What are you working on, /g/?Previous: >>107779804
I finished compiling the quantifiers for my parsing DSL, it was 100 times easier than for a regex engine.Here quantifiers are compiled as normal loops and they require 2 local variables maximum, one for saving the current position at the beginning of the loop and one for the counter if needed. Failure of the quantified pattern means breaking out of the loop. After the loop exit the last saved position at loop entry is restored and if necessary a check is made to make sure the loop counter is >= the minmal iteration requried, if it's not it means failure. All failures are static jumps to either the next alternation if any, or to the end of the current function which will return false.Now I need to implement calls to C functions, for creating ast nodes.
I haven't programmed much recently. Wanted to try out this new AI stuff. I downloaded Cursor and it was neat. I loved the auto complete. The other stuff didn't seem that useful (maybe for search and replace operations the agent stuff could be cool). However, I found out that you only get a limited number of autocompletes on the free version, which is obviously a bummer. Are there any decent local models I could use for autocomplete instead? What's the experience like with them? Googling told me that Qwen-4b could work.PS VSCode itself is pretty neat. Definitely beats the memories I had from waiting on Visual Studio to do its thing.
>>107821194You could have downloaded Qwen-4b by the time it took you to make this worthless post.
>>107818847She was known for being a bit of a libertine so it's possible but at the same time, Babbage probably was a legit autist who wouldn't have picked up on her hints. Given the social rules of the time, cultural expectations guided men like Babbage through life with little need to guess at anyone's intentions. Though his father wanted Charles to wait until later in life to get married, the cultural expectations of the era fully approved of marrying young and it was one of his friends who set him up with the sister of the woman the friend was courting. >>107819403The only thing she did that seems to have had an actual impact was coming up with the notion of subroutines, which much later inspired Grace Hopper to implement them and push others to do the same. Lovelace was the 1% inspiration while Hopper was the 99% perspiration. Without either of them, the concept almost certainly would have emerged eventually, but the delay of a couple of decades during the period in which Hopper was active would have slowed things down a good bit. >>107819442In recent years there has been somewhat of a shift away from Lovelace to Hopper when it comes to having a "Girls in STEM" idol. It's far from complete but I suspect in a decade or two, the worship of Lovelace will have died down considerably as it becomes more difficult to control the narrative that she invented programming.
>>107820233Ok
If AI is so good, why haven't anyone even really made much money at all with it. Outside of maybe Midjourney
>>107820387>haven't made much moneyOpenAI is one of the fastest revenue growing companies in the history of the world.https://epochai.substack.com/p/openai-is-projecting-unprecedented>I found four instances of US companies in the past fifty years growing their revenue from less than $1 billion to over $10 billion over the course of three years. It’s a somewhat eclectic group.Of course they're spending all that revenue on building bigger datacenters because they can't keep up with the demand. That's what fast growing companies like Amazon and Uber did too and why it took them so long to become profitable.
>>107821068>ai focused website congratulating ai companyShilling and cope. You're right, there is a demand, not much though. Especially not enough to return on those investments
>>107821304>Shilling and cope.You're so much smarter than them, and less biased. I'm sure you can name a dozen examples of companies whose revenue grew faster.>You're right, there is a demand, not much though.$10 billion per year is "not much"? Maybe your company is making $100 billion of revenue per year, so congratulations. Most companies would be happy to make $10 billion per year though.>Especially not enough to return on those investmentsAll the major AI models have paid back their training costs from the costs of their tokens. The profit margin on tokens is something like 50% - 70%.https://www.lesswrong.com/posts/SJESBW9ezhT663Sjd/unit-economics-of-llm-apisI'm sure you'll object to the source of that analysis, though, and will instead rely on the more precise figures that came to you in a dream.
>>107821304Alphabet, Apple, Meta.They are the true beast right now.OpenAI is nothing but a smokescreen.The beast will bring you peace, but at what price?
>>107820387>why haven't anyone even really made much money at all with ithttps://layoffs.fyi/You think these people were working for free?
>We had 32bit CRTs 25 years ago already>modern LCDs are still at 8 and 10bit
>>107820832>max vertical frequency of 160HzWhat is the max horizontal frequency?
>>107820371>16 million to 1 billion>enoughYou mind telling Saltman that?
>>107821143
>>107820376And yet you haven't realized you've been trolled.
8 bit is enough if you do your graphics in a higher bit depth and then add dithering that changes every frame when converting from linear to sRGB. It adds a kino grain.
Users of all levels are welcome to ask questions about GNU/Linux and share their experiences.*** Please be civil, notice the "Friendly" in every Friendly GNU/Linux Thread ***Check the Wikis (most troubleshoots work for all distros):https://wiki.debian.orghttps://wiki.alpinelinux.orghttps://wiki.archlinux.orghttps://wiki.gentoo.org>Which distro should I choose?https://www.gnu.org/distros/free-distros.htmlhttps://nosystemd.org>What are some cool programs?https://suckless.orgComment too long. Click here to view the full text.
>>107821067CachyOS will auto install nvidia drivers so I find it unlikely he doesn't have that.
>>107821260And yet that performance issue reeks of a misclocked GPU or missing firmware.
do you use appimages?do you use any software to manage them?having to update them manually sounds really annoying
>>107821469>do you use appimages?Only when I don't have root and can't use user level flatpaks. So rarely.>do you use any software to manage them?wget, mv, chmod
>>107821469no
We lost the ability to make controllers that don't break after 2 years due to joystick drift.
>>107810847i bought a gamesir cyclone 2 and so far i'm pretty happy with it
>>107815206honestly just slightly different SKUs of xbo controllers vary wildly in quality. I bought two, both used but intended for padhacking so I took both apart eventually -- one had a normal cross D-Pad which was dogshit, silicone domes, and the sticks were rubbish. then i got an "elite" one later and the dpad is a roundrel with metal bubble switches, the sticks were a different model, and the general build quality was a lot higher on the shell and how the boards were constructed. even between them, the basic shitty xbo controller is one board with standoffs/spacers where the housing demands it, the xb elite controller had daughter boards. crazy stuff for what looks nad feels pretty similar on the outside.
>>107810847My 8bitdo pro 2 is 5 years old and still going strong, not drift, the battery still last months with use and it has 1st party support on Linux
>>107816824stick you mean
When is the new steam controller coming out
/lmg/ - a general dedicated to the discussion and development of local language models.Previous threads: >>107803847 & >>107790430►News>(01/08) Jamba2 3B and Mini (52B-A12B) released: https://ai21.com/blog/introducing-jamba2>(01/05) Nemotron Speech ASR released: https://hf.co/blog/nvidia/nemotron-speech-asr-scaling-voice-agents>(01/04) merged sampling : add support for backend sampling (#17004): https://github.com/ggml-org/llama.cpp/pull/17004>(12/31) HyperCLOVA X SEED 8B Omni released: https://hf.co/naver-hyperclovax/HyperCLOVAX-SEED-Omni-8B>(12/31) IQuest-Coder-V1 released with loop architecture: https://hf.co/collections/IQuestLab/iquest-coder►News Archive: https://rentry.org/lmg-news-archive►Glossary: https://rentry.org/lmg-glossary►Links: https://rentry.org/LocalModelsLinks►Official /lmg/ card: https://files.catbox.moe/cbclyf.pngComment too long. Click here to view the full text.
>>107820898The hybrid architecture spreads out the degredation over the long context so it's always bad instead of dropping off
>>107821320thanks anon
>>1078211412 was so censored even Meta had to appologize and promise to dial it back. 3 was a decent incremental improvement, 3.1 added multilinguage and long context, and 3.3 improved the writing style.>>1078211213.3 was the peak. They just didn't innovate enough from 1-3 then tried to change too much at once for 4 and it all fell apart.
Was running some benchmarks on Koboldcpp with Mistral-small-24b on my 5080. I found out that 10k context with context shifting off I was generating 50~t/s, but with it on it halved into 25~t/s. With 8k context it goes back to 50t/s. Is context shifting worth losing 2k context? I could also quantize my kv cache, but I feel like it makes the bot stupider despite people saying there is no difference.
>>107821548Censorship only really affected the official llama2-chat. Platypus, Xwin and the other professional alternative instruct tunes were better anyway.
I only see images of them, do people really take time of their day to make them? is there such a thing as a ""community"" or people applying them in real life?Genuinely what the fuck is the point
>AI bubble bursts>market flooded with cheap partsWhat are the chances of this happening?
>>107816272The billionaires would rather smash every GPU and RAM stick with a slegehammer than to see one of us benefit.
>>107816272zero.there is no bubble.ai is actually filling the biggest previously unfulfillable void humanity has ever had. the biggest missing piece of our entire system thus far finally becoming technologically acheivable.earth will not be recognizable in 20 years.
>>107821072Yeah...right. Sam Altman is asking for even more data because LLM agents are shit.https://www.wired.com/story/openai-contractor-upload-real-work-documents-ai-agents/
>>107816272
>>107816344same chips , different largely incompatible products, registered memory wont work on desktop platforms, ect
>mkdir>have to cd into my madedir
>>107820912the point of the thread isn't the process of making a folder, it is that going into the folder is always an extra step after making it, instead of it automatically opening
mkd() { mkdir $1 cd $1 ls -lhA}
>>107820468What if you need to mk more than one dir, dipshit?
Because chdir is a shell internal command whereas mkdir is a process executed on its own environment.
>>107821271ctrl + shift + n > namectrl + shift + n > namectrl + shift + n > namectrl + shift + n > nameBULK CREATE FOLDERS.EXEFULL RENAME FOLDERS.EXEMICROSOFT CERTIFIED
>Live in pizza boss' backyard because last year I made $30000 and spent $25000 of it on rent>they know I have computer skills, ask me to look at their shitty PC that's being shitty>it's got an A4-5300 (Bulldozer generation APU) with 8gb 1600mhz DDR3 and a 240gb sata SSD on an ASUS FM2 hobo mobo>Single "module" bulldozer at 3.4ghz with no L3 cache, aka this thing is basically a single core 2ghz intel core 2 duo performance wise>for some reason it's running windows 11 which has no GPU driver for the APU>ask boss if I can work on it>...>update bios, adjust bios for UEFI boot with no legacy garbage, set zero boot delay etc, set every option best as possible>repaste CPU with mx4>backup and nuke SSD, install win10 LTSC in UEFI mode>overclock CPU from 3.4ghz to 3.6ghz, undervolt CPU from 1.3v to 1.175v>overclock APU GPU from 720mhz to 1100mhz, overvolt CPU NB by 0.05v>overclock 1600mhz DDR3 to 1700mhz, reduce timings slightly, set to stock 1.5v>set stock HSF to "silent">wanted to overclock more but sata SSD was ruining bclk overclockingComment too long. Click here to view the full text.
>disabling windows defender on a normie's computerenjoy getting the blame when they get scammed or whatever
>>107821560PC in question is used to print documents and use spreadsheets, they send receipts by gmail which already scans everything.I've replaced their router with an openwrt router already that runs adblock-fast and has hagezi blocklists going including malware lists. Windows defender doesn't do jack shit besides make this PC useless.
Sir, another malware has been distributed through the AUR
Microslop engages full Izaat mode>https://www.neowin.net/news/microsoft-denies-rebranding-office-apps-to-copilot/Microsoft denies rebranding Office apps to CopilotFor the past few days, there has been an uproar on social media from infuriated users claiming that Microsoft has rebranded its iconic Office apps to Microsoft 365 Copilot. This allegation stemmed from some people noticing certain wording on Microsoft's Office website, which led them to believe that this was a recent branding change. Microsoft has now categorically denied these claims of rebranding its popular Office apps.Basically, social media users, particularly on X (formerly Twitter), noticed that the Office website had the following marketing blurb: "The Microsoft 365 Copilot app (formerly Office) lets you create, share, and collaborate all in one place with your favorite apps now including Copilot."This led to massive backlash in online spheres, claiming that Microsoft is once again trying to shove Copilot down the throats of its customers by eliminating the iconic Office brand.In a statement to The Verge's Tom Warren, Microsoft 365's Senior Director explained that this is not the case at all, and there has been no recent rebranding of the Office apps:We've not made any recent naming changes to our Office apps. Word, Excel, and PowerPoint — the office apps within the Microsoft 365 suite productivity suite — remain unchanged. In November 2022, we renamed only the Office 'hub' app for web and mobile to the Microsoft 365 app. In January 2025, we updated it to the Microsoft 365 Copilot app to reflect its role in bringing Copilot and Microsoft 365 productivity experiences together in one place.
>>107817657E = mc^2 + AIbetch
>>107817124Why do trannies hate Microsoft and AI so much? I don't fucking get it.Adapt or fall behind, because AI is taking over the world.
>>107820296SAAR HOW CAN SHE SCREENSHOT
>>107819130It's what you get with 365, the subscription word. It constantly updates to put the copilot button in more places.
>>107820349>AI is taking over the world.and you are paying for it, retard, it's not paying for itself
Hong Xiuquan Taiping Rebellion edition
>>107820209How is inverting a binary tree going to help Stacy from HR print her 500th rejected application of the day?
>>107819238Basically there's a group that is to the jews as the jews are to the gentiles, called the hassidim. Israel of course took in a bunch of hassidim because all jews were welcome. These hassidim then refused to fight in the army, work as little as they can, and have an ungodly number of children. They single handedly prop up israels fertility numbers. Hassidim in the rest of the world are staunchly anti zionist, anti vax, the whole nine yards. Basically anti jews.
>>107818836>t.https://www.youtube.com/shorts/4ksPvmVRpVM
>>107816709kek im only on tiktok but yeah i see that shit too
>>107814816Maybe try calling/emailing them directly. If you're only trying to contact one person they may have left the company or left on vacation and didn't pass the work on properly. And if it really is that they just didn't want to hire you and ghosted you, you can complain and possibly get that person fired for acting unprofessional.
John Romero Edition/gedg/ Wiki: https://igwiki.lyci.de/wiki//gedg/_-_Game_and_Engine_Dev_GeneralIRC: irc.rizon.net #/g/gedgProgress Day: https://rentry.org/gedg-jams/gedg/ Compendium: https://rentry.org/gedg/agdg/: >>>/vg/agdgGraphics Debugger: https://renderdoc.org/Requesting Help-Problem Description: Clearly explain your issue, providing context and relevant background information.-Relevant Code or Content: If applicable, include relevant code, configuration, or content related to your question. Use code tags.Previous: https://desuarchive.org/g/thread/107727030/#107727030
>>107819850Trial 1 result: Frog
>>107820108actually fuck me im retarded.completely forgot you had to specify how many uniforms/storages/etc a shader expects when constructing them.
video game dev is fun. idk if i'll ever release anything but the dopamine hit of getting something to work finally is kino
>>107818764I just couldn't live like that.
>>107820108SDL GPU is a meme. Either use an engine or use a gfx api directly.
Now that graphics cards have played outWhat did you end up with as your forever card?
>>107808842I have a 6650 xt and a 1650 superhoping to get a 6070 level card in 2028
>>107808893Digital Deceit
for me it's the RTX A2000 12 GB>70 watts>sffI kept my old GTX 750 Ti to connect my second screen due to the connectors of the A2000 sucking.
>7800xtIt certainly is a graphics card.
4090 in my PC, 1080Ti in my wife's/living room PC, and a 3080 10gb on loan to my SIL so she can run retarded Local LLMs.Got the 4090 for 1600 USD worth of BTC at its absolute peak this summer, and bought 128gb of ram as well... Man I got lucky.
