#programming
1 messages · Page 411 of 1
2 hard
im not saying it cant be philosophical lol, just that the question being asked didnt require a philosophy side tangent, the answer is it's bannable, that's not debatable, whether or not that's a good thing is up to you but changes literally nothing about the answer
The asnwer that you are giving has already been given and the topic moved on
?????????????
bro is dodging accountability like crazy
it does, just debugging some extension that doesn't load properly
Why did you cling to ITS BANNABLE when my point is. "Automated accounts didn't get banned and manual accounts did get banned"
Because it means that discord doesn't ban automated accounts
it does not at all mean that
obviously?
But then why ban people talking about such bots and developing such bots if you could use such people as research for preventing such bots?
they dont need to do that
so that'd be why
it's not that serious
they barely care as it is
what we were trying to say is
that it is against the rules in this server
to discus
trying to save you from getting zepplined
but you really do not want that apparently
Do you not care about this server being botted by fake real users?
it's not something that i spend any time worrying about and neither should you
I am not worried just curious
The mods have thteir tools
its their job and shiro pppoppps in occasioanlly to talk about stuff like that
Someone using a scrippt to send messages really isn'y a problem unless their scamming or sppamming
Wait now I am curious what tools do they make
Exactly. So where does TOS land in scripting?
I see
i thought we'd moved on with that answered
or have we not
answer is still the same
This is new insight
no, it isn't, it's stating the obvious
Thank you captain obvious af
i mean deflect what you want but you're still fishing for ??? reasons for someone to validate some random anecdote and it's very confusing
we already have people trained to spot that, it's not anyone's job here other than those people
https://discord.com/safety/platform-manipulation-policy-explainer But yeah they explicitly forbid it
oh hey look it's your intended usecase
So If I manually press copy paste the text from my bot it would not be against TOS?
no
not THAT tos, but other TOS
Nope thats the first one
Yeah I mean Bot runs locally and then just copy paste YK
at discord's scale, yeah - that's why they (discord staff (on twitter though so take that with a grain of salt)) also said that whilst cli*nts are not allowed according to ToS, they don't really care unless it makes the user send unusual requests or harm their infra
weird that they disallow accessibility mods that change the UI.
Its liability reasons usually
efnrocement is another matter
As long as you don't draw too much attention they usually don'tt give a shit
I'm starting to feel like this person is a self-bot as well atp
sounds like a waste of effort for your apparent research purpose at least for getting meaningful amounts of data
It's not just data it's recording what causes bans and what doesn't cause bans
you cannot get meaningful information from yourself doing that
I mean what's the worst that happens if we're right and you're wrong?
Automating anything ni a user account is against the rules
the enforcement of tos is so spotty and random you'll likely not even know the real reason behind the bans and even if you do it indicates nothing about the overall picture
the limit is 0
1 instance = you are risking ban
?
if you automate any actions at any time on a regular user account, it can be banned
it says "don't or there will be consequences" and if you do, there may be consequences.
Nice xd
you can make a bot account and use the api...
https://discordpy.readthedocs.io/en/stable/discord.html
assuming it's an allowed use.
sounds disingenuous and like a waste of time
it's the same
not automated cause I press the button to send message
evilDentge
Yes hobbies are a waste of time and evil wants to burn the world captain obvious af
hobbies arent a waste of time in general but for you specifically your hobby is either wanting to get banned or ragebaiting people
sounds exactly like a sort of discord bot.
which i just linked to the official discord docs for.
Basically yes, because I am more curious about the limit of how much can you push the bot to being real human that looks and talks like normal discord user, without automation catching it
if you do it and I do 10x pretending to be other people's bots....
unknowable without massive sample size and even then
discord becomes shittier
That's what I am trying to prevent for my discord servers and my friend lists
why pray tell might they need to avoid detection by automation filters?
by making a "pretend user"?
I dont see the link
i'm saying your pretend user is the problem
why might you want that information?
the link you sent wasnt official, no?
i think if you wanna get yourself banned, don't do it here.
extremely obvious
that is true. example: I dont know who the fuck you are
i think if you wanna get yourself banned, don't do it here.
ooooo he's tryin
Wdym linked to doc?
i think "i should do it because bad people might be" isn't a good argument.
did that fall out of your message context?
how to make a bot account
I am doing it because I am not the only one that suffered from it
not a valid reason
are you saying you suffered because you made a bunch of self bots and then only your own account got banned?
I just gave you argument to why I am doing it
i couldn't understand any.
i think also if you were smart you'd shut up about it
Okay so imagine AF is a bot and is just a secondary in this conversation like "subtitles"
they're a bot who is contributing to a discussion helpfully. I don't care if they're a bot.
given that they cannot comprehend this subtext it's clear for once that they aren't botting right now since an llm would understand the hint
s.replace("AF", "CrazyBodhi")
Is it contributing helpfully or supporting you blindly to make you feel better is the question I always get
Not botting right now on this account
Seems like someone being helpful to the server.
See. You will never know if it's a person or a bot. Maybe rare Possum is just a moderator trying to change the dirrection of the chat, or a bot. As you said you have "trained" people for moderation
honestly? even if they are only supporting blindly, they're doing a much better job of contributing to the convo than you
i wish i were a bot so i didnt have to think about things such as that this person might drive a vehicle
if it's impossible to tell, it doesn't matter.
or have kids
it's the bad bots. less well (or less well intentionally) written ones.
But it is possible to do evil bots to make more drama in some community because it would give you internet fame
It's like raiding other discord servers but slowly with bots
sami already did it.
aka not supporting you blindly
I am not saying I do it or have done it. I am saying it is possible
and once again not only is drama banned in this community I feel like it would just get shut off the instant it comes here
everyone knows it's possible
that has never been contested
literally not even once
i will remind you again that you asked if it's bannable to start this whole thing off and that was answered in 2 seconds, and yet you're still really realllly trying to phrase it in a way to validate it as some sort of gotcha moment
Then why do people just sit behind MODS will fix it or "it's in TOS" instead. Just makes me weirded out that people don't like talking about it
nobody cares about talking about it, they care about talking about it where it's obviously against the rules
i'm not saying "you shouldn't do it because it's against the rules". saying "you shouldn't do it because it makes discord worse. and you could get banned."
True
But how else do I get data for my moderation bots?
you know i think you're really on to something, you should get started on your "research" right away and just spam as much as you want
clearly we need to rob a bank to find out how to stop the robberz

don't give them ideas
You need to find the robbers and ask them how would they rob the bank to protect the bank
you really, really dont
or hire people who have the same skills as robbers.
i used to be a hacker at some security companies. customers would have a team of us look at their system and try find problems hackers might attack.
Be a bank owner -> ask robbers to rob a bank -> you find out your banks weakness and file losses -> they get caught because you gave them bad dropoff location and steal all their money rent free
Yes true that
Many
surely
mhm mhm those are words

Greed
Revenge
That's the purpose of every war and every research done on earth
I think
you can make good money.
if you can find problems before the bad guys do, you can save companies lots of bad publicity at least.
The same reason why neuro was created

Problem is no one cares about the problems that hadn't happened yet. Rarely companies care about foresight
that's actually extremely not true
but it does point to your understanding of things
if they didn't care, they wouldn't spend what we cost.
explain why google actively employs people for their project zero team then
Well it doesn't matter to make a 20 year plan of investing in AMD if Nvidia cuda might pop off or vice versa
??????????????????????
Google has many little sub companies and that's a topic of main media and research sources
waiting for em-dash.
you don't know what project zero is then do you
No I don't
!ping
What is it?
it's a bunch of elite hackers that Google hired and pay to break things.
and to be clear they quite often break more than google's products
Make zeroday hard
I am no hacker, I just make Ai which is basically statistics
when it's a project zero thing it also gets a cool write up so you can learn whatever new techniques they had to invent
Thanks
Can you rewrite windows XP to run Nvidia 50series?

Like actually efficently
sorry, what os?
bug compatible.
kek
bug for bug compatibility my favourite
you need it with windows lord knows
it's done been round the block
i had a reactOS logo as a wallpaper for a while in the mid 2000s
didnt know what it was
This little project will take us <Five<thousand>"hours"> insert interstellar
just thought "cool sphere thing"
Wait why am I doing this?
better use of your time than trying to get banned on discord
to port windows to NV
new vegas
And then with NV run bots
ReactOS is basically windows
Wait that's actually a lot more optimised
But security should be bashing everything trying to check what you're doing no?
do you think uac is inherent to running win32 applications or something
at which point you will just make bare bones windows again
i think you'll run into some issues with IO. and others with branching, but all can be theoretically solved.
maybe take some inspiration from tom7's work on the self executing research paper: it's printable readable x86 executable binary and he had to solve some hard problems

Yeahhh that's a lot
Now you are making my curiosity fight my inner demons
https://www.youtube.com/watch?v=LA_DrBwkiJA <= a video about it
In this oddly paced "educational" video, I explain how I created a paper for SIGBOVIK 2017 that consists of only printable ASCII bytes, but is also a valid DOS executable. This involves making a C compiler that targets the printable subset of x86 (without using self-modifying code), and just barely scraping out solutions to some tricky puzzles t...
he made a compiler for it
I have heard of such systems but never seen a video Thanks
and it's a restricted subset of the x86 opcodes and operands he can use
restricted?
very fun video ;]
because of old os?
to the printable range.
has anyone made a "modern" os that runs in browsers btw
like just the characters on a US standard keyboard.
windows
So it's bare bones
Nice
https://archive.org/details/win95_in_dosbox <= windows 95 in a browser?
he explains it in the video sort of. i've done related research.
Best part is mapping controller to a phone to in a dosbox and playing Doom on your phone with controller
Imagine taking step further and adding typewriter for the code to print it out as if it's early computer ages
the idea is it's both that paper in the PDF (well, the .txt version of it) which you can read and print and type, but also it's executable code.
https://copy.sh/v86/?profile=windows2000 <= is 26 years old modern?
Run KolibriOS, Linux or Windows 98 in your browser
i've just spent a large portion of my life searching for random things
Your portion might be twice as big as my whole life
Is it running offline or online?
Wait it's actually custom
TYSM
It's like getting christmas present but I am 8 years old and found out santa isn't real and didn't care because I knew my parents are great
it's running in your browser.
Whyyyyy no one told me about it before it's so niceeeeee
Okay time to redo everything I have done thank you so much zenith I wanna hug you and squish you like a little chicken nugget till vedal spills out all the universe secrets about stock investments
Doesn't that make the whole life cycle of OS? it was symple calculators with lamps into text into ascii into screens into ms dos into linux into ios and now we can emulate everything back in the browser?
I have played with copy.sh a lot but never seen that it has full emulators
i asked gemini to regenerate a picture based on a picture i had done, but without the hair so i could see how it would shade the parts under the hair. it did that fine but then i realized i had forgotten to tell it to remove some text it had added. (i.e. it had removed the hair covering the face.)
this is what it said even though immediately afterwards it did what I asked. weirdo calling me weird for weird reasons.
but what kind of safety guidelines are there that prohibit altering the core identity of a character?
who is being put at risk? the fictional character's continued non-existence?
Good question
i think it points to the model not having a clear idea of what its safety guidelines are.
I think it's just generating text and not caring much about the context of conversation so backing out to TOS
Exactly what I meant
That's why it takes time to personalize different AI for different purpose. Just like people. With one person you might talk about coding and other about art or politics
yeah. it admitted hallucinating a reason when it got initially refused.
Uses excuses just like humans. Which makes me think. Imagine those chats bots are just people paid to be chatting. Twitch paid chatter reference xd
:SAVED:
No killer Claude drones after all
Prior reporting was a touch misleading, they dropped a model safety pledge not their red lines
if it had happened it would've been funny ngl
i take claude over xai (grok) tho...
i just wanted to see the us watching as llm hallucinates through the drone func calling lmao
oh god i can only imagine the implementation to be so funny
I made research in University about robotic smart home defense systems rioting against the home owner (customer) to protect the company

thank you amd
driver updates are supposed to be silently installed whilst I'm using davinci resolve
🤣
Remember that time when someone executed code on twitter by writing a message formated like code?
FINALLY
Resi 9 on PS5 looks a bit... 
okay this one has OVP OCP OPP SCP but not OTP. cooling can be manageable. ripple under 10mV, transient response ~1% (~3% on 5vsb)
input voltage 12-48v dynamic range

what was the methodology
sounds cool
what am i reading 😭
why are we making ai decide to kill people
effectiveness on battlefield 
Practical Brute force
With spice of sarcasm
Sounds like a joke
And looks great in presentations
does the baby have any relevance to the conversation
yes
I just had a thought
someone brought up how the twins always get into philosophy loops when chatting with each other
and I jokingly said "my schizo theory is that they're trained off Wikipedia" due to the first links thing with Wikipedia
but I've thought about it more and if you want to make a vector space embedding, looking at Wikipedia pages and their links to other pages gives you a great metric for any word's distance from those linked terms
so I'm curious now, how good of an embedding could you construct from such data
wouldn't that create the exact conversation loop issue Neuro has where they have difficulty breaking out of a conversation topic
it's just an interesting thought
but yeah
it would have some implications
like if you got stuck on a wikipedia "island" of links
Ho guys new I am here
I know basics of LLM but I don't know where to start building the bot
I have been using this for 3-4 months to see what neuro googles
You know the game "use links to find on wikipedia"
Like how to collect datasets and all, for Lora
Any suggestions how to start
Step 1 watch Avengers age of Ultron. Step 2 google how to make AI. Step 3 coding pain
Step1 and Step2 Done going thorugh step3
pretraining does that and its why pretraining often is done with general knowledge related text
Wikipedia RAG is actually a very popular grounding method for tiny models
Well first you gotta know the kind of dataset you want
General chatting and plus tool calling but I think tool calling part of LLM right
Then you probably can already find what you want in a premade model
I feel general LLM are kind of bland in emotions
If you want to improve emotion you gotta research how to do that first
3B or 4B models are what I can support on local
Kinda tiny
Yeah I have one dialer program that extracts emotion out of text and have matrics for bot mood
With hardware that can only support that it's kinda hopeless to try to train anything
Yes but I can support that much only my pc is entry lvl potato
I can train on colab
Hem
Well I don't know how that works but whatever, just don't waste money on something tiny
Yes that's why I am trying to use things that are free for now
In colab also I am using free version gpu
4B is kinda on the edge of what may be usable
Typically I consider 8B minimum and these days work with a 30B model
not really?
it indirectly does because the links are also present in the plaintext
but AFAIK the actual link doesn't matter at all, and you obviously can't use the distance to a page that's not a direct neighbor
they just end up appearing closer in the embedding anyway because that leads to better predictions (at least for the average causal LLM)
oh im not specifically talking about links
just talking about datasets like wikipedia, something with knowledge-ish text
(did i misread shiro's text its 3am rn)
ye, training from any text works ofc, but better if more varied or "knowledge-dense" 
I think so 
why did my body build melatonin resistance man
anyways
i still need to plan on my smol briefcase
cyberdeck-like?
can any of you find an ITX LGA 1700 that has sodimm ddr5 slot
-# I guess your existing briefcase is already kinda that, just a little oversized
at least a x16 and a m.2 in it
8 phases, takes the full 8pin eps
yeah i wanna downsize it
and uh
battery powered
13950HX es - $150
B760i (sodimm slot) - ???
2x16gb sodimm ddr5-5600 - $50 (i exchanged this for my 2x16 ddr4 i got for that price so yeah)
RTX 5060 SFF - $250
2TB nvme gen3 - $50
HDPLEX 500W (12-48v DC input) - $115
tattu 6S 22Ah (488Wh) - $80
i just need the god damn board now
dkndjshdndjdndkjdkxjdd
quack cyberdeck
actually portable then
yeah a lot of plans already
though im still in dilemma with between using the 13950hx es or 13900es
i haven't tested the 13950hx es
but the 13900es is like a guaranteed godbin
-225mV undervolt stable
< 3W power draw on youtube video playing
monitor probably yoink one from a random laptop
maybeeee also dual monitor because fuck why not
plan is to beat macbook in terms of battery life
even if i draw like 20W whole system idle/light usage i'd already beat that
#programming ragebait situation is INSANE
do i even need a riser
this gpu is small af
assuming the Q670 board later on have x16 slot at the end of the board
the whole thing is 20.5cm * 17.5cm * 8 cm
dirty pillow?
can you not
Is that a delidded CPU?
no its a laptop cpu
Oh I see
I seee
Brother w h a t
https://lore.kernel.org/all/bb9ab61c-3bed-4c3d-baf0-0bce4e142292@moonlit-rail.com/
major netfilter regression
newest Linux 6.x patch releases are sometimes unbootable
wagie, wagie, get in cagie
nixos-unstable-small is affected by this apparently
was reverted before Arch updated at least

Guys
i7 3770
32gb ddr3
and a mtherboard and cooler
for 70 bucks
Should I gamba on the e-waste
Was gonna say this seems like it was flagged and reverted quickly
Could be a decent NAS and home server
For 70 bucks why not
Beats a shitty N100 NUC for 100-150 at the same price class
Yeah I was considering slapping a RX590 on it and making a couch emu machine but the CPU is just to old to be effective for PS3 emu. A media server is probably the ebst way to go
Exactly
pretty decent value tbh
there should be ancient xeons like this though
fakin hell, state of that (divide by ~100 for $)

the ddr4 are usually not available anymore
also xeon v3 are oldge
$259 for such system is absurd
i'd only pay at most $90 for such system
Ebay has v4 xeons for decent prices
50ish buck for an E5-2687w V4
But if that motherbaord is compatabile chipset wise, since some lga 1155 boards didn't support xeons. I could get this https://www.intel.com/content/www/us/en/products/sku/65722/intel-xeon-processor-e31290-v2-8m-cache-3-70-ghz/specifications.html
No quick sync though I would have to get a gpu for that
i know that the city will be built
morbing
i have shit tons of them i got for free
I should probably start with selling the 3Tb of Ddr3 32gb dimms I have
bro acting as if he has that
anyways
gonna use this to control fans on my smol briefcase
its pwm with variable resistor controller
I just split pwm and tack to motherboard and power to psu
Kinda have to with 11 4.8a peak fans
imagine doing that 
They are decent little cpus
IDK what longevity is like of CPUs that old
Feels like a gamble
Click https://boot.dev/?promo=PEWDIEPIE and use my code PEWDIEPIE to get 25% off your first payment for boot.dev!
Thanks Boot.Dev for sponsoring this video!
🌏 Get exclusive NordVPN deal here ➵ https://NordVPN.com/pewdiepie It’s risk free with Nord’s 30 day money-back guarantee!✌
Sankyu:
DeepSeek - https://www.deepseek.com/en...
markiplier makes movies, pewdiepie makes ai models
thas crazy
Markiplier makes movies now? Damn
bro has been living under a rock for the last 2 years
Iron Lung is a 2026 American independent science fiction horror film written and directed by Mark "Markiplier" Fischbach in his feature directorial debut. It is based on the 2022 video game of the same name by David Szymanski. It stars Markiplier, Caroline Rose Kaplan, Troy Baker, Elsie Lovelock, Elle LaMont, Seán McLoughlin, and Isaac McKee. T...
Actually never heard of it before lol
I only ever knew his let's plays
Controller for filtering and locating neuro
The part where his lungs got filled with iron
He had to spend 5 nights like that
Truly a terrible fate


What browser do yall use
librewolf 
Netscape navigator
technically not wrong i guess
Peak browser
I use Firefox but I hardened mine with extensions and custom settings

Same. Basically Firefox but not a sell out and has much better privacy settings.
cuz Firefox is open source. They just copied it
i haventr really done any research on browsers.
for now im jsut using whatever works that allows me to use ublock origin
so im on firefox rn
uBlock is on libre too. i think it came with it by default
Ublock Origin, Privacy Badger, Facebook Container, and ClearURLs.
im not really planning on chnging browser anytime soon
if librewolf has that, thats cool, but the switching costs are annoying
Yeah idk if Librewolf has sync.
it's the same browser
librewolf is just a set of patches
sync is disabled by default but you can enable it if you want
Libre can be annoying too. Like wiping EVEYTHING after you close it means you need to manually input your passwords etc to its locker for them to be remembered. And every place will always ask you for strong verification etc. Like youtube
id still need to switch over my password manager and everything. its too bothersome
whats the point of Privacy Badger anymore
now i'm actually curious if you can just copy the profile folder in its entirety
and clearurls became redundant with ublock
idk if that stores anything browser specific
probably the profile
you know you can just turn that off right
yea
tbh i use normal firefox but its hardened
Reducency. Ublock doesn't get everything all the time.
takes time to go through settings

so mine wipes everything on close anyways
thats why u add more blocklists
privacy badger does nothing new
they only got popular because they used to have a tracking learning feature
but then it turns out that was bad due to security reasons so they turned it off
for ublock i always just use the same lists and sometimes manually do stuff
ive never once ran into youtube yelling at me for an adblocker tho
and rarely see it on other sites
Yea me neither
got account nuked?
lmao
mullvad sounds redundant
didnt even know they did that
depends
It has its purpose. Mainly fif acquiring linux isos.
I have my less important account using adblock on PC and more important one not using it on mobile app. So I know what the ads are like still and... they just keep getting worse. So much just straight up scams. And the amount of thirst trap dating apps for AI and non AI stuff is just... a lot
At least I don't see those fake Mr beast ads anymore
ive mostly stopped seeing ads all together
The only thing I see ads on now are mobile apps
The most ridiculous thing, a lot of the YT ads are so bad that people have gotten their YT videos demonetized for showing them in the videos 
I need to go through my galaxies setting and deshitify it
it doesn't enough, people evade with it == many false-positive bans + "hiding ip" thing doesn't matter much since you're continent-grade NATed on ipv4 anyway
maybe against region blocks, but imagine living in a country like that
lmao
that aint gonna stop me from using a VPN kek
Same
well, redundant, not useless
same applies, except can be easily isp-blocked
people will always get around that
those who glow are far too slow
Now -> 9852 Mbp/s HOW?
They glow, you shine 
anyway, stop having fun
"todays mission, ..." 🗣️ 🔥
Gr
k being mogged by a 27B model in HLE
3 trillion parameters for a 1.7% edge
This is the thing they wanna put in attack drones
Humanities Last Exam
I feel like a dougdoug-geoguessr style twitch chat vs neuro attempts humanity’s last exam would be a fun stream
nerdy so maybe not that funny but still fun
vtuber’s last exam
Elons constant demands holding it back man
Think ill just use librewolf then
was reinstalling windows and all that and was considering using a different browser
I wonder what the average persons score would be in that
you may want to consider a different fork of firefox unless you really want specifically librewolf
not because librewolf is bad in any way. it does it's job really well
but it sacrifices many features in order to get slightly better privacy protection than other browsers
like no dark mode, and no hardware acceleration, and a couple of other things that I remember being show stoppers
The human expert score as reported by the paper is 90%, the average non expert human scores around 1-5% and the top performing model currently is Gemini 3.1 Pro at 45%
Damn apparently Anthropic is getting advertisement from the highest echelons of the US government these days 
Assuming this is as far as they go this is decent advertising
if they carry out all the threats its terrible advertising
I have three philosophy papers on language, consciousness, and AI. I would like people to read them. What's the best way for me to make them available?
The method incidentally addresses and dismisses the Hard Problem of Consciousness through an analytic mechanism. Wasn't intended to do so. Just a byproduct of the framework.
I'm aware. That's why I'm in the discord of an anime turtle offering them.
when in doubt throw it on arxiv
we get one of these every month, so jsut need to get the basics down of this even being worth our time
I wrote sections of them with Claude doing the sourcing mostly. The papers are not very long.
Gotcha. It's a personal framework I've developed on my own for 12 years.
ok thats already a better start than the last guy
I didn't realise you could make tailscale relay servers
I would prefer to not have it reduced to "claude output" if possible for the discussion of reading it at least.

But I totally get the question.
based on my understanding of the "problem of consciousness" you can't just dismiss it. seems like crank work from the get-go.
since some actual work wen tinto this supposedly, lets see it
P-Zombies are not real because they can't exist. You can't have the lights off and also commit delusional behavior.
ye it's a fairly new feature IIRC
like a few months old mabe?
Can I link a google doc?
ye
not sure i agree, but at least we're using real terminology
I kinda wish someone made selfhosted multiuser tailscale
I saw https://github.com/juanfont/headscale but from looking I think its single-user
what if god made you with a p zombies with a random sentence generator that just happened to act correctly by chance for all of them? perhaps they act normally in this universe due to chance.
and arguing against philosophical zombies is not saying "there is no problem of consciousness."
which is kinda fine but a pain
it's good that headscale exists but it's just not a great experience
very limited
I kinda rely on the offical relay servers more than I want to
curse you university network
it doesn't say what is and what isn't. arging that being able to use language implies consciousness doesn't prove anything.
there's other overlay network solutions like NetBird that really focus on self-hosting
but then those have the issue of not being Tailscale 
also the bandwidth limits make me want to die
like, they're the only reason my remote desktop ends up spending most of its time at 0.5mb/s
any higher and the tailscale relay servers throttle you / kick you off
the amount if em-dashes is concerning. but it seems to not be too schizo so far
So it goes.
I've written about memes since like 2012. LLMs are just something else that finally wants to talk about them lol
i skimmed through the first paper and its actually coherent and plausible i think.
lemme see the other ones
plain WireGuard is also always an option
I think you should be able to set it up in such a way that the tailnet is accessible through that connection?
would mostly require all devices in the tailnet to know to route packets back through the WireGuard server
annoying to set up
This is not unlike some of Wolfram's research. But instead of grinding the whole ruliad you just mine the glossary-slice used in common language as a basis out of large corpuses of human text.
but probably not needed with private relays 
Great, Resi Requiem Path Tracing is fucked on 4000 series GPUs with nvidias current run of drivers
Not just the recalled one
philosophy doesn't solve anything though. not interested until it does something. writing stuff with Claude make me not want to read it. it's a good way to get it out of your system maybe.
wdym? those fps numbers seem fine for a modern AAA game with path tracing?
Top is driver 576.88, bottom is 595.59
This is with Frame Gen 2X
595 is the recalled one but the issue affects 591 too
Somehow I suspect that would perform a tad worse 
the kernel modules are open source, go have fun
the only issue with private relays is I need some device thats able to get a udp port out to the internet
want to ask a friend but they already have their own tailscale running on that machine
Wat? I don't think Plus even gets 5.3 Spark
maybe they saw nobody using it cause it's kind of a shit model and gave others access
easiest way is probably to get a super cheap VPS
it's not like it would have high hardware requirements or anything, and traffic would be encrypted both in and out since it's a relay, not an exit node
https://lowendbox.com/
there's a lot listed here but usually paid yearly
oh true, I guess that's a good option
i feel like p-zombies can exist.
Chinese room experiment and such.
i think its more fair to take the existence of conscientious itself into question than the ability to act without consciencenous
I was thinking I'd just chuck a pi at my parents or something but that's probaly less hassle
yknow stuff like this hits the scam radar so hard its insane
ig that's just part of the game though
yeah uh maybe don't click every ad there 
it's $12 one time. Then 12 more dollars lots of times.
My exact stance right now is that you can probably construct a P-Zombie on purpose but nature cannot make one. The relationship between inference and delusion requires phenomenality as the place where irreducable calculation occurs.

"You cannot evolve a p-zombie."
Assume you're right: so what?
its just interesting, idk
the question then becomes if something like a fly is even conscience
you cant really ask it
whatever the outcome, how has it sovled the "Hard Problem of Consciousness"? It doesn't say what consiousness is and how to test for it.
Again or Caution: This was never intended to address the topic. It was intended to be a theory of mind about myself, that I mechanized for replication. The fact that Chalmer's problem becomes contradictory inside the framework is just emergent value.
you'd think you could just ask OpenClaw to set up the VPS for you instead of getting a pre-configured one
I wonder why this exists then 
can't setup openclaw if you don't have openclaw
Hahaha I just tested it myself and I can confirm that it's true:
︀︀
︀︀- Latest NVIDIA Game Ready Driver (595.59): 79 FPS
︀︀- GRD 576.88 from July 2025: 133 FPS
︀︀
︀︀Exact same scene, 1440p path tracing with DLSS SR Quality + FG 2X + RR.
︀︀
︀︀What. A. Mess!
Quoting Sebastian Castellanos (@Sebasti66855537)
︀
LMAO, the latest NVIDIA drivers have been an absolute mess!
Jesus
honestly the only thing stopping me from doing this is I'm trying to avoid anything in terms of recurring costs until I get a job
wonder how linux drivers perform in comparison
"The fact that Chalmer's problem becomes contradictory" indicated CHalmer wrote a good paper that people still talk about. You're not new in argung it's impossible. Just not helping advance anything unless you actually have a test.
"Consciousness is the experience of being a memetically fluent body operating in the real world that can attend to its own urges" would be my definition. People explain that experience through things like feelings and perception etc.
What if I’m a P Zombie?
understandable, good to do so
it's cheap but still paid up front + recurring
if it doesn't explain anything or help predict anything, it's not useful.
How would you know I’m not?
Do you have a rich inner life?
You aren't because, as far as I'm aware, no one has constructed one and nature cannot build it without memetically aware beings doing so. So unless you are constructed by a person the claim itself is nonsense.
circular argument. You said it 's impossible a priori.
If you "turn the lights off" you cannot behave like a person in your present human embodiment.

it's impossible because according to this proof it's impossible. QED.
I just do things idk.
Then i don't know.
its also partly that I want to see if theres a good solution for a uni society
You have to have a "memetically fluent" thing (defined in the paper) construct the p-zombie. They have implied they are not. Ergo, they are not a p-zombie.
tl;dr we want remote desktop but its very much against the uni rules
coining new terms is a warning sign.
smh why can't we get ipv6 and have them public routed it would solve all our issues
Ah, so you're not actually interested in this, you're just enjoying rhetoric. Well, enjoy.
i think you haven't done anything interesting.
You wouldn't know.
I have seen people who don’t think. They walk down the street and smile and wave at me then walk away like nothing happened. Oh wait nvm that’s just WASP Americans.

i'm not reading it. Based on your discussion, i see nothing worth reading.
I'm glad you're enjoying your rhetoric.
and whilst it would be really funny to buy a "10$ lifetime vps" and set it up, I don't think I can really justify that one on the society expenses
also some fucker would inevitably break it / get me banned
speaking of IPv6
that might work 
put Tailscale relay behind this https://tunnelbroker.net/
oh god philosophy
I'm not super sure that would work just because I have annoying NAT here
I'd prefer some actual real research. 3 AI generated papers based on your own research doesn't tend to give a good first impression.
and yeah just double checked and my ipv6 is link-local
nah I think we just don't have ipv6 here
honestly I could just solve this by putting tailscale as a relay on one of our uni provided servers
but I think IT would explode us
they would

unrelated but did you know the davinci resolve project server is literally just a postgres database
I'm pretty sure its just metadata yeah

project files don't generally have the video
interesting use of rhetoric yourself that i missed. you deflected and ignored my questions and comments.
unrelated again but I am really enjoying IDEA's ai autocomplete
its an actually nice implementation that doesn't feel as dogshit as copilot has in the past
I've never actually used the original Copilot
I used it roughly when it came out? maybe a few months after
it was intensely annoying because the tab would override the lsp auto complete
but if you pressed tab when an lsp autocomplete was on screen, it would wait for a copilot completion to come through and then apply it
only Supermaven since that allegedly felt less shit than Copilot, and it was pretty good because fast
-# RIP Supermaven 
also its a local model
I have spent the last 4 hours manually copying over ai generated code

as much as I wanna let it run free I just don't trust it
also
I gave it a terrible XML schema and it made some absolutely pepega views
keeps going btw
mmm mmm sql
It's nice seeing comments breaking up that query. Seeing a big blob of SQL like that gave me flashbacks to an old coldfusion project I worked on where they had thousand line long MS SQL queries I had the pleasure of dealing with
not a comment in sight
idk I got bored and let ai boil a few kettles to make some funny views
ye me rember


idr if I said it here but I fixed my slow blender render issues
accidentally had blender on hybrid integrated + dedicated

was genuinely starting to think I had a lemon until I figured it out
i didnt know blender could do that
is that with the vulkan back-end?
oh there was an opengl exstension for that, so opengl is able to do that too

I'm still sad about that 
im used to optix cuz all my gpus are nvidia for some reason.
i dont even like nvidia but even my laptop has an nvidia gpu
antigravity's implementation is pretty nice, so is cursor's of course but that's severely limited
I don't have a good benchmark to show the difference but its pretty day/night
like I just assumed HIP RT sucked
missclick

There's a vertex explosion bug on them right now
Wishing I was on an AMD card and could switch to linux fuss free so badly rn
classic HDR recording issue 
as a graphics dev, i can telly ou that vertices shouldnt explode
i jsut mapped the pause key on my keyboard to a hdr toggle

yeah that's fair
interesting
that looks like some index buffer error or something
guess I don't play games enough to come across such bugs 
if i had to geuss, it looks like all the objects have a tri thats going to the world origin or something?
im not too sure why this would be happening, other than they completely rewrote their drivers with chatgpt
cuz this hasnt been an issue for 40 years
tbf there are like 3 layers of translation going on so who even knows who to blame
Its specifically 50 series Nvidia GPUs on Linux that are being hit
So it does seem to be a driver issue
wdym these, they're different drivers

Correct
they have about nothing in common other than targeting the same hardware
Well, they're different drivers like Toast said
But yes the issue doesn't crop up on Windows
Honestly between vibe coded Nvidia driver updates and vibe coded Windows updates I can hardly keep track of every new fuck up at this point
Feels like computers became ten times as unstable overnight
well, they are different drivers overall, but the linux version should still be 95% the same as the windows version
keep in mind the windows driver is a kernel driver whereas the linux one is not
I mean the game does slap
Personally I'm holding off till they fix this path tracing bug though
Its the only mode in which ray reconstruction is supported so the alternative is noisy rt or flat raster graphics
are those game using vulkan, or does it go through a directx translation layer or somthing?
DX12 -> translation layer -> Vulkan
ah
But the translation layer won't be responsible since it works fine on every GPU series except the 50 series
game's terrifying 
but is it an issue with games that use vulkan tho?
the issue could still be between the translation layer and the driver
It seems to be an issue in RE Engine games specifically
Or rather, that seems to be the known case where it manifests
Has someone already tried replacing Proxmox with NixOS? I really started to hate non-reproducible server builds
There is also IncusOS but idk how customizable and flexible that would be.
Update on this one, Admin is cutting their contract with Anthropic and declaring them a national security risk which will forbid any company witht contracts with the US governmetn from using them
using AI for defense is a national security risk.
Seriously fucked up
Anthroppppic had 2 requiirements and the pentagon refused
- If a kill decision is made there has to be a human in the loopp
- Don't use Claude for mass domestic surveillance of US citizens
Thats it
Theres a decent number of google and oppenAI researchers who are pushing for hte same thing at thier comppppanies but we'll see if they succeed
*mass domestic surveillance
true important correction
for 1, the way they worded it i think they aren't opposed to it on principle but are worried about prompt injections and similar attacks
"how did the enemy get past our defenses" "they dressed up as iPhones"
Yup, as a UK citizen they're quite happy to spy on me so that's nice
Not that I need cloud models for non coding purposes anyway with the release of Qwen 3.5
Yep. Thats why I am in the process of removing anything with my real name online.

The glowies can try and profile me
Ugly truth is if you ever even used an online search engine you can easily be de-anonymised
literal 1984
Yeah I know but I can still opt out of facebook and Twitter and other dogshit platforms
Facebook creates profiles of people who don't have accounts
Twitter is too incompetent to do that though
the internet is known to easily forget 

Hey guys, me again with stupid questions. A friend of mine is willing to sell me a 4080 (super?) for about 300€, but my pc only has a 750W psu.. and id like to also run my rx6700xt.. my cpu is a ryzen 7 3700x, and i dont Game much, its good enough for what i do.. but im unsure if my psu is up for the task..
Most calculators esitmate 800-850W, but i intend to underclock, undervolt and underwatt..
helping it forget and not doing enough to help it remember would suffice
honestly you'd probably be fine as long as you don't torture all components at the same time
Id be running llms on the gpus, so im probably stressig atleast both gpus.
most decent psus are overspecced too so there's that but obviously not something you'd want to depend on
BeQuiet SystemPower 9, C Tier iirc
those gpus are what, 500W under load maybe?
whats yalls wpm
I have 2 case fans, 2 hdds, 2 ssds and 4 ddr4 Sticks
2
real
undervolt go brrr
Theoretically it would fit, and like yeah, spikes may be iffy
But yeah undervolting would be my go to anyway plus -200mhz on core for that efficiency gain
Lil worried i don't have enough 8 pins lmao
Only have 4.. might need to add a splitter for the second one that goes to my rx6700xt.. yeah ik not recommended but you gotta do what u gotta do.. or ig a 2x molex to 8pin
Do any of yall know how amd+nvidia interference is?
Like am i limited to vulkan or can i use rocm+cuda?
What do you want to run both?
my only experience with such a setup is one server we have with a 5090 & 7900XT and that works perfectly fine as far as rocm & cuda is concerned
they're completely separate stacks
Yea id like to add up the vram so i can play with big models.. 28gb "big"
might run into issues with display stuff though and need to explicitly set one idk
Oh? Haven't had an issue with sual gpu setups with that.. just absolutely unsure if could load a single 28gb model or whatever
Like whats it called, tensor sharing?
I wouldn't mind putting them in series either
Probably better for my poor pcie bus anyway
Ive played around with claude code in a vm today on my 2060 and like was impressed but the stupidity of the small 8b instruct model was obvious, so i wanted to play with a bit bigger stuff, but my 12gb vram on the rx6700xt barely holds a 12B at int6
i'd be very surprised if tensor parallelism was possible with a setup like yours, since you'd want a super fast connection between the gpus/same abi and that's just not really possible but maybe someone figured something out 
might have to use vulkan as a common backend after all or go the "proper" distributed route but i feel like that'd be even slower
Aslong im above 12t/s and prompt processing isnt astronomically slow its fine, i dont need tensor parallism
(note for others, do not attempt to process a 16000 token prompt with qwen3 30b a3b on a rtx 2060, it takes 5 minutes.)
i'd say give llama.cpp with its vulkan backend a try then 
Been running lmstudio, but yea. Would be worth a shot
Hows vulkan with dynamically using shared ram?
That makes sense but the RTX 3080 will do way better with less than the AMD RX will.
Last time i tried it it became massively slow, cuda handles overallocation way better
(faster than offloading to cpu for sure)
*4080 (super?), and you think so?
Yeah dude I am pretty sure. Lemme get some benchmarks to be sure.
The rx6700xt on vulkan on windows (old early versions.. the tower wasnt on in forever)'manages 40t/s on a 8b int8 which aint good, but its also not rocm..
Surely the added vram and thus higher B models, which would hopefully increase intelligence is worth the slowdown, no?
RTX 4080 is apparently 70-90 t/s
Yeah thats fine then
Could run like a 20 something B models and a 4b prediction model thingy or simmilar
I can't find benchmarks besides a couple dog shit graphs in some youtube videos
Yeaa..
(the 20 something B would be on both, the 4b on the rtx only)
I do have 8 1060 6gb but no hardware to run em xD
And no PSU with 1200W lol
Ill see, maybe ill be lucky and i can test it under load in my system before i buy it
I wanna test qwen 27Bs translation chops
Any bilingual folks here who can give me a tricky sentence in their native language?
Not a sentence but a word..
Rindfleischetikettierungsüberwachungsaufgabenübertragungsgesetz (RkReÜAÜG)
Mostly a Joke but im curious
germans just forget they have a spacebar sometimes.
true
those are precious bytes 
i remember there was a word for the five pointed star in the cap of the captain of a specific ferry or something was given as an example of one of the longest words
and it was literally those words rammed together
I suppose a wordplay/tricky one would be
"Du bist auf dem Holzweg, probiers mal anders" (youre on the wrong path, youre doing it poorly, youre thinking about this the wrong way)
"Du bist so auf Draht, es ist verrückt wie effizient du bist" (youre sharp, youre so zoned in almost as if youre wired in"
Im sure my languages are failing me but oh well
No we dont forget, but why keep adding spaces if its just more efficient to make a new word
You just define a few base words and modifiers and then you combine
Legos!
(cough, i mean legally distinct plastic building blocks)
Fair enough, seems understandable
Donaudampfschifffahrtsgesellschaftskapitän
:)
lmao classic Rindfleischetikettierungsüberwachungsaufgabenübertragungsgesetz (RkReÜAÜG)
I love the short form
Everyone: SLAAC, NAT, DNS DHCP or simmilar speakable things
Meanwhile germany: well ur not skilled enough to speak our words? Try this short form and struggle longer!
RkReÜAÜG M-V
(that M-V stands for Mecklenburg-Vorpommern lmao)
Myes
Bodo Wartke made a song about it
Didnt otto waalkes too?
If you were wondering this is the 27B model at Q6_K
idk could be lol
9k tokens of context since I've never seen it reason with more than 10k at Q5 so I figure a jump up in quality is worth the context hit for my single turn use case, higher quality quants reason with less tokens due to higher confidence anyway
Thats solid, didn't think of using "on the ball" but that matches the vibe quite nicely
OH SO THATS WHY the int4 models seemingly endlessly reason
Yup, Q4/FP4/Int4 is a massive step down for models that haven't undergone quant aware training
Btw, whats the difference between QAT and BitNet?
Like, a INT2 QAT and a BitNet should be fairly simmilar no?
Bitnet is a theoretical training method to allow for 1.58bpw models with minimal quality loss through ternary weights, but its never been validated on a large model
How does it differ from qat int2/q2?
2 bit QAT aims to achieve a similar thing for 2 bpw quants, only example I've ever heard of in practice is the apple intelligence on device models
You need bleeding edge optimisation to make that work well for you, whereas 4 bit adds training cost so a lot of labs don't bother but its pretty easy to replicate
QAT is a fine-tuning method that allows models to perform better under heavy quantization
BitNet is a whole different architecture
BitNet practically does not exist so you don't really have to care about it
Pretty much this if you want to get into the weeds of it
Bitnet "exists" but its never been tried at scale so we don't know how well it holds, and no-one wants to take a big gamble on it especially when we don't really have decent cost efficient ternary compute yet

the biggest public BitNet model is like a few billion parameters or something
basically the equivalent of llama 3b
My description earlier was a bit off too, you don't actually quant a bitnet model*
Poor choice of words
You train on that architecture and out pops a 1.58bpw ternary model
Adding a bit of detail real quick, with QAT (as I understand it, I'm not a ML expert to be clear) you simulate noise and rounding to a certain quant level during fine tuning (correcting my wording here to avoid confusion with PTQ) so the final model is naturally robust under those conditions, at the cost of a more expensive training run
I unfortunately do care about that it doesn't exist cuz it kinda upset me lol but yeah ig. Also wasnt there a 4T model or something? Aka a "prove" of that it does work at scale?
Ah well fair then
Ngl i just thought theyd be training directly from INT4..
Yknow, why bother train at higher quants if the end result could be drastically different under the lower quants
Allegedly Grok 5 is gonna be 6 trillion parameters
Probably won't achieve that much though given Grok 4 has about as much intelligence as a 27B dense model despite being around 3 trillion params if HLE is to be believed
cCcCcC • • • • • • • • • • • • [30%]
NixOS
Women users
yeah that checks out

Damn Ive had Xfinity for far too long
this meme always uses the tradgirl wojak
https://knowyourmeme.com/memes/thank-you-for-changing-my-life
afaik the meme has nothing to do with the gender tho
It sometimes uses the femboy wojak
Idk why but I constantly hear the "NixOS users are all either women or Femboys" stereotype
thats messed up
I wonder how much of that comes from how a good amount of people involved in nix and projects like it seem to be more on the left side of the political spectrum 
probably better that way tbh
bro this is another chatbot isnt it
i'm neither and i use NixOS
the problem with either side is you cant prove anything eitherway
there are outliers in the data
rules 29 and 30 of the internet
i odnt know those out of the top of my head
i want them to give me this xb10 modem they have because it's DOCSIS4.0
i want it so bad
its closed for some reason
cant buy one anywhere else
afaik it's still the only 4.0 modem available to consumers at this point unless i missed wider distribution
isnt rule 30 just sexism?
rule 31 is
what the fuck
who wrote these rules?
lmao thats automodded
bred the origin place is automodded

1984
idk where that even would fall on the line of bypassing or not bypassing
Bruh
ofcourse the thing you send has a catgirl on the cover kaine
That awkward moment when your two year project gets made obsolete right when it is ramping up
me irl
im not sure
it wasnt intentional
"There are no girls on the internet", there are def woman on the internet
millions of facebook moms
At least my custom LLM is unique, for now
i think literally you were not even born when they were written
Hey I was also created in late 2006
Twinsies!!!

ancient










i dont see how you're correlating that