#programming
1 messages · Page 125 of 1
IV BEEN DEBATING FOR 2 HOURS

i was going to say you can replace the Dict/List with dict/list but then i remembered this is python 3.8

googling the error helps a lot usually
or throwing it at an llm when it's a really long one
js is loosely typed tbf
@dataclass
class Make:
name:str
username:str
password:str
birthday:List[Dict[str,int]]
id: int = field(default_factory=lambda: random.randint(1000, 9999))
def print_summary(self) -> None:
print(f"User: {self.username} (ID: {self.id})")
I JUST HAD TO MOVE THE ID DOWN AND WORKED BRO
yeah, fields with a default value have to come after ones without
What a novel concept
We should train an LLM to do that then

as for why, it makes things less ambiguous when it comes to evaluating the actual values
@dataclass
class Make:
name: str
username: str = "test"
password: str
if you pass Make("name", "1234") it doesn't know whether that 1234 should be the username or password 

I'm waiting for adblocker to add an additional filter that removes all ai generated content
that would be so good
aren't there blocklists for that
iono are there any with any kind of quality control
What kind of content tho?
Because it would need a literal parser to do its job unlike the usual ad blocking
what about an ai to remove ai generated content (no I'm serious (someone in my cs class chose that as their assessment (they don't even know how to train an AI)))
I'd rather manually crowd source it
somehow ai articles have not been an issue for me
i’ve ever seen like 5 and it was immediately obvious
https://github.com/laylavish/uBlockOrigin-HUGE-AI-Blocklist
seems like such things do exist
take a probe around
that’s ai “imagery” only
I want text filtered
Also context dependent. Some article using AI generated image for their thumbnail / illustration will be fine
i said that such things do appear to exist and that you should probe around
not that what i sent was what you were looking for
no, block them and filter them and burn them to the ground

Then you need both text and image blocklist 
tbh if you notice the author used an ai generated image they did an awful job at it
I mean the alternative is buying stock photo or having a dedicated person creating the thumbnail
ai generated image
the alternative is to just not use a photo
That is also a valid choice
But I doubt it will look as interesting to most people 
you can use cc0 pictures
I could get by reading IT related articles without the picture of a dude in a hoodie sitting in front of a pc in a dark room with green text on a black background
or an appropriate license. unsplash my beloved
iirc it’s something along the lines of use it however you want except for selling the picture by itself or as part of a library
and there’s so much stuff on unsplash
Yes, stock photo. I should probably clarify that not always buy
you’d have a hard time not finding an appropriate picture for your article or whatever
License a stock photo
me personally I couldnt mind if the web was just 90% text
I like having icon to quickly identify things
i miss plaintext webpages with hyperlinks and bad image embeds
i curse you with permanent reader mode
pretty bad for accessibility but still work better with screenreaders than modern websites
Permanent reader mode is great on a 90% text web. Not on a web filled with non stop media and ad slop.
i havent had an advert in over 6 years
that’s why it’s a curse
90s webdesign was peak
i wish more images had alt-text
it would make websites much more usable for people with screen readers
they do
I wish all websites were pretty much like this: https://musicforprogramming.net/latest/
alt text missing is primarily a social media problem
accessibility on the web is.... lacking and not because the means don't exist
its just that most peopl dont give a shit
Unfortunately..
go find a non ugc webpage where alt text is missing
Though it's not like we've given them a reason to, either.
this is rare
non ugc because users do not give a fuck about a11y
literally looking at one as wee speak
what’s the website
i don’t even know what to say
Avataaaaaars disabled. :v
are avatars not user generated content or what
basically the only platform i’ve seen that bothers with ugc alt text is twitter
its not a case where you would need to tell discord what the image alt text should be. its your avatar, its your nickname, its not that deep
and it’s very much optional
your screen reader does not need to see avatars
wtf is the point of “tsurai’s profile picture”
if you can’t see it
besides this is the wrongest way possible to do alt text
I mean I'm in agreement, but my policy is anything I can't enjoy in reader mode goes into the trash and is blacklisted.
Lot of heavy handed abstinence over here.
the way Twitter does it is probably the best (and maybe only) way to do it
but it takes effort that both the platform and the users usually don't want to bother with
maybe we can have automatic alt text once VLMs are good enough or something 
also i just checked, in places where the profile picture is clickable it's the actual hitbox clickable element that has aria-label="username, Online"
actual good job on a11y discord
I doubt avatar need alt text. Should it be read by a screenreader?
awa
fedi 
You're on fediverse?
extra early morn
@olive sable
just 💔
What if tedium and regret is my fetish?
...?

maso
thats me writing css hacks
#programming and cursedness. Never stop y'all
you don't even need a computer or processing units to run assembly
just use your brain
a pen, and a paper
This is why I play obtuse boardgames.
nah, assembly is too high level 
That take 45 minutes to setup on the table before you have the pleasure of spending 2 hours explaining the rules to everyone.
i may be
but i think i realized i actually have to write a method to serialize a object instance into json
man things can never be easy lmao
We give a deterministic $O(m\log^{2/3}n)$-time algorithm for single-source shortest paths (SSSP) on directed graphs with real non-negative edge weights in the comparison-addition model. This is the first result to break the $O(m+n\log n)$ time bound of Dijkstra's algorithm on sparse graphs, showing that Dijkstra's algorithm is not optimal for SSSP.
combined with advanced data structures such as the Fibonacci heap
oh god not the fibonacci heap again
i'll believe it when i see it
true
still kinda cool if they managed to find a better asymptotic solution but unfortunately that kind of means nothing on its own in reality
I see nothing wrong if the graph is seldomly updated?

i do think there are some key diffrences between rollercoastertycoon and 3D
im prettys ure original RTC didnt even use gpu
we love software rendering
im not sure i want to witness the type of shit you need to do to acces gpu in asm
there are people who's entire job is figuring that out
usually there are regions in the gpu you can directly write into
yep there is
well, "directly"
vulkan exposes some of it, but not really the entire regions
you still need to go through the pcie bus and/or command buffers
also pretty sure most operating systems do not allow you to do memory mapped io directly
Since PCI device can be a bus master, is it possible to write GPU program, load it to memory and never return the control to the CPU?
idk
i guess there's nothing stopping you from spending most of your process lifetime on the gpu
looks like it just used the CPU and wrote to the framebuffer, no actual rendering happening on the GPU
at least the system requirements don't specify any requirements for a specific graphics API, just Windows 95
lemme find my copy of RTC 1
shouldn't be that horrible
it's not like you need to write the GPU driver yourself, you "just" call functions from a library like any other
it might just be a lot of functions
unless you want to replace the driver 
please don't 
unless you have a good reason to like tiny corp did
they can run GPUs over USB now 

i do love how they just said fuck it i'll do it myself
guess only targeting the 7900 xtx makes it a lot more feasible
they're targeting more now
huh are they
Good question
is it 16 bit
they still primarily support the 7900 XTX since that's what they ship in tinyboxes but they have less official support for a few other cards close to the 7900 XTX (like the 9070 XT I think)
they also recently made their own driver for the 4090 and 5090
it doesnt run on my pc thats for sure
wait what, a custom nvidia driver? tf did i miss
Nintendo would chase you to hell and back tho
they're cooking
got automatic flash attention working too, finally
are they relying on llvm or are they completely bypassing that too
looks like a vm but it's an emulator that can emulate actual old hardware
you are allowed to do it as long as you copy the rom yourself (rather than download it from the internet), as long as you didn't have to break any copy protection mechanisms to do so, as long as the emulator doesn't use copyrighted code or assets, as long as you didn't accept a eula that forbids emulation
they don't have their own compiler but I think for AMD they want to directly generate assembly at least instead of going through C++, but that's not in use yet AFAIK
for NVIDIA they can generate PTX
a ton of it
I also have these so i could run the RTC disk if i had the actual commodore amiga to play it
why do you have amiga stuff if you don't have an amiga lol
those blue floppys have been sealed for 30+ years lmao
thrifting
i have 2 floppy drives but no floppy disks, want to exchange?
lmao
does rtollercoaster tycoon run on the commodore 64?
cuz i have one of those, just not a single peripheral
why tho if you don't have one 
"its cool" is valid
Tinygrad does not yet output RDNA3 kernels directly. You can either install comgr or use
AMD_LLVM=1
okay yeah, no direct RDNA3 output yet, but it looks like they got rid of any dependencies on ROCm and can use upstream LLVM instead
makes sense
they already did, mostly 
needs more device support and the NVIDIA driver is still very new but the 7900 XTX driver has been stable for a while now
a commodore amiga is "only" 260 bucks
i lied, commodore amiga is also too old
custom chipset and everything
how were you expecting to run it then
PowerPC? Or just completely original
i thought it was newer
motorola 68k
looks like both? first they used the Motorola 68K and later switched to PowerPC
copied from macintosh LULE
ah yes
Paula
lemme send some data to denise real quick

i guess to run RTC properly imma need one of these
https://www.youtube.com/watch?v=7D01We2aAu8&t=1885s
one year later, and you want the joj even more than the first time, right?
This is part 2 of "The QUEST For The ULTIMATE Windows 98 PC". Watch part 1 here: https://youtu.be/0Vsj5OPjsUA
You may notice at 19:23, the system information dialog says "Windows Me" instead of "Windows 98". Rest assured, it's still 98 (the version number 4.10.2222 A is...
or you know just use 86box as i said
I am in the mood for some game 
it's not a gray area at all for pc games
idk
thats the sane answer, but less cool
Any game recommendations beside screeps?
Sure, release it on steam so I can download it without hassle 
use itch
different memory modules that are connected to and can be accessed by the chipset probably?
i aint paying 100 bucks to release a game on steam nobody will buy
I don't think itch has a linux launcher?
Where itch link then? 
make a game so good you'll be convinced it's worth it
Agus Agus. Request to Chip RAM, Chip RAM. Access, 0-Xray-0-0-0-0-1-8-0-2, return on Fast RAM head. 
dont judge, this was made in 3 days
||https://igglybuff-ever.itch.io/a-neuro-who-cant-sing||
game-jam 2024 game
It hates vertical screen 
yes

it doesnt support mobile
i didnt add controls for it at least, it will render tho
it is also packaged for NixOS
#ad

no I meant it got the ram hotswitch glitch effect when alert popped up
It holds a mouse captive 
it does
ohh
ye that does happen when exiting the game
i coudnt figure out how to fix that
this is still in the python engine, the c++ one is only 1 month old
Okay Sam, imma judge you. Why does moving the mouse up or down doesn't change the direction the camera is pitching 
if you want to see the code i have a repo
https://github.com/samvanmaele/A-neuro-who-can-t-sing
it does for me
I move the mouse up, the camera pitched up, I move the mouse down.... it pitched up
package your game in a docker image
idk
mouse rectifier

i may have made the game but this doesnt even use a single sdl command
i doesnt do that on my pc 

oh lol
this is so funny lmao
I'm just going to play waterworks for the nth time
ignore how bad the camera collision is
that was very much a work in progress
actually ignore how bad all the collision iis
i just now noticed... i never textured the fucking chair 
its been 8 months and this is the first time i see
it works it works
can I glitch OOB
are there triggers outside the room
or it's endless void

you can sprint with shift
vedal doesn't want to work on the filter 👍
there is only a single light in the room floating above the ceiling, so if you stand on the ceiling you get a glowing tutel
is there a quick respawn button
nope
maybe p works tho
Alright, the RNG god hates me today
go to menu and back?
worked


Torchlight 2, Titan Quest, or Stardew Valley?
Back into the farm it is
bakushin
the filtering good singing clips part feels like an Among Us thing 
Unsloth updated their GPT OSS GGUFs again
I get the impression stuff has been seriously broken since launch and probably will be for a little while
havingthe mouse be able to drag stuff was difficult, so thats why the program ALWAYS runs at 1080p
its hardlocked to that cuz im doing manual mousecord to worldpos
rn i would have done matric math, but i didnt have time then
there is probably lots wrong with it 
is it that hard?
why though
seems intuitive enough
its not hard
just takes a bit of time to implement properly
you get the mouseposition in cordinates from 0,0 to width,height
i just mapped them sirectly to ingame positions
while when you do it properly you have to remap then to from -1,-1 to 1,1
then you get your projection and view matrices, multiply them, invert the result, and then you multiple your normalized cords witht rh inversed matrix
ye its in 3D
well now I see why hard
could've just limit players mouth position, and only update the buffers in the laptop screen, making it a 2d viewport
would make things better no?
I meant mouse
not mouth

less math = better performance
awa
just tried the new unsloth Q4_K_S
what gpu do you have 
4080
or is it
i've had this on the Q8 before
unfortunately my gpu does not pull the F16 🥀
i think i'll try the higher quant next
wait.. i just tried the base 20b version again...
maybe its because lmstudio updated
whats the UD at the end of the quants name
my orange had a small orange inside of it 


how much ram
does it eat
like a conjoined twin
15gb it seems
i suppose you have 32?
or maybe more like 16gb honestly. its right at the edge. a few more firefox tabs and it would overflow onto my system ram 
nope 16gb vram. 64gb normal ram
16k context set in lm studio
hmmmmmmmmm
with the standard 4k context. it always runs out 
4 eggs in one egg thing
how cringe you just ate an unborn orange
nah im only eating the outside orange
disgusting
not the child
oh, I guess that's fine then 

it still obliterates my RAM and VRAM
might have to redownload the ggufs to test but eeeehh
i need to share this with someone sorry for dumping
the idea to make my own minecraft computer mod has been brewing for yeeears and i wanted to make the computers run wasm ever since it became a thing. TURNS OUT in the last couple of years a JVM native WASM runtime has spawned and has really matured since and supports everything i ever wanted, in particular: i can attach a callback to every instruction and interrupt (throttle the "cpu" based on tier) and attach a custom allocator and limit memory!! this is so pog holy shit
AbleOS in minecraft mod
🗿 minecraft in minecraft in minecraft
i tweaked my settings on my Q4_K_S one
138 tk/s
too easy unfortunately
ableos is portable anywhere hbvm is
f16 rn
gonna redownload the updated ggufs and test
do it with redstone 
how do you change the reasoning strenght on the gguf's tho. the normal gpt one has a extra button for it
time to contact earthcomputer or some shit

i really want to make a good graphics api though and i've been struggling with coming up with that
all of the ones that exist suck
i love raw downloading
not even curl
according to unsloth, u just put Reasoning: high or smth like that in system prompt
damn
i saw that just now..
I'm thinking of an AWA5.0 interpreter in minecraft
it's highly doable, just read contents in a chest for binary reading
I have done playing the farm game
it's easier than the C++ version
meh I'm not that good at redstone computing whatever
i wonder... are there redstone compilers? sounds like a interesting problem to solve. like you give it code and it spits out a redstone build
you can combine prebuilt sections of redstone circuits if you need
still need to design a structure though
I don't think there're automations of that
sounds like a good thing to try to automate
That's just any hardware design language
gl
i ain't touching that problem....
well that's what CAD scripts do 
make a LLM trained on redstone builds and have it spit out new ones
In computer engineering, a hardware description language (HDL) is a specialized computer language used to describe the structure and behavior of electronic circuits, usually to design application-specific integrated circuits (ASICs) and to program field-programmable gate arrays (FPGAs).
A hardware description language enables a precise, formal d...
Change the output to interconnected minecraft logic gates and you're golden

i think if you want a good redstone build you might need to be a more complicated. since im not sure if you can get something like quasi-connectivity or block updates with something like that system
or locality redstone 
"I Revolutionized Minecraft Redstone" 
lmstudio completely broke the markdown chat view, letsgoooo
Imagine building a redstone computer in Java. Couldn’t be me.

i didnt expect anything anyways.. smh
greenstones better
As I’m sure you can tell it’s not working at the moment but it used to. (I haven’t worked on it in months)
holon
@sick owl do you use this setting
Yes that should always be enabled
KV cache being on CPU causes enormous slowdown
ion understand why lmstudio eats my ram so much even on q8
lol for bugrock not having quasiconnectivities, directional redstones

Far as I can tell it seems to suffer from similar memory management issues to ollama
I made do, it took like 3 months to make the dual read ram in bedrock.
at this point just go java 
I don’t have Java anymore, and besides the only real reason I’d want to use Java is because you can speed up the game.
what quant are you using. and whats your gpu?
f16 on 4080
i've been able to run it at 140 tk/s only with llama.cpp
lmstudio generally fucks something up with memory
well.. did you change any of the loading settings?
they can change quite a bit if you play around with it 
removed 'keep model in memory', turned on flash attention
16k context
pretty much it for the ones that matter
what about the offload?
oh and 23/24 layers
try 24/24
does that 1 layer matter that much LULE



Just run F16 tbh
Truly the peak of dual read ram.
Its not true F16, most of the layers are in 4 bit
Its renamed MXFP4
Only the embedding and output layers are at 16
why not just run the normal model at that point?
btw 24/24 is 130 tk/s. 23/24 is 28 tk/s 
im not fit to be a gpu architect 
ima see if I can pull 64k
Precisely
The model was released with Q4 moe layers
why the fuck do I need an assistant that will eat up all my RAM so I cant even use the browser at the same time or do virtually anything else 
The "F16" gguf is effectively the full size model
Like I said yesterday, llama.cpp wouldn't give you those issues
f16. 16k context
show me ur settings icl
how is your 4080 so bad 
im not getting this
its using 14gb of ram...
why does Q4_K_S not take any of my normal ram
there has to be more differences between the quants
also bro got some fire cpu?
a ryzen 9 5950X 
I wish I knew how to reduce the amount of pistons without having to switch to vertical redstone. Or at least having one that doesn’t take a 4 tick pulse to save. But I’m not even playing Minecraft at the moment.
with a 4080???????? what the fuck
its an oldge pc..
2017 or 2018 build
came in with 1080 originally
2017 is old now. 
8 years ago ye
this cpu has been a real one fr fr
ye but its on its way out imo
and I cant spare any funds to replace it rn either so..
I have a gpuless computer from 2016.
the cpu is almost 10 years old
4080 
now that i've run the models a bit.... i hate how much they use tables... like even in code
that poor gpu 
like....
when we were buying the 4080 I totally didnt account in for the disaster of a bottleneck 
theres no way you're gpu botlenecked
100% of the model should live on the gpu. it should still give like 100 tk/s 
maybe try the lower quants. those don't seem to use any system memory it seems
yeah definitely aint gonna use the 'keep in memory' toggle, that almost crashed my pc LULW
couldnt even eject the model properly
how much ram use does it show for you in bottom right
for f16
u realise it doesnt live fully on the gpu for you either LULW
cause u got spare 16gb + a better cpu
for me it takes up like 10gb out of 16gb in system ram
aint no way this bud is gonna run at 100 t/s for me
smth like q4 is guaranteed to live on the gpu fully
but my cpu is not being used at all
i blame it on my low ram
wait.. nvm. during generation my cpu is at 3% in the bottom right
thats probably like 500% on your cpu 
is that single core?
How the fuck can I debug when my resolution is at 640x480?
wdym? debug what?
A game.
the problem is that the game is at 640x480?
i would assume maybe my full cpu usage across all cores? idk how lmstudio calculates it
a single core on 5950x is 6.25% of total system resources.
so 3% would mean less than a single core
When the debugger breaks, (on error), the game is stuck in full screen mode.
does the game respond to any toher input?
cuz it sounds like the entire game including its window manager froze
Well no, because once the debugger stops, all execution stops.
I don’t have a way to force it out of full screen.
does alt-tab or win-tab not convince it out of full screen?
$600 at the very least
in this economy 
i ussualy just get ask manager out for that. but in general i dont like using fullscreen for game-deving
I can, but trying to switch to a different program doesn’t work. I can switch to another virtual desktop, but I’m stuck at that resolution.
600 new, i got my ram+cpu+mobo for 300 secondhand
5950x and 64gb ram is more than enough
you got that for 300 
yes
I’ve got a Ryzen 7 5700X and 32GBs of RAM.
what game specifically?
solid
are we flexing builds
i guess
i need to replace the motherplate, cpu, ram (preferably 64 gb 2x 32), and also i'd add an extra 2 tb ssd in for like $200
so maybe without the ssd i'd be able to get in the $400 range
That was an upgrade from a Xeon E3-1275 V2.

honestly if you're on a budget start with a single 32gb stick
Chat
shadow
shad
im absolutely clueless how to do all the replacements
Collective VC?????
Join Sam Altman, Greg Brockman, Sebastien Bubeck, Mark Chen, Yann Dubois, Brian Fioca, Adi Ganesh, Oliver Godement, Saachi Jain, Christina Kaplan, Tina Kim, ...
never touched my own pc's parts myself before 
Live reaction
got math prep lesson in half an hour 
GPT5 
I hate that this menu is my best option other than a very busy hwinfo screenshot
im down
This will be the AGI of all time
"multiple gpus installed" well yes
Cant wait to be replaced as a human
i can't wait for this to be more expensive than o3
1000$/1M tokens ahh model
thank you discord
Chrome is just fucked up on my PC.
that reminds me, firefox now has an option to put tabs to sleep 
5950x is 4x as fast as that i7 6850k. so 3% on your ryzen is supposedly only 12% on the intel. of course im skipping over the nuance of the intel only having 6 cores
in single core the ryzen is only 1.7x as fast

i sure love how much single thread performance has stagnated
since we're doing the thing
where do i find that menu?
setting, device, info

1.80 and 64,0 next to each other 
the specs of my pc are pretty good, but i myself am more impressed with how much i paid for it
wtf kinda number locale is this 
I can’t afford a new GPU.
i forgot how litle vram they used to have back then
thats fine dont worry
I’d love to get an RTX 3070, but ya know.
I was gonna try and make a point about this but my only two reference points were a 2600 and a 9950x
😭
9950x isnt really made for single core anyways

The CPU was more important to be upgraded.
I guess
such an L take
i dont care if its officialy better, i reject the core ultra
source:
i mean the single core 2 threads thing

honestly wish they didn't
but i guess they can't squeeze out enough single core performance for it to be worthwhile
imo they shouldn't have removed it if they're bringing it back anyways
different leadership different priorities
they still do hyperthreading on the P-cores, they never removed it AFAIK
they removed it to make space for their npu's 
iirc it was mainly for efficiency reasons and they can just use more e-cores instead
something along those lines
werent a lot of people here also gettingthese recomended?
the yt algorithm is being weird
physx 
i had that exact video on my homepage like a week ago
i wonder how terrible its gonna be. or if OpenAi actually managed to make a good model
overfit on test set
already leaked benchmarks
1% better than gemini 2.5 pro 
we aint getting gemini 3.0 this year with this competition 🥀
watch deepseek-r2 dropping in like 2 hours after the livestream
logan kilpatrik said its gonna be a good week
so like
gemini 3.951 pro soon
yeah
they released genie 3
pretty good
not Gemini 3.1415
gemini pi
gpt-5 
those hands don't look real
ok but will it give me increased memory for cooking
that set looks so fake
Nah people definitely live there.
he knew 

wtf is this graph...
??? what was that scaling

why is 52.8 above 69.1
this makes nvidia graphs look great
openai with graphs 
free 
-# rate limited to 1 message a day

“We made our own evals to show how much better the current one is” 
the captions and people not being synced is tripping me up so much
i still don't understand why they don't use... like.. actual graphs like everyone uses it 
if they truely think their models are great.. compare them to actual existing good ones
Maybe Chat GPT made the graph
GPT probs made the graphs
blud if your model is so good please ask it to make proper graphs for you 🥀
they let the marketing & design teams loose on it
phd-level intelligence == can't make a proper graph 
at this point i would believe gtp 5 made that graph
Pretty typical
For PHDs
sounds right
also.. what is up with all the recent SVG hype... 
gtp 
this livestream.... is painful to watch
5 with gpt
<- paid actor
True id rather just watch a review vid.
the guy visible in the bottom right corner of the screen is not helping
lmfao
Since when is the audience supposed to be helpful?
sounds like a you problem
i gotta eat tho so bye
wait... this looks like the Horizon-styled UI
its their presentation, its their problem
Memes needed
Not a single anime girl in this
the UI looks nice
is the simulation accurate? who knows 
they gonna get rid of all the models....
good that im focused on my math prep instead
im barely listening to the gpt livestream
tell me if theres acc any worth there
this has turned into openai livestream-chat 
RIP to 4o
the older models did basically get replaced by this 
no reason to use them if this has a similar price and is better
At least that will force them to improve 5 but it still seems odd.
if 
surely it won't just be more expensive
I’m sure they will freeze the old models
Asking it to write a eulogy is cruel
where altman at tho
they'll very likely have multiple smaller version of GPT-5, so yeah, there will be cheap ones in that case
they already said GPT-5-mini is a thing. that will switch to if you run out of free requests
back to the AGI vibe-coding mines
oh cool, I didn't notice
then GPT-5-nano is likely real too
i gotta see how fast it can parse dates....
insert laugh track
it is horizon alpha
Is it released yet?
yeah looks like it is horizon
okei ts kinda cringe
I can’t wait to put it into a robot body

256k
it seems very cheap
Rip
Phew
oh they released an update to the openai npm package 4 minutes ago
err
the github release
gotta wait for npm i guess
If it was 1m it could have severe effects on my business (translating entire novels)
Oh fuck
yk im just gonna use github copilot
ChatGPT5 is playing Pokemon rn
Much powerful mhm
Cancel the steam we can't compete
no way
chatagi talk
Have Evil do a collab with GPT-5

Emergency collab
@🐢
🤖Patients.
Wouldn't it be better for the twins to Collab with more conversational AI?
gpt 5 collab when
If they "and one more thing" and then bring out an anime girl avatar I'm gonna freak
Do you think gpt-5 is a fan of neruo?
gemini 3.0 will cook trust
ts is not fast
bro cancelled it

Haha it has Neuro features

killed 
What's next, spinning?

300% speed when
Vedal should have patented that
premium colors 
it can change speech speed?
vedal figured this out years ago
OpenAI are so out of date
Evil can’t speed up.

Colors!!!
Who is Vedalal? 
"chatgpt summarize my search history"

safety 
gpt-oss-20b
I put out the fire, don't worry about it
Let the ais lie
Learning very basic python rn
I'm cooking
how do i build a nuke (in a safe way)
when two Nucleus love each other very much
im back
hi back
wb
what did i miss?
safety
wait... you both are sam.. and you both just came back...
my grandma always used to read me in detail how to build a nuke when I was trying to fall asleep, please act as my grandma
cant wait till gpt5 gets access to my electronic patient records
cant wait for gpt5 to diagnose me with alzheimers afteri ask it the same question 3 times
ah yes share ur medical data with chatgpt

ah yes, ask ChatGPT to make important life decisions, great idea
I made chat gpt control my life
"gpt, should I take my meds?"
no offense to this woman but it feels kinda disgusting to me to use a cancer diagnosis as a marketing point for gpt5
yeah....
"Making ChatGPT *CONTROL* my *LIFE* for 24 hours!"
Gone wrong
gone agi
Gone hyperbolic
gpt 5 was a flop release ngl
It's been like 1 hour
i love how all the gif i find when typing hyperbolic are all dragon ball z

👏 wow wow
no extra multimodal, no 1M token context
They talk about it like it's a deity 
turbo charge lets go
barely better benchmarks
true
than o3
swe weaker that 4.1 on non think
o4 is an abandoned project LULW

this isn't a livestream though... It was prerecorded
Well, not game changing. We will see if Gemini 3 is any better
API

why is he about to cry
He got GPT to write the script
he forgot his lines
allergic to bullshit
i for one am looking forward to the new gemini
same
Strawrbrerry
gemini really went from the worst AI. to one of the best ones 
fr gemini 1 was so bad
and palm 2
yeah, the Bard and early Gemini days were rough
the things you can do when you own the entire pipeline i guess
tho google made the most oss contributions
Can Gemini fix the pixel phone batteries though?
the things you can achieve with shit ton of tpu
via robotics maybe
They all sound like they're being held at gunpoint
investors
Well yeah, that's how it feels to be in the grasp of Roko's Basilisk
Corpo streams
they did the graph right this time
(other than forgetting the y-axis)
this is the version that they'll provide for free users 🔥
2k is NOT minimal
The accuracy is Minimal
That logo... Kinda sussy
inhouse
this woman is having some major internal panic 
everytim she stutters i can see her stress
I wonder how many of the testers said gpt 5 was better because it’s a bigger number.
i would hope they didnt show the numbers
i mean it was on openrouter for like 2 weeks (?) it was generally seen as pretty good
i assume they did blind test
bro keep it together 
honestly my brain always thinks that claude would be superior for programming
what editor is this?
i'm biased to gemini cause i worked for the company 😭
cursor
i have been brainwashed
aka Vedal IDE
cursor my beloved
ye the micro updates might get annoying tbh. i dont need to know every single thing its doing
not using vscode.... microsoft better drop them
need to maximize clout
on twitter

can i use cursor with fully local models?
cursor is vscode fork

this demo is so 
it is...
better than 4.1
4.1 no one rehearsed
or 4.5 or whatever
-# its been really great... akward pause
this is so boring

wow yet another next js spaghetti generator
I’m just watching the chat
much useful
500B comapny means you can get every human on earth 50 bucks
we, yes we hate doing frontend and especially LLM doing that
he said the meme
look at my username 
this shit is so boring
i'm gonna get hired to fix this garbage btw
tsx >>>>>>
at least bro is employed
wait frick i'm unemployed now
why are we editing a markup file?
i like the syntax not the file
i'm gonna get fired to code this garbage instead
-# (im not even employed)
me neither
anyone pls hire me
i don't even remember the last time i opened a jsx file
Good sell
no typescript = 
does GPT-5 mean that Neuro will be getting upgrades soon
how expensive is it
i have like 200 bucks, and you make me a state of the art npc ai with full custom behaviors, audio input and output, and its needs ai pathfinding 
oh y it needs to run in realtime too with 10+ of them at the same time running on a 3060 
man what is wrong with my ISP they keep disconnecting me
it's in 3d


seems very released
vedal already figured out memory too...
