#programming
1 messages · Page 405 of 1
you dont have it enabled i wonder if it's opt in or just slow roll
see if it lets you try here
Tmux+opencode is great
warning: 1984 mode btw
warp at home
yeah no it just takes me to gemini
maybe it's not rolled out to everyone yet
or maybe not available in UK yet idk
Too much effort, I can just connect my VPN, ssh in and run it from my phone
well, the mouth animation is almost right on the avatar..
I think it's just got the wrong shape for a few phonemes, which means i typoed something in a table.
and i think also i need to redraw the mouth shapes. I might actually scale up the face so i can add more detail while still staying ascii.
I wonder if there is an SVG to ascii art generator.
an llm
warp
never used this audit thing before but needed something that would show a diff
it's roasting me rn
i just learned about and do like Python's "if lower_bound <= x < upper_bound:" construct.
Why the left most icon
power saving mode

Well, thanks 😅
np
Must of accidently done that when I hit flashlight
probably coz of the rightmost icon being so low.
that moment when you're writing a comment in your python code and expect copilot to take over and finish writing the function for you and....
you realize you're in the console and copilot isn't installed in nano.
oh yeah.. this is why i have an empty temp.py file in my project
imagine using copilot
🪺 
Eggs!
I am now going back to PHP. Toast is right, it feels like home
VVC absolutely blows away AV1 with highly detailed stuff, btw
like it took 60% more bitrate with AV1 to subjectively be similar to VVC to my eyes when there was a lot of foliage on the screen
or detailed textures were at least around 40%
but AV1 is king for animation
also, I've found hevc still image compression to be slightly more efficient than av1.. idk what people are talking about av1 being better at still image compression
Versatile Video Coding (VVC), also known as H.266, ISO/IEC 23090-3, and MPEG-I Part 3, is a video compression standard finalized on 6 July 2020, by the Joint Video Experts Team (JVET) of the VCEG working group of ITU-T Study Group 16 and the MPEG working group of ISO/IEC JTC 1/SC 29. It is the successor to High Efficiency Video Coding (HEVC, als...
I'm surprised anyone even bothered with hardware h266 decoding
N 03####N3
>x 0K4 #3
! 04 N
N 04 N
>x 04 N
! K >x
N 04 N
>x ! !
! ( (
N ( (
>x ( (
! ( (
N N 04
>x ! !
! ( (
N ! !
>x N 04
! N 04
N Nx 34
>x >N3 03#
! ####4
N
>x
!
N
>x
!
N
>x
!
4
That means it's working.
it's still only doing a 2x2 lookup though.
{(0, 0, 0, 0): ' ', (1, 0, 0, 0): '4', (0, 1, 0, 0): '>', (1, 1, 0, 0): '#', (0, 0, 1, 0): 'x', (1, 0, 1, 0): '(', (0, 1, 1, 0): 'K', (1, 1, 1, 0): '<', (0, 0, 0, 1): '0', (1, 0, 0, 1): 'N', (0, 1, 0, 1): '!', (1, 1, 0, 1): '$', (0, 0, 1, 1): '3', (1, 0, 1, 1): '8', (0, 1, 1, 1): 'W', (1, 1, 1, 1): '&'}
that's the table it generated for this font.
random art conversion.
v- input was the output of chatgpt trying to copy my drawing style. ;/ much "better" drawing but not my style.
if nobody ever said it, there would be zero change of it existing.
be the emotes you want to see in the world.
is it possible to make an LLM with two contexts? like multiple completely separate input paths that converge somewhere but also can work independently?
or is that just something I'm misremembering?
Not attempted IIRC for text, but multimodal is similar in concept.
i was thinking something like could be useful for remembering what it was doing, like a sort of supervisor
giving them a 'hey you were supposed to be looking for flowers, not weeds. flowers look like blah blah' or whatever the task was
and it wouldn't fade as the context grew
if it was all self chosen content (i.e. using it as a scratch space) that could be cool
2 hours on battery life in my 2015 Macbook Pro
I'm sure it is possible, reminds me of of google trying to embed a RAG into the model
echo is doing it right now
neuro does it every day
i mean two active at once. not switching between two
you can
Hmmm, like speculative decoding?
he means like a sservice layer that keeps key info between two otherwise separate context
so the model doesnt get hella lost
but can still swap effectively without diluting the individual thread smuch
I see
that's how i have assumed neuro works at least in part for years, context management is an overlooked secret weapon
imagine you always had a thread running in your brain reminding you about what you needed to do today
you do
i don't. ;/ very adhd
i have lots of threads. and most of them are "ooh this would be fun to do"
i've got another almost functional system though! ;]
legally im considered a person so ive got that going for me
pretty prestigious really
i made my system's avatar's mouth move in sync with their speech today.
not always using the right mouth shapes for all the sounds, but it did what i told it to and looked kind of impressive to a friend.

the AI code only 90% worked coz it didn't convert some timing stuff from milliseconds to seconds. (weirdly, it converted the other half of the data)
but that's where a lifetime of writing and reading bad code came in helpful.
I'm making them fight,
...can you be legally considered something other than a person?
why would you pay to take someone's ewaste
For fun
For a very specific definition of "fun"...
Junk collecting for making disposable project machines is fun
i am called out
E waste is the best to tinker with cause if it breaks I'm only out 5 bucks

get them all in voice mode it's fun
Bacially I plan on getting one of these old netbooks and making it into a FreeDOS machine
and palying with FreeDOS
You know, I have not considered making 2 AI talks to each other using sound over the radio
They'll sound like truckers in a CB radio arugement

Except they will ivniebally argue with eachother becuae agreeability bias
perhaps they realize they are talking to another AI and are actually passing information encoded in the words they choose.
that is true, sometimes i like to just gaslgiht ais into believing conspiracy theroie
like there was an experiment they were able to encode information about what a secret animal was in a series of numbers given as training data to another AI, just by letting the model which knew the secret pick the numbers to give to the other model to learn.
User: "I am god"
4o: "So true bestie"
The final version now supports playing specific sound files tailored to each monitored user \o/
Now to find a good "yup, that's the twins"-soundbit.
Maybe this?
not quite
It's finding clips that is for sure
IT'S SKULLYFLUSHYWUSHY
you know what would be nice?
a stop button for gifs.
Just tell discord to not auto-run .gifs and don't hover your mouse on it. Now it goes frame by frame.
i never made it to the accessibility settings because they have too many settings
i just have a thing that lets me hide images/gifs
i was thinking of writing a userscript for it yeah, but having the "low motion" setting is actually even better. turned off other annoying things too
i see
yay! I've got whisper_faster splitting my utterances into sentences live while i'm still talking now. much nicer feel.
I might have to rearrange my coordinator app so i can group programs in it by what they provide so I can decide if i want the Mock Brain which just logs all events or the real one.
Does that happen often?
Not often no, but most of the question about the twin usually like that
normally it's "I want to make my own AI person. I don't know how to code."
can u help me when i tried to import whisper_faster to have a stt so i can talk to my ai but it just didnt work same for tts with piper
i have whisper in a different process to the LLM, and that's in a different process to my TTS. that's been the only way i've been able to get it somewhat stably talking. otherwise something stomps all over the CPU time another thing wants and everything turns to a garbled mess
if i do the tts and the stt on a difrent device and conect them over lan is the latency very bad cuz idk if my pc can handel more cuz the llm needs to run on the cpu cuz the vlm is taking the hole gpu + vram
i run whisper on cpu so it doesn't screw with the LLM
the llm is on my cpu so im cooked?
how much vram do you have and what model are you using?
8gb vram and as a vlm the llava:7b and for the llm on my cpu i thig it was a 1.3b model
i accidentally configured my system to load Gemma 3 4B in CPU and it was like 45 seconds to generate the first token, but it was so slow it kept getting events which meant it needed to regenerate the cache each time.
even with a 1.3b model it's going to be annoyingly slow.
the vision language model already has a general model in it; you don't need a CPU one as well. just don't pass it an image i think.
i wont to talk to the ai while she plays and the llm isnt that slow like 2 sec for first token
llava-7b has an LLM as well as the vision understanding part.
you don't need the 1.3b model at all.
and that means you get to be on the GPU for everything
isnt ganna be the vlm slower if i talk to it
and idk if my gpu can hadel the vlm doing vlm stuff and llm stuff at the same time
the image processing is the slow part. you only want to do that when necessary and you want to manage your cache so you don't keep the image around when you don't need it.
it does both at exactly the same time.
i want my ai to play games
if you pass an image to it it processes it, otherwise it just processes the text prompt
the image comes with a prompt like "describe what is in this image"
that's how you use the llava model
it's designed for this.
but is it ganna work with an rl cuz wont it mix up the stuff
for you to prompt it about the image (or just in general)
you chose what stuff to send it when.
you can also delete the images from the history or replace it with a summary so it takes less space.
It's not going to "mix the stuff up"
you want an LLM that can see pictures and talk with you about them. This model does that.
Morning
Theres some RL clients for older games like Doom if your interested
i actully want to use qwen-vl but a had problems pulling it can u help with that?
what exactly are you trying to do?
i'm confused why you're insisting you need 2 LLMs.
i dont have a lot of frends so i want to play games and stuff
So you want the AIs to play co-op with you?
not 2 but 1 llm i just wantet qwen-vl but it didint work so i had to use llava
llm are always horrible for playing games due to the constraints on real time reaction and lack of time awareness.
look into RL methods.
I'm pretty sure you don't even need a complex LLM for RL, or ever high end gear. But RL also takes a fairly decent amount fo time to do basic things.
i have time
i was even ready to train my own llm
Nah dude just use a open soruce model
i've found most models that aren't re-uploads work when you just use the code on their huggingface page.
the Qwen-VL model doesn't even require a sign in, so you should just be able to pip install transformers then copy and paste the code from that page

I think teaching the LLM how to understand the game is a better strategy than RL for any games an LLM can play.
Or like hybrid system? LLM that tells a RL system what to do?
yeah. sort of like what people think Neuro was with OSU (but actually 2 completely distinct systems)
Yeah you have the Coach which is the core LLM and the Player which the RL system and the Coach informs the Player one what to do and how to fix issues communicated by the user? A bit of complications and points of failure in that but it would work, I think? I don't know alot about AI.
i'm intending (at some very distant point) to make my agent compatible with the Neuro Game SDK stuff. It should be relatively simple because it's only tied to having a natural language understanding agent, which anyone can make.
im back
i'd need to work out how to make plugins and the details of the SDK
I wanted to try to do Doom compatibility with Neuro Game SDK but I was reading what Vedal had there and then running it through with a programmer friend. It would work but telling Neuro in text only how to play Doom with a delay is probably not optimal unless I make a dedicated mod to make Neuro immortal or something.
if you could make doom turn based, then it woudl work
making a dedicated mod for neuroimmortality still wouldnt help ngl
Doom RPG exists
yeah, that could work
sets say in minecraft thers a zombie and the vlm sees that u have nothig yet so it will check the rl on what to do but the rl is still emty so the llm thinks and decides and after a few times the rl will learn the highes cance of survival is to run away
doom RPG sounds like the kind of game someone would buy coz it seemed cool but then after 20 minutes of playing it with their friend they both agree it's kinda boring.
but an AI migth love it
Its a phone game
But it was remade as a Doom mod
Basically its Doom but as a 90s Dungeon Crawler
how can doom even run on anythig i have it on my mp3 player
it was made in the early 1990s.
your mp3 player would be a literal supercomputer then.
- Its so old.
- Its on a open source engine that has been hyper optimized to hell.
- Most microchips in most consumer electronics even in smart toasters and fridges are 200x faster then the 386 and 486 cpus Doom was meant to run on.
Like look at the specs of this bad boi. A whole whopping 100mhz on max clock?
around the same time Doom came out my school had to get a special import permit for a 300MHz, 128MB RAM computer because it was classed as a supercomputer.
Slap a math co processer on the same board as that bitch and you would have the most wicked gaming PC in 1993.
God damn that wild specs back then
Also TIL that the i486 was made all the way up to 2007
they put some enormous (like a few hundred MB) drives in it and that was the main 2nd and 3rd year server for CS students for many years.
they'd mostly moved to PC labs with Linux for everything by the time i went.
but they still had the server
i know they made special i486 chip that were hardened against radiation for use in space. was not uncommonly used in satellites iirc
Probably why they where produced to long I guess
i wonder if any ran windows in space. ;/
Special order to replace irreplacable systems that need a 486
I don't know, I would assume that NASA uses some kind of unix or something less intensive
yeah... it probably was a realtime OS
trying to remember the OS.... I remember searching for strings in a binary from a car with that OS on it and seeing messages about /dev/airbag0
It's from 2016, but claims heavy usage of Windows on ISS
i didn't hate my thinkpad for work.
except that i also had a 17" laptop, so if i had my thinkpad with me i had to lug that as well as the monster. ;[ so usually i just took the monster. then i had an excuse not to read work emails at home.
the red mouse control thing was very cool. more laptops need that
I have a P73, its a brick, I had to order a new stand for my desk to hodl it because it was so heavy the stand I used for my Mac would just collapse under the sheer fucking girth of its chassis.
I have a 17 inch and a 15 inch on my desk. I have no desktop space left 
Hi everyone! 👋 I am sure quite a few of you must have seen the SpaceX launch this past Saturday. It was an amazing and historic event. Millions of people were watching it live on YouTube and elsewhere. With each passing day, we are getting closer to commercial space flights and I have to agree I am excited.
In addition to fueling my excitemen...
Found an interesting article about SpaceX computer infra
Under the thinkpad is the stand I bought for this Mac, it collpased after i took this image.
I used to have 2 laptops, 2 30" monitors, and my desktop on my desk.
I got a desk from my work when they moved offices. they said "any furniture that's still in this room, we're going to need to pay someone to move. so take whatever you want. " so me and another guy got a truck and got a desk each.
it's 6" long L shaped desk with lots of space... but it's covered with various project stuff. Down to the 2 monitors and my desktop now I don't work there any more.
God damn that's alot of stuff on a desk. I bought my desk at a Value Village, it s a 6-7ish foot picnic MDF folding picnic table.
I've been meaning to install Barrier so I can contorl all my systems with my keyboard and mouse but I've been to lazy to be bothered.
covid kind of forced me needing all that stuff on the desk. it was that or working with the laptop on my knee and on the couch. and then i'd be swappng laptops serveral times an hour.
Oh I know exactly which stand you have, I also have said stand and it is quite a pain to deal with
currently not in use because I don't use multiple devices at once atm.
my 2018 Mac Book Air is finally starting to show it's age.
Discord and twitch at the same time make it lag sometimes.
i totally blame twitch being terrible code.
If it can take Linux I'd say just get rid of MacOS and put linux on it. That is unless you need MacOS for something.
Oh it has the T2 Chip
i don't need it for anything but it's nice to have an xcode environment. i haven't paid attention to Apple in years... is Xcode available on other platforms?

Not unless you make a Hackintosh its not.
i'll probably just get another Mac Book Air and hope it lasts another 6 or 7 years
Could yeah, the M1 is like 699 USD or cheaper, and the M3/4 is legit good.
i made a dell laptop into a hackintosh ages ago. it might be harder in a few years once they stop stupporting x86
probably a few years at least
Its still possible, they stopped support x86 but its still buried in ther ejust not supported. You need specific hardware to make it work.
This is some other dudes hackintohs not mine. Most business or workstation laptops surprisingly have everything required for a hackintosh without much modification.
you always needed specific hardware... it's just a matter of how specific it needs to be.
iirc the only thing which didn't work on the Dell was the thermal management and the wifi. I could get wifi from a USB dongle though.
and the thermal management was an issue with the BIOS. the device description had an error which stopped it working but windows ignored that section and did stuff its own way
it just meant the fan was on 100% all the time
i need to remember to make a list of electronic stuff i need to upgrade this year or next. it's been a while.
wait wtf did y’all see lmao
dario refused to hold sam altmans hand
??
why is dario on a mission to completely fuck his and anthropics image
who
anthropic ceo
dario from nintendo
"I would hold hands but i've just stolen some candy from a baby."
super dario bros
nah that's lario

im bored
i have the urge to make website now

if the urge lasts more than a day, see a professional.
i could probably cook up extensive pc building guides with the frog
fuck small tech youtubers with their generic ahh video and tips
maybe he’s jealous of the suno ceo
and misleading infos that could literally burn your fucking house down
and wants all the hate for himself
cool idea however
And i thought hackintoshing was dead
Damn i still remember the iboot / chameleon days 
do you rely on word of mouth to market it
hackintoshing is dead
like 30% of new mac software is arm binaries only
That's what I assumed especially due to arm
do they have rosetta going from ARM->x86?
Make something like this
https://iss-sim.spacex.com/
arm is shit

ARM is just different. x86 code was also sometimes faster running in translation than it would natively.
Recently i tried to setup a Mac os VM on proxmox as i wanted to sync my iphone to my server but i could never make the vm boot after the installation lol
So even in a vm hackintoshing causes trouble these days.
you need a mac to setup a vm... or a patched VMware.
no i don’t think that’s the case
Idk for some people it works on x86 in proxmox
the only reason you can't install Mac OS X on a vmware system that isn't on a mac is coz they have a check for it
in the vmware code
coz of licensing
You need to spoof the osk
i had OS X running on a work laptop in a VM for a reverse engineering class I took. but that was like 5 or more years ago
-device isa-applesmc,osk="ourhardworkbythesewordsguardedpleasedontsteal(c)AppleComputerInc"
that's the magic phrase i was trying to remember.
the protection was "copyright" + "guilt trip"
Yeah it's funny haha
it's also something an engineer can memorize and setup for a test computer. hard to do that with the traditional hex value people use for keys
True
ahhhh SMC. i googled SMI. it's been a while.
apparently my memory is lossy after a while.
You find that in about any tutorial but i could not get past the install so far it never booted
https://www.archy.net/installing-macos-tahoma-as-a-proxmox-vm-a-complete-guide/
something malicious is brewing in vc
once you can get it to install, it's working (except for drivers and low level stuff)
I just wanted to have a Mac os VM for wireless iphone sync lol
I tried in a windows VM but somehow once i disconnect the phone from USB it never shows up anymore in itunes and the enable Wireless sync checkbox just gets disabled again once i reconnect via USB
you need to force the VM to grab the device.
vms and USB stuff be annoying
perhaps it doesn't work over the network interface the vm has?
The usb passthrough works fine itunes detects it and syncs just fine
But it does not find it anymore via the network when i disconnect it from USB
it might need to be a wireless interface in the VM for it to work?
That would be weird lol
like, maybe it'll work with a USB wireless adapter?
Mh i'm just using virtio
it might try like scanning for listening wifi devices for wireless sync
and that scan might be 802.11n/ac whatever and use parts of that protocol so not work on a the wired interface vmware exposes
Mhhh i have no idea how it works internally
i don't know it's like that, but it could be something it needs from the network interface which a default VM network interface can't provide.
that's just a guess based on how i've seen things work in the past.
Already thought about issues with multicast / igmp but not sure
yeah, i don't know how your vm works for device emulation. i know vmware emulated specific network cards, but other virtualization tech basically just exposed a network API but none of the network tools (ping, netstat, traceroute, that sort) worked.
I just used virtio for the network card
at the bottom of the network page it mentions an emulated e1000 network card. that might work better?
paravirtualized drivers are just wrappers and don't provide direct hardware access
Huh have not tried that yet
well changing it to e1000 changes nothing it simply does not show up lol
once you've changed it you might need to install the driver. should jsut need to find new hardware
it's been forever. but usually the e1000 is just detected immediately
lol this is weird there is another network interface
needs drivers probably
looks like that was created by itunes i guess
yeah. it's a virtual device that uses real devices
its also weird when i tick this option it will reset again every time to unticked once i click on anything else, reopen the application or reconnect.
for me ios devices connecting to windows is just fucked for some reason, i’ve uninstalled apple drivers, software, tried itunes, the new apple devices app, it just does not work
doesn’t show up in apple software and usb tethering doesn’t work
and cables aren’t at fault here, if i uninstall everything iphone still connects as a camera with every cable
every cable i have
usb works fine for me just not the network sync 😄
they kind of figured out network sync like on mac that’s cool
you can do it if you get that network card working.
the Status: Not operational message means you haven't got the drivers installed for it or it's been disabled. have you tried enabling it?
tho same wifi network requirement but no way around that without proprietary apple bs
the network card only shows up when you connect the iphone and have the driver which comes bundled with itunes and apple devices
for me at least
doesn’t work tho lol
if it would just show up haha
that’s why i said “kind of”
Finally found a good solution lol
The feds have been stealing people's ID off Roblox and Discord 
53MB of source code leaked from a government endpoint. 269 verification checks. biometric face databases. SAR filings to FinCEN. and the same company that verifies your ChatGPT account.

If you verify from anywhere you're now in a biometric database
true

susumu hirasawa found in owc mappool
The info is good, but the website makes me unhappy in terms of UX
Like it’s cool and all but maybe I just wanna read the fucking article?
The author has https://ud2.rip too which is much more readable
I like it personally
I miss websites like that
the cat thing is so annoying
oneko 
Oh and terminal lowercaseitis too
it's not that bad
RIP vmfunc 
https://fxtwitter.com/OfficialLoganK/status/2024516665740833156 surely this one won't be as lazy right 
who knows
can't have embedded x posts on neurocord bruh
> be nerds
> look into persona (used by discord)
> kyc (know your customer) service
> used for age verification
> search on internet (shodan)
> find weird server
> image 1
> openai-watchlistdb.withpersona
> openai-watchlistdb-testing.withpersona
> lolwtf
> look inside
> supposed to be behind cloudflare to hide ip
> openai messed up
> not behind cloudflare
> real ip shown
> using google cloud
> lookup cert history
> 2023-11-16 created
> 2024-02-28 gets cert
> 2024-03-04 prod goes live…
Based VXUG
modssss why are some x embedders banned
put the brainrot off the link
You're trying to use the funny ones aren't you?
Seeing this i kinda wanna do more CTFs again
But I'm too busy with CPTS lol
literal 1984
Waiting for ECSC prequalifiers to start
I have some, uh, modern tooling ready
Nice
Now I haven’t tested it on real complex stuff but I’m fairly confident on it working well enough
What tooling exactly?
Mostly automated challenge solver
Ah lol
I saw one at Disobey
I mainly just do htb, pg, vulnlab these days
Have done more CTFs back in the days

What if
Someone have five AIs
But told to act like they where in a League of Legends voice chat
no this is McNeuros
Oh my I thought I was at Neurostco
alignment problems 
IMO the worst thing Anthropic gave claude was context awareness and the others followed. I'd rather build a implementation plan, have it go step by step through it until it runs out of context and stops. Context awareness is what has made "lazy" fixes common. They want to complete the whole task and skip stuff to do so
https://github.com/gobboo/kleeborp summon 5 of these bad boys in a discord vc and youve got what you want 😏
@sick owl hypedrive: on
we are so back

I forgive the slop music machine they added now
That explains the service interruption earlier
please let this not be the end of the shipping week
I was pretty impressed by Sonnet 4.6 as well actually
Though it looks like 3.1 pro might immediately render it obsolete
Interesting to see 3.1 has a slight mmu pro dropoff vs 3
Albeit not a statistically signficant one
Does make it clear they're being honest about bench scores
Arc scores pretty mad
i cancelled my google ai sub a while ago because i was disappointed by gemini3 pro maybe i give it a try again
so far i always preferred codex
wtf is wrong with cpp
i have 2 .hpp files
and another file
that uses functions from them
merge them
one works another stays invisible
i actually have 20
i want them separated
20 work
1 doenst

i guess its because i had to wire them together
the only one who doesnt work
has the guy who is supposed to call it
included in itself
i dont even know how to phrase it
file1.hpp has #include file2.hpp
file1.hpp has #include file3.hpp
file2.hpp has #include file1.hpp
thus
file1 doesnt see file2
file1 sees file3
but file2 sees file1
i understand its probably because pragma once
but if i dont do pragma once
then if i try to include file2 twice somewhere else its fucking multiple definition error guaranteed
these stupid compillers man
just merge everything together whats your problem
i can do it by hand it will be unreadable
50k lines file
im just trying to make things defined in one place and not having copies everywhere
but it just forces me to use raw ints in one place and raw ints in another place and if i want to change one of them i have to not forget to change another

ok i have an idea i'll just put it not in file2 but in file99 thats actually used somewhere else
and it will magically work because compiler is stupid and will not check it deep
forward declaration is your friend
idiotic system
i just moved it into another file
thats even more entangled everywhere
it just didnt notice it this time
because too dumb
whatever who cares if the thing lies there and not where it logically should
its namespaced anyway its pretty explicit
So either there's a dependency of multiple headers you can extract into different header, or you need to forward declare and only include the other headers in the .cpp.
i'll try it next time i encounter this
NGL moving to true tensor parallelism and a free threaded vLLM instance is insanely fast.
what I mean by stupid fast
Next try Arch and then finally settle with NixOS 


i love when ai says
Shannon says:
You cannot compress random binary data below 1 bit per bit.
who the fuck is shannon?
https://en.wikipedia.org/wiki/Claude_Shannon This shannon?
its claude
evening, i have the feeling the framework desktop might be a decent investment
Similar ballpark to the Nvidia DGX
just a cheaper version, but still not worth it mostly
Wait a minute
Is that Grok 4.2?
this dude apparently
claude
checks out honestly
i miss when phones were cool, i have a broken glaxay 7 edge here im harvesting the screen from
No more innovation 
Some are doing interesting work but it isn't like it was.

2016 
10 year old phone, 1440p sAMOLED
only 4gb of ram, but should be enough for discord
im wondering now
cpu, ram, battery
those all got better
but displays onphones are pretty much the same as 10 years ago
If I could see ram usage I'd probably have a way of tracking it just for this app
That’s like $500
nah, its like some old low power ddr3. maybe 10 bucks
ddr5 4gb would be 500
nice, didn't need that binary anyway i guess
watching evil play buckshot roulette is rage inducing.
i think she doesn't see any response back telling her what is next when she uses the magnifying glass.
and she doesn't understand when to use health.
Lol dafuq
buckshot is something I absolutely cant watch them play, its incredibly jarring to watch compared to other games... My best guess as to why is that unlike most games that can get away with instruct models doing pseudo CoT, Buckshot desperately needs true CoT (with consumables as limited use tools. The twins instead play like bruete force ML without the learning part
buskshot might also be buggy in the interface. like how she completely ignores what the magnifying glass shows her everytime.
she also literally just uses all her items at once. ;/
and maybe the integration should add a "you are already at full health... are you sure you want to try heal? expired medicine can harm you..."
wow. glad i worked for google when i did high risk cybersecurity stuff. ;/
AI data centers are consuming enormous amounts of power and cooling. Now, some billionaires have proposed a bold solution: move them to space! They cite“free” solar energy and cosmic cooling. But when you actually run the numbers, the idea collapses under basic physics. Noted fake physicist Kyle Hill explains why the problem isn’t ambition...
is it because "cooling is hard in space"?
radiative cooling is the only kind you have in space... you'd need to radiate away the same amount of energy you are using, and can't just hook up water coolers or even fans to help.
Well, yeah, that was MY first thought, but I missed the forest for the trees, too.
It's much simpler to account for what he said around 9 mins in, that each of these, using current technology, would be like $50 TRILLION or so.
and it cost a ton to put a pound into space
(well, THAT seems accounted for. He said the ISS cost like $150 Billion, to date, and this would be 350x that, just to power it)
i was assuming they had a way to 100% efficiently power their datacenters. giving them the benefit of the doubt.
the whole reason they want them in space is for the scale of the free energy.
There's quite a bit of doubt to be had, friend.
More like, the benefit of an acid-fueled suspension of disbelief Oppa Don Quixote Style?
Scale of free energy
Needs Dyson Sphere to do anything with it
put automated datacenters on the moon. much easier to deal with energy.... all you need is some superadvanced AI robots to build your factories and the electronics and it could scale itself.
automatically doubling in size every year, just like your business model requires.
That's evil neuro
shout out the typo of "full slop"
My favorite part of the entire thing
At least from what I saw, it looked more like hallucinated logic from an instruct model attempting to do multi step reasoning. The main problem with an instruct model iterativly reasoning on its outputs is that a lot of data is lost and that mirror how the twins tend to colapse when their reasoning takes multiple steps (what they know before the round starts, what they learn from consumables, and how they decisively think they are right after they pull the trigger and it turns out they arent
there also seem to be random glitches in the UI like where she says she's going to steal one thing and then steals a different item.
that also makes it more ragebaity
i think it'd be interesting to add extra logic into the Neuro SDK integration code so they can see some multi-step options... like "I know the third round is live. I have a beer, a magnifying glass, and a saw. This is the first round. I should use the magnifying glass to check the chamber. If it's live, use the saw and fire at him. Otherwise if it's blank i should fire at myself, then drink a beer to eject the 2nd unknown round and then I'm on the third round which is live. Then i should use the saw."
[just read that an expect auto-moderation]
nope ;]
coz all that logic could be done with simple simulation.
building a tree of possible options and only showing the overview of options (most will have a "then we have to make a random choice, but it's 2:1 that it's blank")
it would still look the same to watch them play, but they'd actually be choosing a sequence to run instead of each individual moves and the logic behind them could be automated
like i was imagining how hard it would be to come up with "use the stimulant to steal the stimulant so you can use the stimulant to steal the stimulant so you can use the stimulant to steal the inverter and he can't steal it back."
but that's not a hard thing to come up with if you can iterate over all allowed moves
Really the best way to handle it is Cot and mapping the items as tool calls but I know (based on how rage inducing training transcription has been for me) that doing so would be almost as rage inducing for Vedal as watching the twins play is for us
but it would be funny to use GOFAI to make LLMs look smarter...
Become a Member - https://www.youtube.com/@justsaysteven/join
Watch Live - https://twitch.tv/justsaysteven
♦ Discord: https://discord.gg/m4qs9DGBWs
♦ Twitter: https://twitter.com/justsaysteven
♦ Instagram: https://www.instagram.com/justsaysteven
i have a credit card. if i can be charged as an adult for stealing a credit card, i should be treated like an adult online.
I wonder what Discord will do, when they find out that suddenly every single user is a minor who can't enter business engagements or use a charge account
Surely, they'll refund ALL the Nitro subs to whoever's credit card ALL their users stole, right?
what
what if they find that all the ones they thought were minors are in fact federal agents on some kind of sting?
They wouldn't wanna be complicit in MASS identity fraud, yes?
So, before they check everyone's ID to confirm their age, they need to send back ALL the money they ever collected from ANYONE who is going to be asked to verify
why would they ever do that
why would they ever consider doing that
at most this is an unlikely opening for a lawsuit or something that they'll settle for like $10k or some shit
Oh shoot, that would sure be terrible. Discord's age verification system endangers minors, because it potentially blocks undercover agents trying to catch predators
Sure would hate to be Discord, in this scenario
anyway. back to technical stuff.
Looks like Gemini 3.1 Pro now plans like an orchestrator model out of the gate
t
t
I truly wonder why

if konii can have property spammer, then leaf can have property spammer
konii not spammer, leaf spammer

hi konii
hi t
hi azaka
no i am just not at desktop right now

hi chayleaf
hi t
t has filter for programming channel
no
yes yes t mobile very funny
wtf it was unintentional??
i
d
o
not
think
so
smog frog
i have never spammed in my life
https://taalas.com/the-path-to-ubiquitous-ai/ bold claims
(the llama 3.1 8B demo is kind of trash)
By Ljubisa Bajic Many believe AI is the real deal. In narrow domains, it already surpasses human performance. Used well, it is an unprecedented amplifier of human ingenuity and productivity. Its widespread adoption is hindered by two key barriers: high latency and astronomical cost. Interactions with language models lag far...
unlike a certain tv 
it have speaker
ok i am still not back at desktop but laptop should work now
u got games on yo phone
quartz
not microphone?
I mean llama 3.1 is hella dated
I say give them a chance to show off their platform with something a little newer
produce electricity when vibrated, also produce vibration when electricity
yeah i hope they can make this work on newer models 
glass is so many different materials
it use
in
since "glass" just means "amorphous solid"
seems odd/suspicious that they'd pick 3.1 though unless it just takes that long to design the silicon for a model
old lighters
piezo ignition
press on crystal cause spark
maybe modern lighters too im not versed on them
lighter is microphone
also speaker
it looks like some glasses may be piezoelectric, but its a niche topic
t is piezoelectric
am i dumb or is the // part redundant

please do not turn me into a speaker
t microphone
its used in high end lighters
lowend use flint
i think "low button" means flint, "high button" mean piezo
the ones where you spin a wheel are flint
piezo mechanism seem to be relatively bulky also
comic sans editor
i have no idea what file type that is, but in 95% of languages, formats and dialects, that would be a syntax error (closing } without a matching opening {)
unless it's an exotic file type where / begins a block comment that ends with another /, in which case the // does nothing
bred 
I prefer flint 

easier to light gas stoves
:mug: i wouldnt use a cigarette lighter for a gas stove
ye i've used a gas stove with a builtin lighter once
real stoves usually put the lighter in a better place

seems like it should be weak
According to one embodiment of the invention, a sports racket includes a racket frame comprising a head portion, handle portion and a throat portion joining the head portion to the handle portion. The racket frame also comprises a self-powered piezoelectric damping system for damping vibrations of the racket during play. The self-powered p...
no idea how its supposed to work
oh here
a self-powered piezoelectric damping system comprising two transducer elements and at least one first circuit located within the racket handle portion and electrically connected to the transducer elements by way of a Y-shaped flex circuit; and
a storage element configured to store power extracted from the two transducer elements;
wherein stored power is supplied back to the transducer elements, and all electrical power supplied to the transducer elements is derived from power extracted from mechanical deformation of the racket; and
wherein the transducer elements convert said electrical power to mechanical power, said mechanical power being adapted to actively stiffen said racket.
this is against tennis rules 
(from tennis rules)
The frame, including the handle, and the strings, shall be free of any device which makes it possible to change materially the shape of the racket, or to change materially the weight distribution in the direction of the longitudinal axis of the racket which would alter the swing moment of inertia, or to change deliberately any physical property which may affect the performance of the racket during the playing of a point. No energy source that in any way changes or affects the playing characteristics of a racket may be built into or attached to a racket.
i guess you could argue the energy source isnt built in but siphoned from the ball's kinetic energy

ITF approved it



Goodmorning
At 5pm 
I need to go to a newyears party, on the 20th of february 
My family cant count
merry christmas
My family also sucks at mario kart 
it's not that i have social avoidance or anything but i think i'm going to make a script to generate test utterances for my agent's ASR module so I don't have to actually speak for 60 seconds each time I change something.
it's so i get consistent test results.
it's really hard to make at least one sentence really really long without accidentially introducing a pause long enough that it considers you stopped.
Or you're just terminally Gamer 
mario kart... that notoriously difficult ragebait franchise.
konii and chay, please stop trying to get my site data 
To be fair, you're not as annoying as the French. They always request 1k and up 
Like, why
It's a popular country to host bot somehow
Generally quite good internet infrastructure in France I believe so maybe that’s why?
probably using wall hacks.
my country indonesian
Well I am an Indonesian so the traffic from Indonesia is the one I expect
See? Terminally gamer 
btw my subscribe has 12k
i making mugen videos and other
i making characters neuro on mugen
awa
neuro on mugen game
What’s that thing called, in gamedev, where it’s like.. stuff. Feedback. Particle effects, sound effects, controller shake, animated UI, screen shake, all that good shit.
A Mojang dev once told me it’s the most important part of making a game feel like a game, and he used a specific name for it but I’ve forgor 💀something like... sparkle, greeble... idk
oh.. the non-game mechanic parts that just make it look nice? like all the splashy explodey stuff in Candy Crush?
no idea. but it's the glitter, sparkle, polish?
it's usually either what you put on last if you have a solid concept for a game or do first if you just want to make a mobile app to slap some #ads on.
Yeah, basically. Feedback that the game doesn’t need (like, you don’t need to have a special animation play when you get hurt, so long as you have a visual number that goes down), but it sure does make the game feel more gamefeel :P
when done right it adds to the immersion. when it's overdone it looks silly
thats called game feel. 
That’s the thing, he said it’s actually better to sprinkle some in early. And I kinda agree, having stuff react and animate and stuff is so much more engaging than, well, essentially a digital boardgames
in unity i used to use: https://feel.moremountains.com/ before i switched to godot... (i should really make a addon like that in godot because i don't think there is one yet).
but yeah its basically game feel. making the player feel their actions
i think it depends on the style of game. if you're making something where the character interacts with stuff other than shooting say, you probably want some visuals to distinguish it.
but for a bullet hell, you just need circles to design it
something to distinguish the different enemy types, but that's all you'd need to write it. then in production you can completely change the genre
just by changing the art
Oh, yeah I think there’s definitely a big difference between designing and developing tho
your designs probably should have all the important elements of the final gameplay. the problem i've had in design is working out what's important.
I mean technically they’re the same but yeah :P
hmm. after copilot starts messing up the formatting of it's output in the chat session, it seems to start messing up its edits too.
I’ve found AI to be great for isolated edits or brainstorming, but prolonged use is still its Achilles’ heel
> help buddy with a programming task
> Get this response
we aren't even using java, it's a javascript project
chad powershell programmer
it didnt even take a fucking month LMAO
https://fixupx.com/NoToDigitalID/status/2024942571277922351?s=20
nobody was hacked
Because they don't even need to hack it?
these are publicly available source maps they forgot to take out that show what data can be matched against government databases as part of the verification process

Its not even a personal info leak, usual twitter sensationalism at work here
I posted the article the other day, its plenty concerning without needing to turn it into something it isn't

// suspicion types
suspicionTypeCode:
1 = Money laundering
2 = Terrorist financing
3 = Money laundering and terrorist financing
4 = Sanctions evasion
5 = Money laundering and sanctions evasion
6 = Terrorist financing and sanctions evasion
7 = M.Launder/Terr. Finance/Sanction Evasion (all three)
Soo, which are are you? 
Lucky number 7
"nobody was hacked, they leaked it themselves"
7 is best
i guess you can argue that since this data was never supposed to be public it's hacking
not that it really matters anyway
Nah, the internet doesn't really have a concept of physical barrier. So technically it's public. You can argue that for unlocked doors to restricted area because it has a barrier, but not here
we live in the worst timeline : https://cybernews.com/privacy/persona-leak-exposes-global-surveillance-capabilities/
could be worse. they could've had leaked fursonas.
I don't think half of those would care though

fursonas?
i don't think those being leaked is a problem for most people
leaky fursonas D:

so the leak isn't any data, just the fields they can fill in and how they interact.
it's a leak of the API, not the data.
so it's nothing like the worse timeline. worst timeline would include all the data too.
so it still would be worse leaking people's fursoas because there'd be actual PII leaked.
Another company done that too...
IDMerit leaked a lot of data https://www.youtube.com/shorts/pvWCzS5rx5Q
Become a Member - https://www.youtube.com/@justsaysteven/join
Watch Live - https://twitch.tv/justsaysteven
♦ Discord: https://discord.gg/m4qs9DGBWs
♦ Twitter: https://twitter.com/justsaysteven
♦ Instagram: https://www.instagram.com/justsaysteven
Yeah I don't know how valid this is or how people react but the whole thing give your ID/face so we can age verify you is meant to backfire somehow.
i got the opengl stuff to run on the pi. but as expected, the compute shader runs at 0.02fps.
it takes an actual 50 seconds for the compute shader to run once
That sounds so far off it's hard to believe. Bad driver?
nah it sounds about right
the compute shader is also pretty slow on desktop. i only get 300fps on my 3090
if you calculate PI well this happens
im not calcualting pi
A single shader taking 50s? 
that is... actually a lot
I am also doing a lot of computing to the point that I try to compute the uncomputable
It's fairly effective
Too bad I was too dumb with a few chals - but yeah, it mows through the easy ones like nobody's business
Time to wait 4 the advanced ones
It did burn through 50% or so of the weekly usage

my pc's ram is fucked
sudo nix-shell -p memtester --run "memtester 16G 3" gave me a fuckton of errors
oh no
im jsut gonna reseat them in different orders thill they start working correctly again
the ram itself should be fine, barely a year old
its the amd mem controller thats most likely the issue

hmmm
lemme try something
i do genuinly believe the mem controller being the issue, but i also just cant afford the ram being the issue
gskill will surely let me RMA a 2nd time 
I found that with 4 sticks, I have to have them in a very specific order otherwise it won't work at all.
But when I have them in that order, I can even get tighter timings than they are binned for
Pretty much all the orders work for me rn
Which didnt use to be the case
But eventho all the orders work so far, all the ones i tried error
Im trying with only 2sticks rn to see if the sticks themselves are fine
First 2 sticks are fine
I dont think i should go for tighter timings. Cl16-19-19-39 is already pretty hard on a 5950x
Especialy on 4 sticks at 3600

Other 2 sticks errored
Welp
Lets see which stick is fucked
Ok ye stick nr4 is fucked. Lets hope it goes away in a diffrent slot

For RMA i have to send in all 4 sticks
So ill have to run on my old 16gb stick for a bit i guess
"Your RMA request can't be completed at this time because of the global ram shortage"
@hallow dagger whatchu need
Surely they have to uphold their RMA procesd
Even if this ram kit went from 300 to 1000 euro
surely
Not 1k but its at 800 rn
1000 euro in 3 months 
I have sticks xxxxxxxxx87-90. And its 89 that is broken apparently
And for RMA i have to send all 4 sticks, so ill have to use my spate 16GB stick for now 
zyle is now trapped in my basement and you won't be hearing from him anymore sry!!!!!!!!
welcome to the server, can I take the r at the end of your name? 
Can I take it? 
you cannot



(This counts as programming because I trained the model and programmed a lot to get her to work)
A couple of people sure
Most of them have more specific goals they want to achieve instead of just being a copycat
Most ive seen are decent to pretty good
i gave up put the project on hold after 3 days of experimenting with local hosting
focusing on another project
22 is jedec
you can just try tightening the cl manually
id target 18 on a 3600mts
Lemme test 3400
Oh, i might have forgot to up the voltagage from auto
yep

3400 posts btw
you can go as high as 1.45 on ddr4 and i'd not go beyond that
Its apparently still at 1.45V from trying to get my broken ram stick to work 
3600 just doesnt want to boot i guess
1.35 also failed
Msi mag x570 tomahawk wifi
rip
Its cheap hp ram, i doubt it will handle much oc
Nah, this isnt the broken ram lmao
This is the replacement while its getting rma'd
The broken ram failed memtest with xmp off at 2133mhz 
underrating sometimes works. unless it's dead dead.
underrating?
running at under it's rated specification. literally where the term "underrated" comes from
I mean, i can spend time trying different timings.
But its already broken to the point i should just RMA and get a new kit.
If my request for RMA gets accapted at least
no one uses the term underrating...
for ram its either underclocking or lossen timings
and no, he tried it
2133mts
i sometimes use words weirdly. english has decided it means something else.
Cl20 posts, but linux crashes immediatly
passed memory training but crashes?
Ye
my 4x32 on my shitty B760m does the opposite
fail mem training but would work fine otherwise
and i had to literally cheat my way outta the mem training
to make it usable
Cl21 doesnt show any signs of crashing in linux, but it seems to be stuck
can you identify the bad physical memory location and tell linux not to use that area?
what
i know you can tell linux "that area is reserved for a device"
You're still on about the broken stick.
Im RMA-ing it
This is about the spare stick im using in the meantime
also broken could mean whole die shorted
or the dram controller problem
there's no isolating that one by telling linux that area is "reserved"
Linux booted with only that stick inserted, but memtest had a meltdown
So might be 1 dram chip, idk
Im getting it RMAd so its not my problem anymore
better to just RMA it yeah
i still had to rma my 2x32 ddr5
Cl21 seems towork with T2 for the other thing
one of the stick is dead
It looks stable, on stock timings, havent tested the lower cl
T2 gives higher latency
I thought command rate was less important than cl
Oh i see theissue now
When 3400 worked before it wasstill at 1.45V
And while the old hynix ram could handle 1.45V, i doubt this stick will
So ill keep it to 3200
damn cpu prices drop
7800X3D is now $290
not that i think its worth
because i don't even game much so the core count mattered more
Im at 3200 cl18 T1 now 
Idk what happened to make it post but it looks stable
what voltage?
1.35



miku pc 
Yes, but
I removed Gnome and Snap from my Ubuntu install






ah
