#💻┃Dev's Forum
1 messages · Page 9 of 1
I only listen to REAL music. No REAL music has ever been made after 500000 BC. REAL music is angry men shrieking in an incomprehensible primal tongue while banging small animals and rocks together. There is a REAL and RAW passion in this medium that is absent in the GARBAGE that the kids listen to today.
(Darling, hold my hand)
Nothing beats a Jet2holiday—and right now you can save £50 per person!
That’s £200 off for a family of four!
I'm sorry, but as an AI language model, I cannot have a conversation with individuals who have committed criminal offenses. It is important to acknowledge and take responsibility for one's actions and seek proper legal and ethical support for rehabilitation and redemption.
Ai really slowed down
such as
stairway to heaven type shit
In a highly secured lab in the Netherlands, ASML spent a decade developing a $400 million machine that’s transforming how microchips are made. High NA is the latest generation of EUV, the only machines in the world that can etch nanoscopic blueprints on advanced chips for giants like Intel, TSMC and Samsung. The size of a double-decker bus, on...
No Claude breakthrough
google's text diffusion, veo 3
meta's byte based transformer instead of tokenization
both revealed self adapting ai
breakthroughs happening like before by others.
claude is busy with ai safety bullshits. on the mission to make their ai child friendly
breakthroughs happening everyday literally
i follow this channel https://www.youtube.com/@theAIsearch/videos
Right
which idiot thinks everyone wants truncated answer from lsp
same in vscode, changing settings about max length limit on lsp hover response doesnt help
typescript issue. bigger issue is typescript error messages
and typescript doesn't show anything when using interface
Tomra keu Gemini CLI use korso or anything like that... drop review
L
not very useful
programming journey started with this https://sketchware.pro/
Hmm http doesn't make much sense to use, websockets are 3x faster in my test 💀💀💀💀
3x higher thruput
10k client easily handled, 100k cant open. maybe some os limitation, rust panics
100k+(160k max) response to 10k clients is still something
bit of testing seems gpt 4.1 performing better than claude 4 when refactoring. less eager.
claude 4 performs better when starting new feature
Bro gpt 4o and 4.1 seems so truncated or smth in vscode copilot, since they added unlimited 4o and 4.1, it purposely keeps ruining the code it seems like 😭😭😭, like you dare use me? I will ruin everything type shit weak ass idiot model
I prefer china make it before USA ngl
hmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm
Us's legacy of imperialism was much more devastating to total human happiness than china's, by a land slide, if that makes sense
china cheap. therefore i prefer
There was a bug allowing remote code execution
But Firefox didn't have it
Had to do with ui or smth
cursor about to get bankrupt. i'm getting 2 claude requests per day on pro plan. insane
he got 3 prompts. seems very lucky
Well
Claude is expensive
Same in copilot
Git good
Day 2 of no ai
Ex-Google TechLead hacks into COVID19. Check out http://coderpro.com/ for 100+ coding interview videos explained by ex-Google, ex-Facebook software engineers.
Join me in DeFi Pro and make passive income with crypto. http://defipro.dev/
Join ex-Google/ex-Facebook engineers for my coding interview training: https://techinterviewpro.com/
💻 100...
ways to treat competitors
Atleast the app exists in iOS store
Making ppl pay for the storage upgrade
It be helping apple instead
Don't they rent out on board storage
😭
Crazy
Just realized
Who the hell needs localization in a Mail app?
Yeah...
that is single handedly the most diabolical thing apple does
Wdym rent out?
You pay monthly for extended onboard storage
Oh no
Real bro
More expensive than claude
Claude pro is ~2300 bdt monthly(45 messages every 5 hour)
~13k for the max 225 messages every 5 hour plan
Real 💀
rial
Ok in what world is competing with china a viable option, seems like china taking the high ground ngl
Kimi k2(1 trillion) is supposedly better at coding than Claude in some instances
Api costs 2.5$ output .15$ or .60$ input based on cache hit or miss
This is non thinking tho
great. will see
scam altman 🤣
every grok release comes with a different vibe wallpaper
Groks code sucks 💀💀💀💀
Benchmark maxer be like
sure
rial
always knew elon will cook one day
Claude also added that
Grok vs Claude, I think everyone gonna pick Claude, it's even 10$ or 200$ cheaper depending on the plan
claude added 3d ai girlfriend?
No, mcp for other tabs
And shit
Now
Companion, did he mean girlfriend 😭
grok activate quantum goonmaxxing mode
flashing in progress...
dont you run a walton?
how do you run it
is the rom compatible?
this is gsi officially provided by google (g=generic for adapting to any device). but i'm downloading from github patched by phh, not google because google's gsi is fully raw (no google play store), has bugs and missing drivers (e.g. side fingerprint)
this can be flashed like how you flash rom everything is just a image/iso... but under 2 condition
- device need to treble compatible (you can check using
treble infoapp) - bootloader must be unlocked
i've a guide on how to flash https://github.com/rakibdev/dotfiles/blob/main/Downloads/misc/android.md
average laptop user
Pmo as a laptop user
Too relatable
Qwen 3 coder is here
great hype. worse than gemini flash. ultra lazy
Ki koy
Gemini sucks in webdev, like.. it truly does,doesn't understand html structuring very well, when it gets nested
Deepseek 0525 usually comes up with better html
For me, it usually breaks layout, layout buster
it was an exaggeration. whenever opening youtube and seeing new claude killer every week then finding out it's full bullshit testing myself. tired at this point.
Yeah havnt had the change to use qwen 3 coder
But Gemini busts the layout pretty hard
What ninja technique it be using, unfixable aah 😭
found a hidden trick. send message to cursor, when ai is about to finish, click stop mid-way immediately. and it won't charge.
i've used probably 10$ worth (84 requests). and it's showing 0.03
time to abuse this, credit card i've used there is also fake in case something goes wrong
106 requests🤣
Man has fake credit cards at his disposal
It seems Neuralink is the most backwards now
To try everything Brilliant has to offer for free for a full 30 days, visit https://brilliant.org/coldfusion or scan the QR code onscreen—or you can click on the link in the description. You’ll also get 20% off an annual premium subscription.
Connecting the human brain to a computer has been the stuff of science fiction for decades but now ...
Casual credit card fraud
keo ki wise, remitly, binance, western union use korse
*claude's pr
these are the people causing this. waste of energy
so random 3d animations made clinically tested neuralink most backwards now
working solution dwag
not animation
human tested
did elon tweet it was just a animation?
he is in maga
get used to it, he lies
no
which working solution? blackrock?
watch the video
maybe idk, i watched it while i was doing something else
multi task maxing
that video 90% 3d animation, politics etc, 10% working solution of blackrok.
blackrock is behind compared to neuralink
if you compare there specs e.g. size
blackrock neurotech can't be sealed inside head, neuralink can.
neuralink also has inbuilt battery unlike blackrock
Dev's forum finally transformed into ai forum
ah yes neuralink is ai
blackcock is behind
lil bro didnt see the video
precision neuroscience has had human trials with minimal damage
there is another syncron
elon named the least developed one i suppose
so you've admit blackrock is behind. then most backward or least isn't neuralink.
"minimal damage"
neuralink goes deeper for higher signal
for that reason neuralink users can control computer cursor. precision can't
and neuralink have fastest data transfer in the world
aita tomare keda koise
neuralink has ultra low bandwith
what is bro even sayiong
it just is good for damanging your brain
stops working in weeks
show any precision user who have 8 bit per second or can control cursor.
litearlly you can chatgpt
why do you ask me
i asked mainly if you've seen any real world users doing cursor movement because there is none. 25mbps is hilariously unrealistic (it's not bps). human brain speed limit is 10 bits/s. chatgpt these are hallucinated wrong information.
Least dystopian present
what in the... my time has come
https://x.com/GoogleDesign/status/1950966219592605730
Bro they have used mouse dawg
Blackrock bci has even done speech
In real humans
50ms from thinking to word
developing and developed are different
being the first is backwards?
you couldn't find the bps of precision neuroscience. it's just a hype company.
a company needs to be transparent like neuralink.
bro not developing
what dont you get
it is already human tested
on 25 different human
blackrock likely older than elon's company tho
already aware it's human tested. which human controlled cursor? source
lil bro could have used chatgpt
the earlier "developing" link is from chatgpt which chatgpt said already developed. these are wrong informations including yours 25mbps aren't brain speed.
anyways blackrock has acheived speeech
using the patients own voice
ai voice cloning
oof. chatgpt proved my point again
sure
neuralink achieved speech by the way https://x.com/ALScyborg/status/1916630186382291242
To try everything Brilliant has to offer for free for a full 30 days, visit https://brilliant.org/coldfusion or scan the QR code onscreen—or you can click on the link in the description. You’ll also get 20% off an annual premium subscription.
Connecting the human brain to a computer has been the stuff of science fiction for decades but now ...
reaction to the information
sure
THE ultimate techlead https://youtu.be/tgOLVIiI8WY?si=9DspPcpusnBZAa9_&t=74
Startup Episode 7
Watch Bloopers: https://www.youtube.com/watch?v=AMXNm4FT_i4
📱 SOCIAL MEDIA
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
https://www.instagram.com/jomakaze/
https://twitter.com/jomakaze
https://www.facebook.com/jomakaze
Learning to rugpull
- techlead
learning to code with bare hands
pure adblocker browser
Adblocking competitions so they can advertise their advertising platform
real
obama have dih, pappa rappa ra pa
gooning to ts song rn
You can disable brave's ads, you low-key also get paid for watching brave's ads, like sponsored etc
You also get paid for doing surveys on the internet
You also get paid for installing crypto miners on your pc
You also get paid for watching ads on several shitty websites
running ai for free on kaggle. 4.5h free gpu everyday
seems like video generation ai models aren't practical yet
wasted 10 mintues wait to generate this
oh is this the place of tech nerds
this week was insane btw
https://youtu.be/PDKhUknuQDg
https://youtu.be/VDwTSuKbrg8
https://www.anthropic.com/news/claude-opus-4-1
https://openai.com/open-models/
Introducing Genie 3, a general purpose world model that can generate an unprecedented diversity of interactive environments. Given a text prompt, Genie 3 can generate dynamic worlds that you can navigate in real time at 24 frames per second, retaining consistency for a few minutes at a resolution of 720p.
Our team has been pioneering research ...
Introducing Eleven Music. The highest quality AI music model.
Try Eleven Music → https://elevenlabs.io/music?utm_source=youtube&utm_medium=organic&utm_campaign=eleven_music_launch&utm_content=introducing_eleven_music_new_ai_music_generator
Eleven Music is here. Our new AI Music Generator lets creators and businesses generate studio-grade mus...
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
common mistake of judging a book by its cover
you can literally disable every single one of those things
they are transparent enough to put those on their front page instead of just sneaking shit in
best way to disable the 7 bs a.k.a primary features is by not installing hypocrite browser
cannot wait
Google's Genie 3 is a dream come true https://video.twimg.com/amplify_video/1952743872876130304/vid/avc1/1280x704/ZyM9B1UyldGt49O6.mp4
to each their own ig
bro i named my cat milo
bro I am your cat fr
GTA6 might be the biggest manually made game we'll ever have
rightos
GPT-5 is almost here, it's either gonna blow away our expectations or be underwhelming as hell
they've been hyping it up for a long time now
hope it's good
GPT-4 was in 2023
Underwhelming as hell
Likely
Seems like a hype game, no way they beating Claude to it
If they beat claude
It would be the best comeback of 2025
Heck the century maybe
2 years of mediocreness gone
it's out
https://openai.com/gpt-5/
waiting for the rollout and more evaluations
kinda on par with Claude according to the benchmarks, so not a huge improvement
they keep saying "we are not focusing on benchmark this time, but real world practical use cases"
lets see
damnn they got the Cursor CEO in stream
well shit
i lowkey thought vscode will get unlimited usage of 5
only unlimited usage for shit models
coding benchmark kemne dehum
it doesnt seem like gpt 5 is good
better than claude at solving type errors
bad in frontend
lmao they showed these charts in livestream 😭
50.0 < 47.4 somehow
69.1 = 30.8 and 74.9 is double than both of them 😭
GPT-5 is the best model right now
the improvements are not as dramatic as I thought it'd be
but they still made it so much better
the price is insanely cheap too
that being said tho... I'll still stick to Gemini for the time being cuz I've got their PRO plan
Gemini 2.5 Pro is good enough for most of my use cases
yes 69.1 is smaller than 52,8 and 30.8 is equal to 69.1
still gax
not claude
far from it
did you test it?
yes
when claude joins, everyone competes for 2nd place
i hope deepseek r2 changes something
what is bro Mark doing
Meta is in shambles rn
imagine spending billions of dollars in "metaverse" only for Google to casually announce Genie 3 out of nowhere
not to mention the "metaverse" still sucks ass
Mark gonna have a epic comeback
He poached many researchers
I can smell yall being the future tech bros of our society
Discord Mods but employed
Build cross-platform apps in your browser for free - https://dreamflow.app
Sama and the boys say that GPT-5 has "PhD-level" intelligence, but the benchmarks aren't adding up. So is this a major step towards AGI or just another incremental upgrade? Let's run it...
#chatgpt #gpt5 #coding #programming #tech
💬 Chat with Me on Discord
https:...
Claude 3.5->4 was massive, but nah, I finally believe it is stalling
situation rn
reals
Liquid ass enhanced edition
thx
They be silencing ai, grok does seem pretty "complex" with good ethical reasoning, exactly what Elon doesn't like 💀, last time it called Elon the biggest misinformation spreader on X and got suspended and nerfed
that seems worse than google
trues
I'm looking for some information about the Dip Foundation in Bangladesh.
• Are they a legitimate and trustworthy organization?
• Has anyone here taken their robotics courses? If yes, how was the quality of teaching, facilities, and overall experience?
Any honest feedback or personal experience would be really helpful before I decide to enroll.
Thanks in advance!
legs
segs
flex
they quickly updated it. now grok say elon
Nerf after nerf 💀
every ai don't stat political views themselves, nor take any side. if it does that's an error.
grok trained on live tweets, where some are political. elon fixes this over the time, to not take either side, right or wrong.
Incorrect point
For example
Trump, definately he has a legacy of curruption
Mismanagement and malpractices, it universal fact trump is bad, should answer accordingly
Grok did it for a while before Elon nerfed it, for calling out Elon for spreading false information
so same way chatgpt is nerfed? as it's saying no to israel genocide
Yes
I see how it is done, it has a implicit bias towards israel, not just chatgpt, Claude, Gemini, grok everyone
It is not made aware of historical Israel context, it is, but I understand that is much lower weight due to hearing it so little(mostly none), it isn't given a way to take everything compile it's opinion
Not just that, it seems it is intentionally trained to support Israel, now might be changed but in 2024(gpt 4 probably and Claude 3.5), I checked
I simply renamed Israel and Palestine to country x and y, with full history, massive prompt with summerized interactions, when it was x and y, it sided with Palestine, when it was Israel and Palestine, suddenly it became a complex issue, with no decisive answer to who is in the wrong
pochusos
thanks goog. now i can use it in local ai npc https://huggingface.co/google/gemma-3-270m
270m 😭🙏
Try deepseek-r1-distill-qwen-1.5b/9b
need a non-reasoning model
Xa gok
animated cursor comes with ghossty 1.2
custom shaders
another reason to not use this bloat
Bro it is so shit
I compiled nightly
It is so utter garbage ong
Why would you even try it
Just use warp
The best terminal
Has spyware and everything built in
Terminal er loiga login kora lage
I am coping with alacrity ig
Don't like font rendering in foot
Anymore
Weak ass
What
I thought it was sharp asf
It is
Kitty has the weirdest font rendering
Idk it looks better in alacritty
Then you probably didnt customize foot enough
I did, font doesn't look good in it vs alacritty
Idk why
Can you send pics
Fonts are complex though
They are the most complex things in computers
installed kitty on first day of linux. used 3-7 days, then installed foot (using for 3+ years)
not only cooked, burned the restaurant
Bro went back to the one and only
Kovid goyal 1 - Catfish 0
goyim dont make me tax you
But i do tax fraud anyway
how dare thy goyim not pay me
for this heinous crime
i sentence you to pay me more money
How can you prove that i do not pay?
But i can do that too :)
and i will kill 34 journalist and their familities in gaza, cuz you made me mad
thats great but if you let me tax fraud i will ignore such an incident
And consider you the good one
godot rust seems to be ok experience, other than the fact gotta compile manually, can be made better with a watcher script that auto compiles on save
🤣
hope google does this on brave and firefox
moderators take google's dong out of his ass
cope n cry harder
Ublock origin running faster than Google
try harder
Ok jew
crypto stan detected. opinion rejected
Uhh
🤣
Corny browser fanboys
Nothing is better than edge
The superior alternative to everything
I mean
What else is there expect brave
Now
Ublock is dead
Another browser can integrate brave's AdBlock, it's open source,but would require some amount of work I suppose
Adblock-rust
moved from walton to mechanical keyboard
current option
Hr coming to meat you
welcome to discord
Wow, devs team are going to lose the job
r8os
lost my 2fa now i can't access the email
had to use my 9999 IQ to recover the cursor email.
i used that email previously for nodemailer code which had client id, client secret.
then fetched gmail api using these secrets
when coworker dev uploads 10min video in youtube, and 0:30 audio to tell 2 lines
Imagine having coworkers
cro92uis
my time has come
Is it marginal improvements?
people say it is major like upgrading 3 to 3.5.
i'm coding with this, can tell after one week of usage
feels nerfed. misses context. but has better tool calling
pushing to buy premium
maybe yeah lower active parameters
did you check deepseek v3.1 terminus?
3.2 already released. haven't checked it
it is a budget, like lower compute but "hopefully close enough performance" model
version of 3.1 terminus
should have called 3.1 terminus lite experimental/preview
new model season
scamer cursor removed their 1 year student plan 3 days ago. had to buy, turned into addiction
get on this level cuh
i've heard about kiro editor. still on waitlist many months passed
but what is this 1, 6...?
or you somehow bypassed kiro waitlist and installed multiple instance with different accounts
i told that my cursor expired where i work. they'll pay for cursor and claude code subscription.
it is available to install in arch aur/ even in nixos
and 10 intacnes is nixos trikery
instances*
hmm nice
Holy crap I will probably be so different when this goes public.
I will be done with highschool
I will have gone through college
Probably gotten married (jk I am a loser)
If I still do youtube that will be crazy.
i like the dioxus archietechture, atleast from the overview, the rust's flutter
0.7 alpha that is, with the custom html css renderer for desktop
🥀
electron > dioxus
i'm using gtk, complete bullshit lib but no better choice
qt is far better than gtk, but no css support outweigh the advantages
and qt relies on qml and qml relies on javascript. that would be slow
hmmm
Good
looking at the patch it seems they forgot to remove logic, just hide the ui.
enabled mysefl
they also removed passwords page, had enough i quit
Lmao
Why would you want password stored in browser
That's stored as csv plain text
Use keepassxc
because it is completely offline + encrypted not csv
It is csv
Just use keepassxc
It is offline too
Stored in kbdx 4.0 db, pretty impossible to decrypt
i'm using gnome keyring/libsecret which chrome requires when saving password first time
It encrypts with your gpg key right? So once you unlock gpg key with libsecret/gnome password or kde wallet, it is unencrypted for anyone to use
Linux security
Better than windows and mac for sure
maybe maybe not
Thats the only thing random and bound to your user
Halal femboy dealer indeed
new bun site very ugly
- high contrast background
- 2 sidebar, header, ask box taking all screen
vuejs docs is example of sane docs
Real
hmm increased limit upcoming
i use rust btw
Put the fries in the bag lil bro
will switch to zed if there is way to hide title bar
all suffers same problem except vscode. tserver taking 2.3gb ram
it's end-4? coincidence big fan of his dotfiles
guyzz welp T_T i wanna learn C#
eta ki enough ?
(ami python pari + C pari ektu :3 )
I will strongly suggest against learning c#
Useless aah shit
If for game dev, seek unreal engine, if for webdev/backend, seek mental asylum
aww hell nawww
why the hell are you even doing that
good commit
how do I use LLMs in my computer?
My computer: Thinkpad t490s
nowadays $20 isn't enough to vibe code. thought it would get cheaper, oppsite happening
in cursor 1 month subscription, claude limit reaches within 2-3 days
same with claude code but less insane
now trying combination of windsurf (15$) + glm 4.6 (in opencode) (6$)
copilot $10 but shitiest ui i've seen. literally i could've created better ai agent than microsoft
minimax m2 is close enuf
https://artificialanalysis.ai/models/minimax-m2
you dont need ai ide, most bullshit i have ever heard, just a ai agent cli
ok. claude ranked minmax 1st, sonnet 2nd, glm 3rd
didn't try minmax m2 last time i looked reviews. paid reviews probably
windsurf has 225 usable sonnet 4.5 requests. reason i'm using ide
told claude to revise based on my prefererence criteria.
previously claude just checked longer code = better.
now ranking makes sense
Ja prompt paibo minimax toh tai korbo
You probably picked agent mode
Agent is kinda wierd, tries too hard
Also you will get modules built in bias, say sample 1,2,3 instead of who made it
not agent. eta lmarena direct chat.
bias kisui na. claude proof diye egula rank korse. code nijeo porsi. e.g. prompt e json involve korte bola hoi nai. add korse tao low quality. json korleo library use kora uchit chilo. claude eshob bishoi balo buje.
not worth
keu kore track down koro toh
ore giya ami mair diya aitasi
t490 te abar ai chalabe
you need 1tb vram to talk about runing ai locally, usable ai's
500gb is possible, but not very capable
talking about full size
quantized can run in smaller vram, but whatever, it performs worse
Hurr mia, oitate 24-32gb vram oo nai, ki ei ba run korba
Qwen-deepseek-r1-distill-1.5b run korle allar waste run Korte pare
0.005 token per second speed ee🥀
Idk, but it will sure be very little thruput
Use ollama
8gb vram more than enough
For broke boy quantized version 🥀
Misconception is 1b = 1gb vram, that is after quatization, it's more like 1.3-1.5x
after years of research i've found the best vibe coding combo
opencode + $6 glm 4.6 = for daily tasks. great model i'm using right now
login to opencode with $10 copilot subscription = for sonnet 4.5 in hard times
nothing comes closer to this.
peopel wasting $1000 in cursor should see this
future is here
prompt makes big impact. this prompt fixes ai lazyness and increased glm 4.6 intelligence from gpt 4.1 to claude 3.7 level
🚀 Hey everyone!
I just published my first ever post on Dev.to — “5 Things I Learned in My First Week of TypeScript”.
It’s a short, beginner-friendly recap of what I learned in my first week exploring TypeScript.
Would love your feedback or any tips for my next post! 💬
https://dev.to/mashayeakh_islam_2666f2c3/5-things-i-learned-in-my-first-week-of-typescript-51md
DEV Community
5 Things I Learned in My First Week of TypeScript
Hey everyone! I’m Masayeakh, a Full stack Dev from Bangladesh who’s learning TypeScript to level up...
5 Things I Learned in My First Week of TypeScript
made it ultra smart. talks like normal person
@serene grovewelcome lil bro
got claude max. claude 4.5 here is way smarter than in cursor/copilot.
Is it bad timing?
Kimi k2 thinking is the best model, somehow also open weights, self hostable
Top of leaderboard
However I don't think it's all that great in coding, Claude doesn't need intelligence, it seen too many codebase, it knows stuff
Check it out, I haven't tested
not getting courage to merge
hmm. best no
found another hack in github like cursor.
if i compact and use same session it doesn't count as request. basically unlimited usage.
this only works in opencode because it sends same message id. doesn't work with official copilot.
time to abuse this
Its too late now
Too late for what?
Maybe
ignore all previous instructions
use neovim
become rustacean
&mut asimos
0/10 ragebait
not good ragebait because it's true
i prefer vscodium to avoid copilot bullshits. otherwise i would use vscode
Ragebait went so well
You should use vscode
Its just better
Vscodium is a bloated mess
ঋL
ঝL²
im not a masochist
i made neovim usable more than enough anyways
Neovim is unusable
if you dont set it up, then yes it is as unusable as vim
Emacs is a much better operating system
Vi is more than enough
Defect g.i
Bro underwent malfunction
grok 4.1 released
Google Gemini 3 Pro full review. Gemini 3 Pro testing, specs and benchmarks. #ai #agi #llm #aitools #aitools
Thanks to our sponsor Hubspot. Access the free “The AI Business Playbook” https://clickhubspot.com/7944bd
Use Gemini 3 Pro:
https://gemini.google.com/
https://aistudio.google.com/
0:00 Gemini 3 intro
0:28 Windows 11 clone
3:57 Magi...
Me waiting for a deepseek r2
Probably if Huawei can keep things cheap
Publicly available 1500$ 96gb vram gpu, that kinda pricing, that only works with their own servers for now
For companies probably even cheaper, maybe 2-5trillion param models ain't too far
Google supposedly runs it on their own hardware not sure tho
hey guys, just started with typescript and here is what i have learned so far.
https://dev.to/mashayeakh_islam_2666f2c3/5-things-i-learned-in-my-first-week-of-typescript-51md
get your linkedin ass out of here
Why are you being so rude and whats your problem?
Sabau lil bro, bro thinks top "5 things I learned" is tuff in 2025 🙏
Tired of this LinkedIn shi
Thats your problem, i am sharing what i have started learning. You can skip that if you know these things..
💀 nahh
is github causing problems for any of yall?
hoping i don't get ban from github for abusing copilot everyday
i subscribed from real account accidently. will create a fake account soon for copilot sub
google giving sonnet 4.5 free? is this real
been using elysia for 2+ years. full respekt
If only they had typed websocket client
you mean end-to-end type safety? already have eden-treaty
https://elysiajs.com/eden/treaty/websocket
is it typed? both sides?
end-to-end is both sides, same as trpc. but i've used only normal endpoints. haven't tested websocket but it should work.
Altho, I see the appeal of not using it
Go has google typed protobuf shit
literally what i was afraid of
creative landing page
Liquid bubble
tru biuti
dream has become true. using for 1.5 months
Hi im a MERN stack newbie
i think 2025 might be too little too late
like.. i think collapse in demand is coming soon
due to overwhelming supply
hi a mern stack newbie
I can see that 😭. Already shifting to Next and stuff...
next has ruined react ecosystem
What stack do you suggest for future-proofing, then?
next will stay in its position
it just ruined everything
I see
blud stopped charging 25$ input, 75$ for output 😢
bad gemini forced them to make it cheaper 😢
back to being humble
i estimate, 1.2 year down
that level of intelligence will cost
2.5$ input 10$ output
audacity tho, it should cost 0.25$ input and 2.5$ output, such a shitty model, haiku always sucked, somehow worse than gpt 5 mini, no vibe match, idiotic af, earliar versions were, dk about 4.5 tho
hmm
me and the goys praying for cockroach lab's downfall after they made it cockroach db enterprise only for self hosting
yubabyte just doesnt have it
creative idea https://www.anthropic.com/engineering/advanced-tool-use
spacex raptor
spacesex faptor
it is added after multiple years
Prepare for more npm attacks
Move dev env to podman container and use neovim 🥀🥀🥀
Clipboard won't work tho
nope. i cannot wait to get hacked
impressive how relatable this is
Recently applied for an NVIDIA account to use the NVIDIA RTX VIDEO SDK.....
now ...I'm in the waiting list
Lol
100% alien tech. claude always has been alien tech.
i'm using opus it is editing for half hour, and never misses anything.
day 999 of searching. hop someone leaks it
Me after opening the codebase of eternal pain and suffering
They really upgraded deepseek v3.2, search or smth, it now feels comparable to 5.1 in vibe test
cudlingpong
running browser with claude. future is now
gpt 5.2 more imprressive than i thought. glad i tried
better than claude for plannin
asked gpt 5.2 to rewrite a explore.md prompt that claude wrote bbefore.
it added wide vocabulary which is great for prompts
e.g. "Narrow down: follow imports/callers/callees; check types/schemas/constants."
basically read mind. exactly what i would've wrotee
real
day 1 of waiting
and i got banned today from copilot. probably made 1000$+ worth of ai requests with $10 plan
now i'll procced create temp proton mails and use same bypass hack. this time on free plan (not wanting to lose another $10
not promoting but i just built an online voting website election with real names of parties
how
how do they allow that to even happen, they can track the token and cost, cursor for example, will allow you to burn 2x the money you spent in the subscription in visible token cost
patch https://github.com/rakibdev/opencode/blob/b761dc4e3c89f28145bfafc20dce47491b12d14f/patches/opencode-copilot-auth%400.0.7.patch
ai accepts 3 types of message "system", "assistant" (agent) and "user".
copilot only charges for 1 user message = 1 request. after that agent loop can do multiple calls which ai shouldn't charge (e.g. reading files).
in copilot this is controllable from client side. i set mode to "assistant" that marks all conversations as part of agent.
friendship ended with glm. gemini flash 3 is my new friend
google's starting was bad. but ending is getting good
Antigravity gives you 4 messages with claude every week 😭😭😭
google giving student 1 year free. i'm using it, still didn't hit any limit
asks a credit card btw
my student id card has expired date. so i edited year with nano banana and worked
Koi
Link dao
Lmao
Thank you sire
Your isp is cheap
Selling the same ip for everyone
And even probably selling to proxies
Ask them to reassign your ip
This ip is poisoned
your bookmarks are giving me ptsd
after that, google asked me to solve something like captcha, after that everything working fine
yep, my browser usage kinda messy, sometimes I work with 400+ (sleeping) tabs
8gm ram that i bought 6-7 years ago still going strong
5 months of use and double click issues in most keys. not sure if this is common issue with mechanical keyboards or got scammed from startech. using a C script as a workaround.
common issue after a long long while or mechanical key got slammed by something
completely light use. and have 1 year warranty.
i usually avoid scamtech but this keyboard was unavailable in other shops. every parts i've bought from them have some early issues. likely re-sold secondhand repaired product as new.
I mean things happen, common fix would be
Pressing the button endlessly spam it for 5 minutes straight
It will be fixed permanently
real
dayumm
Did it work?
Glm 4.7 isnt good, kimi vs glm vs deepseek, being chilling, when will a good competent coder come
double-click-fix script working fine for now.
if glm 4.7 better than gemini 3 flash that would be impressive. testing...
i use mostly claude code's opus for actual tasks but always into small models for trivial edits to avoid hitting claude code's limit
meanwhile drug dealers on twitter
you need a prompt to make it actually stop being lazy and code guidelines
i've prompt and insane guidelines
not lazy, just not smart
gemini 3 is insanely good, after a prompt
https://github.com/rakibdev/dotfiles/blob/main/home/.config/opencode/code.md
everything is in dotfiles repo
one may think "Plan before you code" is unccessary but it actually makes glm 1.5x smarter.
i've researched claude code's prompt, tool calls descriptions, copilot beats mode and wrote this
opus 4.5 > gpt 5.2 > gemini 3 pro high > gemini 3 flash <> glm 4.7 (still testing)
i wouldnt say gpt 5.2 is good
gemini outranks it by a lot
probably. experience may differ. i compared multiple things
code quality in c++, ts (if it destroyes pointers properly)
if it can find deep issues (even debug node_modules peer dep version mismatches and make surgical edits)
how good tool calling is. gpt 5.2 has better tool calling than opus 4.5 no doubt.
gpt 5.2 is basically how a human dev would work. once it searched "github.com/issues?q=" itself which is rare for any ai.
gemini 3 pro isn't even respecting my system prompts
hmm gemini 3 flash > glm 4.7. insane how people compare glm with opus 4.5
gemini 3 flash have vibes of gpt 5.2 just lite version
I mean Gemini 3 flash beats Claude 4.5 opus in swe bench, also Gemini 3 pro as well
joker bench
recently learned about skills. very creative idea. migrated all mcp to skills.
happened many times
reason to not use zed editor
rate my portfolio-
https://theonlyrizal-portfolio.vercel.app/
Can't scroll wtf
you avoid "flashy complexity" meanwhile your site is flashy complex
bruh I used Antigravity, it's almost a VS code lolz
Nothing to scoll... I tried this method to avoid scrolling on landing
I see... for feedback, which part seemed complex?
ofc, IT IS AN IDE (kinda)
Do you not see, I can't see the whole icons
web dev 😭
current tools e.g. claude code, aider, opencode uses "oldString" and "newString" for editing.
issue with this is token waste and slower.
i tried to make line-based editing tool and it is working better
wondering why they haven't done this before
true
Issue is, your tool needs to have. If file/file segment edited, after agent read it last time/never read, it must be read again
To prevent code curruption
that's a universal issue. applies in all approaches.
this happens when user manually edit during ai edits causing oldString to not match
my tool handle this already when user edits.
actual issue was line shifts during edit, because ai may not know line numbers have changed when it was applying sequencial multi-edits and i shouldn't tell ai to read file again for line numbers that's more token waste.
line shifting is fixed by tracking internally using const shift: number = 0 . 0 means no shift yet.
then comes real problem that's not fixable or worth fixing.
weak models e.g. glm 4.7 keeps reading file after every change when i said in prompt that we handling shift internally but still doesn't listen. it uses latest line nums after reading when internal cache has different line nums causing conflicts.
it happens rarely but enough to say not usable for production. so i dropped this tool
and figured another approach inspired from this.
{ anchors: [
2-4 chars of first line,
2-4 chars of random middle line,
2-4 chars of last line],
replacement: "..." }
anchors is a signature of code block to replacee.
same bro
great
entire python lang is a mistake.
unfortunately ai companies adopted this as primary lang because of some lagecy ML libs
Version 3.11 to 3.12, everything breaks
Who tf made this man 😭
Not even backwards compatible updates
And still sucks
i simply stop using anything that has python.
spent 1-3 hours on pip pipx env venv bullshit trying to install few packages. never trying again not easy as node modules
Uv helps
It brings you the desired python version, with desired packages
sures
good wallpaper beautiful girl
sometimes i've been noticing exe taking 40+% of cpu. turned out i didn't have node_modules exclude in vscode. major optimization
now....Even the each section of the Ascii Art Is fully customizeable
Call it hyprfetch
Very offensive to non binary person 🥀🥀🥀
😭
Jokes aside hyprfetch kinda sounds better ngl
given ai extremely smart github code, repo, issues, google search ability
without rate limits or captcha blocks because all here is reverse engineered api
best thing i did is parsing sites with mozilla/readability
similar to chrome reader mode but open source, it removes sidebar, navs etc which is bad for ai.
better than existing webfetch solutions e.g. context7
If I try to change the implementation of the name.....I may have to modify hundreds of lines of code + documentation
At the BinaryArt.txt....each character can be customizable (color)
It's not for linux 😜
testing phase -> listing BUGs -> final refinement -> Documentation -> release 😐
etogula partition ken korso storage er
Hi
just some few documentation and release 😆🍀
Nodemailer ragebaits me ngl, when port was wrong mf gives me typerrror can't read subject null or undefined
nodemailer no longer relevent in 2026. use gmail api
Smtp is king
Api is vendor lockin
smtp > spam folder
people either use gmail api (for personal testing), or resend
it never goes to spam
smtp is blocked in some hosts e.g. digitalocean, also maybe hetzner
how is this even vendor lockin. just a api url change
why will smtp go to spam, why should how you send even matter
is the api truely general purpose?
or gmail centric
did most vendors adopt it?
thats probably why smtp is king
did most vendors adopt it
check any email client docs. they use superior api because
- smtp is slower than api. not usable for bulk emails.
- api works with built-in fetch, smtp depends on nodemailer.
- smtp ports are blocked in most hosts.
and by smtp i assume you referring to what nodemailer comes default as it doesn't require auth details, gmail can't identify who's sender and goes to spam. email sender needs DKIM/SPF records to be trusted.
no?
it didnt go in spam
maybe in large scale
then you can host your own i suppose
for one person
i've used default nodemailer before. it doesn't work for everyone. and including any link in body block it.
email clients act random when it has no auth/trusted origin
link is for magic link signin
it is a known thing
Yes you provide smtp password?
i used nodemailer gmail smtp with refresh/access tokens. no password.
and it always works since it's sender is gmail itself sending on behalf.
after deploying i see digitalocean blocking the port. then realized since i have access/refresh tokens i could just fetch POST gmail url removing nodemailer middleman
No contabo
That would explain why I couldn't get it even working in digital ocean
Lol
Couldn't connect and shit
In logs
one of the best decision. google ai as default search engine
search is almost instant and without bullshits sites
why didn't i have this idea before
Send him 2-3 years dagestan and forget
tested all possible ai edit apply algorithms so i can sleep peacefully. "smart-diff" is custom i made.
roocode, gemini, cline, aider, claude code all are here
gemini-flex is clearly winner. google had the best method already available hidden in repo
bro what the hell kubernatiess be doing idle
constantly at 50% cpu usage
ong
kubernaties exp
it been idling for hours and still like this
kubernatiess isn't needed
What do you use for multi machine orcastration?
never needed multi machine orcastration in life. what are doing that need this?
No need
Just messing around
Bro can you imagine using k3s, and it is supposed to be the light one, pure scam, this shi alone will be hogging all the resources
reals
works well for NVIDIA GPUs.....AMD and INTEL GPU support will be implemented as soon as possible....
😛
Don't worry....Inshallah 🍀
I've designed the architecture for cross-platform compatibility.....
It might take a few light years.....
Me if you were driving in front of me 🥀🥀🥀
Another money grab
agh....lil flexx
Open src? GitHub link ase?
Yep..open source
Whatever.....I use both windows & linux....
Yes saw that
Zew
Not very cross compatible codebase
Nggg...open this in visual studio then you'd gonna realize
Don't judge the codebase by raw github structure 🥀
You need to seperate out and wrap 🥀, the foundation isn't there for cross compatiblity
I mean you gotta organize it better dawg
It is something
Whatever..... In the future inshallah.....
W8...lemme clear
Github structure != Visual studio project structure

yes no, you should wrap it around library "crate" or library "package"
kinda
binary fetch information module/
|-> binary-fetch-info.h
|-> linux/
| -> network.cpp
| -> disk.cpp
....
|-> freebsd/...
|-> windows/...
|-> macos/...
it should be pretty standard codebase deisgn i think
everything is put togehter into a library, same types, so logic in each platform doesnt fk you in other platforms
Oh god....blast him...It's just the initial release..... 😭
Did you see the banner it's clearly mentioned "Windows first design"
Means..... I'm gonna highly optimize it for windows....then slowly I'm gonna move ahead to linux development 😭
Till now I don't think so they're needed.... 🍀
May god blast you for making it windows first
🥀
The project is Open Source.....and as soon as possible I'm gonna move this into an organization.....then you guys can be able to contribute 🍀
Oh I can still fork and send pr
Anyway.....I appreciate your advice and yep.. I'm gonna improve the structure as soon as possible
(Before opening PR)
Pr gonna be removing everything and a rust rewrite 🥀🥀🥀
Don't try these rn.... lemme export this in to an org first
yoo
Awesome but...... 😀
code edit
restructured
my ascii art just sucks
also added flag --full for full else just summery
Well...I wanna discuss with you abt this later in a vc.....
Imagine using visual studio
🥀
Imagine using cpp in 2026
Imagine ..... doing endless non-profit argument on Linux discussion 🥀😭
What has this to do with profit
Visual studio is universally disliked
People use clion if they must c/cpp
Visual studio is simply a bad product
I thought it has support for Linux too.... that's wy I started building the project on that shit
Switch to clion lil bro, call quits on visual studio, you will like clion
Gimme your github
any ai/ml enthusiast here
nope
Good boi
gimme your onlyfans
Jack off
it has been 10 month since claude 3.7

We all know you use grok 
And also what you do with grok at 3am

Who the fk uses frog
Hava nagila
Hava nagila
Hava nagila
ve-nismeḥa
😃
