#Programming and Technology
1391 messages · Page 2 of 2 (latest)
yes!!!!
no!!
yo chat
3090
4070
7800xt
6900xt
if they were all at the same price
what would you get
5070
nah its out of my budget
all used cards btw
i don't play AAA titles
i wanted to learn to edit videos, using blender and AI
so thats why i didn't just pick amd in the first place
Then don’t buy a card at all
You should lease cloud computers from sites like lambda labs
how much are they
wdy guys things flutter or react native
personally im very confident with react so react native shouldn't be too hard to use but i also wanna learn flutter
native
never used flutter
try expo
300-450$
but incredibly sketch
sort high to low in graphics cards
and you'll find all of them
sketchy asf
ill try
but it'll be hard
because i broke my laptop and pc
and yk
i feel guilty
that lowk looks really good
@tidal plume enable image perms here
the 3d thing looks really cool
but you should prolly preventDefault() when the user is scrolling on the 3d part, since it scrolls both that and the page itself
What do you mean ?
.
like the Cosmic Creations thing
Ohh
when you scroll down while hovering on that, it scrolls the page as well
Yep I will work on it
😊 ty
Hello
Do you have projects too ?
I had no idea I would get so many upvotes on a reddit post about my repo
@wide lintel he dropped cs a month ago look at this
wdym
oh
😭
its still useful to
tho
like not for learning
but acedemically for universities
He doesn't do cs anymore dropped in A2 and look things that dude does
Ohh his is 3
Yes
am not to sure about univeristies
but if you enter as a cs major
dont you get put in foundqtiion year instead of first since u didnt take cs a level?
I don't think so
Coding is a soft skill
I know people from unis who took a major in Physics but are now working as software engineer
how's this landing page
Be an academic weapon!
I recommend dark mode
it just looks better
shittt its so nicee
great jobbb :>
samee!!
Pretty good, but the labels shouldn't be fixed or sticky
Was that a design choice
Also the bottom GIF seems a bit out of place with its width and height being 100%
And you can add some more design to the hamburger menu
I'm on phone
it's not a gif
🫤
I don't have any gifs on the site
lol you're missing out on the good experience
💔 will check it out after I get home
Oh it's just a large animation but the white background looks odd
Make it RESPONSIVE 😤 😤 👺 👺 👺
its u
smallbrownman has 170 rep
Alr daddy
hello fellow tech ppl
made this website as an alternative to pretty much every past paper website out there, but free and without ads, and a hell of a lot more features. For now includes some O and A Level subjects, but the way i have it setup it isnt that hard to add more subjects.
would love to get any feedback you can give me on what I can improve in terms of UX and all. It's still in a very early build!
Highlighting Features in no particular order
- Includes a great mcq solver, which allows you to solve past papers yearly and gives advice based on which topics you got questions wrong in
- A 'explain this question for me' button using LLMs and a lot of context, to make sure your explanation isnt wrong.
- Includes a yearly past paper automatic solver done through LLMs. You can annotate with your mouse or apple pencil or anything such and write over the paper, and it will automatically grade you and give feedback, and save persistently to your storage. it will also remember your scores and display them, for mcqs and structured papers, and how much of a paper you've completed. a little hard to explain through text, would recommend just trying it out through your tablet or whatever
- Also from my knowledge one of the largest topical banks for physics, chemistry, and biology.
it can also be installed as a PWA (progressive web app) on your phone, ipad, tablet, or desktop, and then be used like a regular app on your device!
is it open source
oh shit
this actually works
UI could be better but from a functionality standpoint this looks great
it works
I'm happy to work on it as well if you're looking for another dev
there are a few errors with the app
like this for example https://i.imgur.com/n8t6CEk.png
i hopefully want to, but the data is like 3 gigs
the frontend is just vue so i can probably get that out
you mean all the images
yeah
ye ye
a little messy rn, but 100% want to open source it all
if you want to contribute then i can probably just make the machine pull from the git repo, and if you want to help you can make some pulls which would be appreciated :)
are you using bootstrap and tailwind
yeah I'm good with that
I think you're getting 404 error because you're using the subject name instead of the code
I'll probably dm you because I can't attach images here
yep saw that, dumb mistake lmao
fixed
had to fix some other weird error but shouldddd be good
only issue i can now think of is possible abuse of the ai endpoints since i have a quota, but i think i have a temporary solution for that in mind
im using bootstrap but otherwise just custom css, and everything else is just vue (router, state)
this looks sick dude, m def gonna use this
ty!
That's cool, how'd you get the topical data? Are the LLMs custom trained or just GPT?
i gave it a knowledge base, but no i didnt fine tune it
that is so cool bruh
oh nice
if you want to do something similar, i'd recommend o4 mini, i used to use 2.0 flash / 2.0 flash thinking, but it was very inconsistent
Long time back I tried AI stuff, but without api's from gpt and all, tried using this thing called spacy in python
made it work, but training and fine tuning it got the best of me
spacy for data parsing?
13 yo me could not handle the smoke
yep
ncie
not if you know where to look 🤫
but if you use it through official apis, which i did for a while, i had to optimize my proooompting a lot, amount of tokens wasn't too much of a worry. if you look at o4 mini most of its major cost is output tokens rather input tokens.
if you're looking for a slightly cheaper alternative, 2.5 Flash but put a limit on thinking tokens to 1k
it cost me around $0.005~ per paper i proccessed through the official api for o4 mini
which isnt that bad at all
it could get more expensive to around a cent though depending on the amount of context i needed to input and the image context
too be fair you were working in like the stone age of LLMs, so understandable
fine tuneing is a whole lot easier now
you can literally fine tune a LLM on a discord chat for a few cents today
yeah, ill dabble in AI once again soon
OH
but dont use o4 mini high whatever you do
you will cost yourself a fortune..
so youre using a public api key or something?
no of course not
shouldn't hurt individuals over this
i had a subscription which had unlimited o4 mini requests, so just used that before it ran out
hahahahaha
all you're going to do is abuse google ai studio
reporting you to google rn
google ai studio 
i do have a openrouter api key connected as well for gemini 2.5 flash as well
but haven't had time to connect it up since yk
caies ✨
literally made this website in mocks
damn youve spread it over a year
did o levels in mj 24
oh
im now in AS
valid
when you need free ai integration on your website
all you do is slap gemini
brackeys is back
Frfr
Gemini and groq have probably faced more than a hundred dollars in compute from my usage
use cerebras
thats what i've been checking out now
i already use cerebras on my site as well, just waiting for a vision model from them to fully convert over
bro it's been ages since
I got cerebras access two weeks ago
Don’t get me wrong - the inference is great
But its tool use is exceedingly buggy
buggy how?
haven't used tool use yet, but its inference has been pretty great for me
It’d just throw a “tool call failed” error 80% of the time
you sure you weren't just using a dumb model?
did you try the new llama maveric model
maybe give llama 4 maveric a shot? though tbh scout should work fine
did you ask them what they think
they have a discord
haven't gotten to doing it
aight i'll try that
@lucid ember ```json
error: {
message: "Failed to generate tool_calls. Please adjust your prompt. See 'failed_generation' for more details. Exception: 'arguments'",
type: 'invalid_request_error',
param: '',
code: 'tool_use_failed',
failed_generation: '{"name": "shell", "parameters": {"command": ["ls", "-l"], "workdir": "/path/to/folder", "timeout": 10000}}'
},
the model uses parameters instead of arguments and i can't really fix that
also the workdir is /path/to/folder, which is kinda stupid considering this is llama 4 scout
they don't have maverick yet
yikes
found something really fun for ai studio piggy backers, if you allow data sharing with grok you get $150 worth of api credits every month, and grok 3 mini is currently the 3rd best model out there now
great day for the unemployed
oo damn
you need to pay at least $5, but after that everything upto $150 a month is free
yeahh i just checked that
hmm we can't send images in this chat
Only teams that have spent at least $5 are eligible to enroll in this program.
exactly
they're so desperate for training data 
yep thats what i've done
thankfully have a lovely paypal and card so just paid it
so worth it for $150 of grok 3 mini
$150 is $150
and im going to abuse the hell out of it while i still can
they even support function/tool calling
only issue is image proccessing, but you can just use google gemini to describe the image for grok
tool calling when thinking too?
will be great for those who want to solve math / physics or anything technical
because flash thinking is being removed soon
and flash is pretty bad at that
man i'm desperately waiting until either groq or cerebras add qwen3
lemme see
cerebras is not adding tis bro
they are so slow
and groq is expensive af
well not that expensive but for the quality of model.. yeah
they still have llama 3.1 
groq has very, very limiting rate limits
whats the rps?
per minute
lemme check wait
you've been checking for quite a while
Ah yes I forgot :)
don't you think people should be able to send images here
Ah that reminds me
I think I replied in off topic
wha
I need a good 2 in 1 laptop for studies
@empty torrent you might find this cool
it’s kind of funny tbh
interesting
nice hack
just waiting for twink Altman to bomb me
there was actually surprisingly a lot going on here (codex)
but until then I’m using this to its fullest extent
Perhaps do it in a more legit way - gemini and cerebras do have generous free tiers
neither can be used very significantly via api
woah, so no api key to use code to access gpt?
ye but you need ChatGPT plus or pro
thanks man
ty
wtf 491 stars is crazy
people love free stuff
and surprisingly was the first to get on this
nice job lol
guys
is there a way to use the esp32 c board to turn music into vibrations but like any song instead of coding each vibration separately for each song
I don't know the nuance but you'll probably use FFT and pick out the lower frequencies
artificial super inteligence is gonna kill us all before nuclear can do that
woah
real tho
humans have been working for ages to develop nuclear power
but at the end of the day
it's just heating of water
🤦♂️
i justfelt like putting reactions idk
does anyone here use a raspberry pi
why not both?
both :3
been getting into frontend development for fun
i only know basic html and css rn
made this in my free time
anyone who can review it or give any sorta feedback?
oh yeah the footer part btw where theres like 4 columns of "Get to know us" i didnt bother editing it i was too lazy
heres the github repo
old but ye
which model
rp5
runs all my sites and tools
pretty nifty little machine
nice
i have the rp3
it used to be my dads but he barely used it
mine isnt powerful enough to run the windows software but can still run quite a bit of stuff
ive yet to make more projects with it tho
so far ive only made a blinking led 😭
why are you trying to run windows softw?
are you using it as a main machine?
ah well windows itself is pretty bloated so its not that great of a benchmark
if you have it run linux its a great little guy
at least on my rp5 have been able to run ml models locally with great inference speeds
thats awesome
i connected mien with my laptop and did the screening thing os i could code on it
i got this little camera module i still haven't had the opportunity of playing around with :(
you should get a screen
i dont have it personally but have heard of so many cool projects ppl have made with em
want to get one the next chance i get
i have one
it gets heated up rlly fast tho
and when it heats up it starts lagging
i also wanna get a camera module
aw fair enough
you should try and get the pi 5 then
its a beast, albeit over priced
but its cool overall i have these glass borders i use to connect on my raspberry pi with the screen and keep it intact
maybe but ive gotta get good with the pi 3 first
i only know the basics
e
hi
Hiii!
I'm working on a text based visual novel(?) i was experimenting with base printing styles to use in it, and what better way was there than to print out my poem on fish going insane?
woahh
lmaoo nice
ehhehe thank youu :3
i currently only have the typing thing left, aftet that I can start working on other mechanics such as jumping forward, backward, etc
oo damn atbb
:D
can i ask airbus to make me an A380
sir are you okay?
22 KB, bloatware!
nooo ? sob
write it in assembly to save 2 MiB RAM and 1 KB storage
💀
yeah right buddy
I ain't touching assembly with a 20ft long pole with a gun to my head
I mean you can put it in your cv...?
and they wont give you the job bc you wrote it in assembly, they will because they know you can stay up for phone calls at 3AM
what even is that
ptsd
also, put it into github, why are you sending a whole source file into discord
because I don't like github
fair point
ikr?
😔
