level explanations of devices whose designers are long since retired.
And we're supposed to believe that no one in the world can figure out a batch-processed COBOL accounting system? Please. The talent is there, it's just that no one responsible for these systems wants to hire at that level. But in extremis they will, and the world will survive.
Figure I might as well drop a quick review after 2 years with the lemur pro 11
Pros:
* Most things "just work", which you only appreciate after working with other linux laptops. For example, I can seamlessly plug this laptop in place of my work macbook with just one usb-c cable. That being said I think things have generally gotten better in the space so this may not be as much of a selling point anymore. Additionally this laptop doesn't have an nvidia gpu, which means its job is easier.
* Great compatibility for building software between my desktop and this laptop, makes my personal dev work a lot more portable.
* It's quite small and very portable.
* Nice keyboard
* Moral points for supporting a small company that focuses on security (whether this is actually significant is up to the reader)
Cons:
* Battery life is a lie, especially since it drains almost as much battery closed as it does open.
* Not great screen, terrible trackpad, and silly webcam considering the price of the laptop.
* As mentioned no gpu, while costing about the same as razor laptop.
Overall, I think I am probably going to switch back to a macbook after this, not being able to go a day without charging and your laptop always being on low battery is a bit anxiety inducing. Also (and this doesn't matter to a lot of people) I really value a laptop trackpad and this one is just plain bad.
> Most things "just work", which you only appreciate after working with other linux laptops. For example, I can seamlessly plug this laptop in place of my work macbook with just one usb-c cable.
I think any laptop sold in the last 4 or 5 years or so is plugged and charged with an usb-c cable and can be docked that way.
Agree, battery life is atrocious. I get typically get 1.5-2 hours on a charge, after 2.5 years of ownership. I always have anxiety about plugging it in. It's the biggest problem with this machine.
Other than that, I have been very happy with it. Keyboard, trackpad, screen - all adequate for me. In every way other than the battery, it pretty much gets out of the way and gets the job done.
I've had the complete opposite experience (I replaced their Pop OS with Arch Linux). With light usage I get about 10 hours, if I'm playing games or something then yeah it drops a bit but not that much. It easily lasts the duration of an intercontinental plane flight which is my primary use-case.
Maybe it's because I don't actually use it all that much, so my battery hasn't had many cycles put on it. I only use my laptop for travel, normally I have a desktop. That's why I went for a small, highly portable model.
My experience is the same. I replaced Pop with Arch running a pretty low-resource desktop setup (i3 and generally lightweight programs) and I can get roughly 10 hours with very heavy usage: Firefox with a handful of windows/dozens and dozens of tabs open, Docker, Spotify, etc.
Not terribly impressive compared to something like the newer Apple silicon MacBooks but also not terribly offensive considering I don't often work far from an available power source for super long stretches.
Wow, I'm surprised to hear this about the battery from you guys, as that appears to be one of their top selling points: "Most battery life!"
I see that the article describes it as repairable, but is it really easy to get and put in a new battery? I don't see them for sale on system76.com/components
I almost bought one of these in late 2021 when I was in the market for a new Linux machine. They were one of the few manufacturers that actually had stuff ready to ship. But I wasn't interested in PopOS and Framework seemed to be offering a slightly better deal, so I ended up waiting a month for a Framework DIY edition. I've been happy with it despite Framework not being truly Linux-first.
As far as I'm aware, every System76 laptop is a rebadged OEM laptop with an opinionated set of expansion components that they've effectively certified as functional in Pop!_OS by ensuring good driver and DWM compatibility. I think the OEM is Sager, but I'm not confident on that, or if that's uniform. Anywho, because of this, it should be reasonably possible to source replacement components upstream.
Yep, Sager/Clevo. Clevo L141AU has the same specs (2.5 lbs, 14 hours claimed battery life, 65W barrel charger, same keyboard layout, i7-1355U option). The case looks close, maybe identical, to the photos in the article. https://laptopwithlinux.com/product/clevo-l141au/
EDIT: Customized the Laptop with Linux Clevo to the same exact specs as the review unit and got a total price of $1,507, so about the same, give or take $30.
I wonder why. I'm running Ubuntu 22.04 on a Dell Latitude and battery life hasn't been an issue for me. I probably don't really do power-hungry stuff but just browsing, editing code, ssh, email, etc. it's been fine.
I think this might have to do with aggressive CPU throttling and backlight brightness. I've noticed both of those play a pretty big role in the discharge duration of my Dell Latitude.
I'm planning to send it in to them this week for some repairs (USB-C port stopped working, replace the rubber feet, and replace the battery). I have to pay for the battery replacement (~$100 I think). Hopefully it improves the battery life. IIRC, I was getting 4-5 hours when I first got the machine. Never got anywhere close to 11. Maybe it's how I'm using it.
Hang on, I thought this was still Clevo stuff. I bought a Clevo-ODM laptop a decade or so ago, the keyboard was atrocious in feel (no click left at all on most keys, just linearish sponginess) and activation (e.g. Space was quite difficult to activate, A would very regularly double-activate) within two years. And I know I’ve heard similar complaints regularly since then from others.
yeah - as seen from a few comments on my post, everyone has different opinions on non-objective stuff related to keyboard/trackpad. I feel like the keyboard is pretty responsive while the trackpad isn't, but the only way for you to know is to have tried it I guess.
I’m not talking about subjective things, I’m talking about keyboards objectively wearing out 3–10× as fast as any reputable brand’s. I’ve deliberately filtered for objective cases like my own, including querying specifically on the nature of what was bad in at least a couple of cases. I’m not talking about opinions, I’m talking about a keyboard becoming decidedly spongy in less than one year (compare the feel of the least- and most-frequently-used keys—on ones like ASUS and Microsoft, it’s taken much longer than that before I can readily discern any difference), and and almost uselessly bad within two years, in ways that other companies’ haven’t failed in four years (… though others haven’t been without their problems, but I’ve never had one get anywhere near as bad in general in three or four years as the Clevo one was after a little over one year).
Keyboards don't have to be non objective. I think everyone agrees that ThinkPad keyboards were great (though since the T14s gen3 they too have lost their quality)
A friend back in college absolutely hated the ThinkPad keyboards, (thought they were too deep and bouncy) and loved Macbook's butterfly keyboard when it came out, marveling at the short travel and crisp feeling of it. It's indeed quite subjective.
System76 is working with Clevo but increasingly with other vendors where they have some more input. Most new products aren't Clevo is my understanding.
Glad you are! I don't want to be overly negative. I am comparing my M2 work macbook trackpad with my 2 year old lemur pro trackpad - it's not really a fair comparison. The problem is that the pricing is similar enough where it's hard to justify the hardware downgrade.
I should also mention that my free time to work on projects has dramatically decreased in the past few years, so I am valuing the ability to seamlessly switch between my desktop and laptop on personal projects less than I used to.
> As mentioned no gpu, while costing about the same as razor laptop.
Consider the lack of GPU a blessing. You absolutely do not want a hybrid NVIDIA GPU laptop, unless you want to sit with it plugged in at a desk while the fans try and keep the GPU from melting through the case. Worse battery life. With absolutely not a single other tangible benefit.
Unless you are using the GPU for machine learning or w/e, in that case, the only utility it has.
Interesting, I've found hybrid GPU laptops to be pretty practical. I've had two over the years and on my newer one I installed PopOS, and there's a setting which puts it in hybrid mode, so by default the discrete GPU stays off, but you can run a program with some environmental vars and get it to use the Nvidia GPU.
When I'm not running a game, I get plenty of battery life out of it (4-6 hours or so?) and when I run a game I get decent performance. Exactly what I wanted. I haven't tried ML yet but I don't see why it'd be any different.
> but you can run a program with some environmental vars and get it to use the Nvidia GPU
Other than for gaming, which isn't worth it to do on a laptop anyway, I have seen no reason for this. Most discrete GPUs do a better job with drawing UI elements than a GPU anyway.
Usb-C connections being unstable in Linux is definitely still a thing. My Asus laptop (amd cpu/gpu) works about 50% of the time, and the other 50 it keeps blinking the monitor on and off.
I can fix this by plugging in hdmi first, then back to usb-c. Some sort of hardware reset gets executed that way, I suppose.
Re: Battery life, are you using stock Pop OS or did you put something else on? I have also gotten bad battery life but I'm pretty sure it is because I installed over it; the battery life claimns come as a result of power-saving features built into Pop OS.
These are product for certain type of people. I could’ve been that person maybe 10 years ago, or whenever I thought compiling Gentoo on a Pentium 4 was worth my time. But as I got busier with life, that opinion changed. To have to pay for a product that “mostly” works? (That doesn’t really once you read more details about it)
Yeah, that's probably why GP was compiling Gentoo on it ten years later: By then it had become a little too tired for Windows, but still quite viable with a crisp Linux distro on it. In case you thought you saw some discrepancy there, I think you're mistaken.
>Most things "just work", which you only appreciate after working with other linux laptops. For example, I can seamlessly plug this laptop in place of my work macbook with just one usb-c cable. That being said I think things have generally gotten better in the space so this may not be as much of a selling point anymore. Additionally this laptop doesn't have an nvidia gpu, which means its job is easier.
This might be your experience with System76, but it hasn't been mine. My Adder WS had infuriating software problems.
• It would regularly hang when disconnecting from AC. The only fix I ever found was a hard-boot.
• When disconnecting from AC, the CPU would sometimes get stuck at 800 MHz. The only procedure I found to reliably fix this was to reconnect to power, wait a few seconds, disconnect, wait a few seconds, and then reconnect.
• It would regularly fail to wake from sleep.
• The screen randomly flashed bright white when suspended, so I had to get in the habit of shutting the lid at night to keep it from waking me up.
• The fan would get stuck at 100% even when the temperatures were at 30°. Fixing this required sleeping and waking the machine.
Maybe some of the problems were caused by Nvidia, but I don't much care. The fact remains that I've been using Linux laptops since I was in high school and I've never had problems like these. Debian, Ubuntu, and Fedora all worked well on Dell, Framework, IBM, and Lenovo hardware (Yoga, not Thinkpad), as well as on my home-built desktops with Nvidia GPUs.
Also some hardware problems:
• The barrel connector was cheap and the power cable would regularly fall out. This exacerbated the software problems. The machine would get stuck at 800 MHz at least once a day and hang every few days.
• The machine was generally cheaply-built. The rubber feet fell off, the case creaked and flexed, several keys cracked, small plastic bits broke off, etc.
Credit where it is due, System76 support was responsive and replaced the mainboard promptly and free of charge. But that didn't fix the problems, so not all that much credit is due.
I wound up installing Windows on my System76 and giving it to my cousin as a gaming machine. I owned it less than two years.
>Moral points for supporting a small company that focuses on security (whether this is actually significant is up to the reader)
System76 rebadges Clevo machines and isn't very forthright about it. I find that questionable enough to outweigh my preference for small businesses.
Want to also add that unless you want to stick with Pop OS!, custom firmware will need to be installed, that custom firmware conflicts with packages from many distros. Without that custom firmware, your fans won't scale properly and you will struggle with hardware sleep, battery life, performance in general, etc.
Ironic that distro hopping on a linux-first laptop becomes difficult. But, priorities, I guess.
Maybe being stuck on Pop isn't an issue for some, but for those of us who don't like a UI locked in brown and teal that isn't being updated because they are writing their own entire (also ugly) DE, it is a problem.
Idk inventing convolutions and a bunch of other stuff we use everyday is somewhat significant. Machine learning, if we call it a field, is a whole lot newer than physics, and so a few people have contributed quite a bit in recent memory.
I have to agree. Especially given the very real possibility that your ML project won't be cutting edge research grade. At that point someone who doesn't have bias and is willing to search for a reasonable looking approximation to the problem and try a canned solution may actually be an optimal candidate.
Considering the number of problems that could be plugged into a random forest with good results, data proficiency seems more important than strong ML experience.
Depends heavily on the application once you get to more specialized domains.
I wish there was an easier way to label roles differently based on when you just need to throw X or Y model at some chunk of data and when more specialized modeling is required. Previously it was roughly delineated by "data science" vs "ML" roles but the recent AI thing has really messed with this.
They do have this, you can choose to bid for a target conversion cost (I want to pay at most $x per conversion). In general I find all of this pretty silly as a Google employee.. I don't think you can distill the complexity of any change down to one article. Say for example that Google did just raise how much advertisers paid per click across the board, why would that lead to billions of dollars in profit? I would bet that the majority of advertisers on Google already spend their whole budget, so that would mean that now they would just be getting less clicks and conversions per dollar of budget spend. Logically they would then move some of this budget to other advertising channels such as Facebook. Charging more per impression always has second order effects, but who cares about that on hackernews
But thats why modelling businesses has such a large opportunity. Some businesses will be prepared to pay more if you send them more conversions... Others have a fixed budget. Some use many platforms, and certain things might persuade them to move more budget to you. Some might be persuaded by "tricks" like handing out airmiles, vouchers or free stuff to the marketing team in return for ad spend.
All of those things require a higher level strategy than just doing everything per-impression.
I agree in general, but most people that run 50 miles a weeks also eat healthy. So there's a lot of correlation there. IE show me a person that runs 50 miles a week but also eats mostly process foods and drinks lotsa soda.
That demographic is Marines in bootcamp if they're from Appalachia.
Most people with a postsecondary degree who do something like use a gym or jog probably read Omnivores' Dilemma, may shop for organic food, and are more likely to be vegetarian. This demographic is largely incongruous with behaviors like tobacco and substance abuse, and frequent consumption of fried and ultra-processed food.
If one is running 50 miles per week one needs to orient eating around the runs. It's difficult to run right after a full meal. The result tends to be restricted eating to have a comfortable run in the first place.
PS: A bunch of extra work is near mandatory to maintain weekly mileage like this. Proper stretching and recovery--maybe some yoga. Lots of distractions from eating too much in the first place.
You don’t really need to do anything extra at 50. 50 is a basic baseline for a lot of folks running 3-3:30 marathons. I stretch for about 3 minutes after I run, and little else, for instance, and I’m at 60-70mpw, old, and a bit overweight.
I know very few people at this mileage doing serious “extra” work.
Meta is a company that makes money off of users endlessly browsing content. It would follow that making it easier/faster to generate content would benefit Meta.
When I asked it to help me practice French, Claud let me ask clarifying questions about specific phrases it used, with background on specific conjugations/language constructs.
I do wish that it's responses were more "chat like" though. I feel that its default response to even a simple phrase... "Merci!" - is something like
paragraph -> 5-6 bullets -> paragraph.
While this makes sense for technical questions, it quickly makes the experience of "chatting" with Claud pretty tedious.
This paper feels significant. If chatgpt was an evolutionary step on gpt3.5/gpt4, then this is bit like taking chatgpt and using it as the backbone of something that can accumulate memories, reflect on them, and make plans accordingly.
I think you're ignoring the work that went into this as well as the useful technology that came out. Prompts make or break interactions with LLMs and ChatGPT especially. The difference in output from a naive prompt and a well crafted one is huge. This paper is one of many explorations of what happens when you design such prompts to work in an iterative fashion building upon the previous conversation text to produce emergent behaviour. These are the seeds of the next programming paradigm on a completely novel architecture. It's incredibly exciting to be present for the beginning of this field, this is what mathematicians must have felt like when they helped design and program the first computers.
Oh yeah I agree that it _could_ do all those things, but it would be a bit of overkill to always send every observation an agent encounters into the API/chatbox, and ask it to spit out an evaluation or action.
This paper does a nice job of separating the "agency" from the next word with context type predictor. I think that's why I like the paper, it is just chatgpt, in the same way that pizza is just dough, sauce, and cheese.
Yes, but I think this was a fairly obvious conclusion to imagine isn’t it.
If you were going to seriously consider using ChatGPT for AI in a game, you would need each instance of GPT to only know certain information it has gathered. And you would want it to reflect on observations to come up with new thoughts that weren’t observed.
Still, I’d argue you don’t really even need GPT for any of the above. GPT is useful if you want thoughts expressed as natural language, but you could easily code observations and thoughts into an appropriate abstract data structure and still have the same thing, except it’s a bit harder to understand since asking an NPC something in a language it understands and getting back a query result isn’t user friendly, but it can be just as amazing if you know what the data represents. The imprecision and fuzziness of an LLM leaves room for fun weirdness though.
Time for my yearly tradition of making it to day 4 in Haskell and feeling very smart before completely giving up on the first hard problem that needs regular arrays :)
I know exactly what you mean. I'm going to try AoC this year again with Haskell, but this time around I'm just going to cheat and use it as an imperative language.
{-# LANGUAGE BlockArguments #-}
{-# LANGUAGE Strict #-}
stateful s' f = runState f s'
forEach xs state' f = foldM f state' xs
solve input = forEach input initial_state $ \st current_value -> stateful st do
...
I tried that last year, and got worn down by how often I'd Google how to do something and get an answer explaining how recursion works from first principles as opposed to just pointing at the deciding standard library function that did what I needed.
I want to try again this year, do you have any advice for how to get going with Ocaml as someone who has programmed long enough to understand how recursion, linked lists, and all that jazz works, and just wants to figure out how to be productive in the language?
Also you can use F# notebooks with .NET Interactive and the Polyglot Notebooks Visual Studio Code extension, which is great for something like Advent of Code.
I'm not an OCaml expert either. I've used Haskell professionally a little bit and also know OCaml. I find OCaml more practical because you don't have to fight with "pure" functional programming when you really want an imperative algorithm. For example, OCaml provides mutable arrays even though it encourages functional style whenever possible.
One downside of OCaml is that its standard library is not very powerful and there are multiple third party replacements that are popular. Jane Street's base/core, Containers, Batteries Included, to name a few. So you have to make a choice.
Having said that, I think you can live with the built in standard library to solve most of the AOC problems.
I did OCaml the passed two years. GPL productivity in OCaml is easily accessible, but you need opam packages. Its not a deeply batteries-included language at all. With the exception of eio, my goto set of base packages are: https://github.com/cdaringe/protohacks/blob/main/dune-projec...
containers is my preferred “add batteries” std lib extension. Also, the ocaml discord chat and discuss forums are very active and welcome to questions
