programming

[mighty thorn]

I think we should jointly hack into anthropic, raise the usage limits, then exit out without stealing any data or doing anything malicious

[fast pagoda]

... raise the usage limits... without... doing anything malicious

[silent cloak]

Didnt implement the make it work feature

[nocturne olive]

So why may my filter be nonfunctional

[glass flower]

bad object type?

[nocturne olive]

It would have a compiler error if the type was incorrect for what the function expects

Am I just supposed to sleep with a random useless Creep sitting around doing nothing because the filter doesn't work

[glass flower]

you could solve it for now and then fix it properly later by using native kotlin filter options

and then calculating the range yourself

i found this on the screeps discord

[nocturne olive]

I guess this works for now

[glass flower]

performance should probably be about the same i'd guess..

but that it works

[nocturne olive]

My repair creep works now

For some reason the code feels silly though

Cool emergent behavior, the harvesters are feeding the repairer because earlier I told them to give energy to creeps with space if they and all storages are full

Now I should sleep

[glass flower]

or improve your screeps further

[nocturne olive]

I've been watching the other room with the one exit that's nextdoor to mine, and it looks like they're about to lose a level on the controller thing

[glass flower]

they are at that point for a while already

[nocturne olive]

The creeps in the room are all running around uselessly repairing a single container it seems like

[olive sable]

[glass flower]

when i last looked they were around 3.6k or something

[nocturne olive]

130 left

[glass flower]

[opaque wharf]

Almost done

[nocturne olive]

I may just watch what happens if a controller downgrades because I'm curious

[glass flower]

and the way that guy has setup his creeps is the reason i like single-role creeps instead of super creeps that do everything

[nocturne olive]

It do be hilarious to watch

[glass flower]

there is a creep going to the upgrader

is he going to get there in time

[nocturne olive]

Na it's just gonna repair the container

[glass flower]

he turned around

[nocturne olive]

Or a random road

[glass flower]

[nocturne olive]

They're just running around repairing things

[young plover]

The huge build pipelines at work are defined in Kotlin and the syntax is black magic
lots of this:
https://kotlinlang.org/docs/type-safe-builders.html

[nocturne olive]

The syntax is super simple

[glass flower]

[nocturne olive]

Obviously this is Screep behavior code

25

[glass flower]

i love how they are like right next to it. but obviously the code isn't prioritizing it so they just move away

downgrade

[nocturne olive]

The extensions turned red

Or started flashing red

I assume that means they're nonfunctional

[glass flower]

i think so too

look at the room above yours

[nocturne olive]

Oh and now they go put stuff in it

[nocturne olive]

[fast pagoda]

before in your pic, after i'm done :

[nocturne olive]

My code has a null error

[glass flower]

[opaque wharf]

[nocturne olive]

What am I looking at

[glass flower]

spaghetti

[glass flower]

also the flashing red... that room is abandoned. the controller fully downgraded. so all the structures in that room are now free to take

[nocturne olive]

[frozen igloo]

I wonder how feasible it would be to have Neuro or other llms write genomes for horsey game.

Either tell them what each gene does to see if they can make something functional or don’t for the funny

[nocturne olive]

I caught the error once and now it's suddenly gone

[glad path]

what happen

my wine is broken

steam wont staart

[mighty thorn]

Yes

Very good job

[glad path]

does anyone know how to fix this

it doesn't render the ui

[tropic spindle]

There's your problem

[glad path]

kindly leave

the problem is most definitely not that i'm on a mac

if it were i'd never have had been able to run wine in the first place...

k

[mighty thorn]

I have an idea!

[glad path]

i would use linux but i cant

[glass flower]

whats the issue in this picture anyway?

[glad path]

i think its steam

[glad path]

steam

its all black

it wont load properly

[glass flower]

oh

[glad path]

over wine

[glass flower]

yeah it does that sometimes

i had that on windows even

[glad path]

how do you fix it

[glass flower]

close it and start it again

[glad path]

i did

it won't go away

[glass flower]

like kill kill it

[tropic spindle]

classicA CEF

[glad path]

i did

[glass flower]

even the steamwebhelper processes?

[glad path]

i killed all the wine processes

[glad path]

yeah

[glass flower]

well...

[glass flower]

do you even need to run steam through wine?

[glad path]

yes

i can't play a lot of games otherwise

[glass flower]

what

[glad path]

most games dont have macos support

[glass flower]

well.. but steam should then use proton no?

[glad path]

no

that's for linux

[glass flower]

[glad path]

macos only has wine

nothing good for macos really

[glass flower]

surely there is a way to select wine inside of steam or something...

[glad path]

what

[glad path]

i can see

no

its not a thing

[glass flower]

i personally only see proton that i can select.. but why would i also need to select wine so..

well.. then i can't help

[mighty thorn]

@olive sable check the fps on that square!

[glass flower]

try reinstalling steam or something... idk

[glad path]

i dont want to have to reinstall games too

[glass flower]

only steam

not the games

[olive sable]

i have been pinged

hello

[mighty thorn]

Square

Look at the shape, triangleboy

[olive sable]

since its an image, id have to say approximately 0 fps

[glass flower]

isn't it technically 0.000000000000~1 fps

(with a lot more 0)

[olive sable]

1/inf fps

[glass flower]

[tropic spindle]

*0.016666666~ FPS

[olive sable]

prove it

[glad path]

what is it doing

[glass flower]

its best

[glad path]

like

[olive sable]

thats not how fps works 🥀

[tropic spindle]

Actually wait that would be per minute

[opaque wharf]

Moar cabling

[tropic spindle]

Stfu it's 1 AM

[opaque wharf]

Whoever needs to maintain this will curse me tho

[olive sable]

frames per second. 1 frame per ever, so 1/inf

[opaque wharf]

It's me. I am the one who will maintain this

[tropic spindle]

Just eat the cables and then you have no more to manage

[olive sable]

this woudl be the frametime of a 60fps video; 16.666667ms

[tropic spindle]

If iron deficiency is anemia, surely there's something for copper deficiency too...

[tropic spindle]

What is this mysterious 60 FPS you speak of

I only know 7 FPS

[olive sable]

wdym?

[tropic spindle]

https://www.intel.com/content/www/us/en/products/sku/81071/intel-celeron-processor-n2830-1m-cache-up-to-2-41-ghz/specifications.html

CPU that I daily drove for 7/8 years

[olive sable]

huh

its even worse than the celeron N4000 i had for 5 years

[tropic spindle]

buh

Cheap ass government secondary school laptop

[olive sable]

how tho? the N4000 was 2 core 2 thread 1.1GHz
the N2830 was 2 core 2 thread 2.16GHz

it should be better

oh

boost clock

4000 boosts to 2.60, 2830 boosts to 2.41

[tropic spindle]

Boost clock more like roost clock

[olive sable]

what do you have against boost clock?

[tropic spindle]

That my current laptop refuses to fucking stay on it

[olive sable]

i hate these cheap laptop cpu "4.8 W" like bruh my phone's cpu uses more power probably

[olive sable]

ye thats why its a boost clock, it only boosts to it when ti has the thermal headroom to do so

[tropic spindle]

It's not just a thermal limit

[olive sable]

power limit and such aswell ye

[tropic spindle]

It's an artificial power limit

Changing the IMON slope was SUPPOSED to get rid of that

But now it's apparently a "current limit"

[olive sable]

545 in passmark

[tropic spindle]

ElivGlare

[olive sable]

now with my current cpu next to it

only 83 times faster

it has 8 times as many cores tho

so only 10 times faster per core

[olive sable]

ddr3 is fine i guess, but only 4 lanes of pcie2???

pcie 3.0 came out in 2010

[tropic spindle]

And would make no difference in this system

[amber fractal]

Only 4x slower than the average SSD

Well maybe 2x depending on who is running a old PCI gen 3 SSD like I was for awhile

[tropic spindle]

This is my current main system CPU at least

[olive sable]

decent i guess

[tropic spindle]

Which is much better except for the stupid ass fucking power limit bullshit

3.90 GHz my ass

[tropic spindle]

Something is wrong with that test

A 5950X should be at least 50-60x faster in single core perf

[olive sable]

its jsut passmark being shit

in geekbench single core, the 5950x is 13x faster

it wont get to 50-60x

5950x is mainly multicore focussed

[tropic spindle]

Yes but the Celeron is nothing focused

[tropic spindle]

Still seems very low

[amber fractal]

Celeron is "Low powered Ass" focused

[olive sable]

i doubt the testing conditions were the exact sine, but the celeron has no hyperthreading so that accounts for it partially

[tropic spindle]

That CPU was like 7W, I've seen lower wattage with more horsepower

[olive sable]

out of the 105W going to the 5950x, thats only 6.5625W per core

[tropic spindle]

105W?

[amber fractal]

That is why the last part was added

[tropic spindle]

I thought that was like a 200W CPU

[olive sable]

tdp is 105W

you can OC to 150W with some luck

[tropic spindle]

Fun fact, that CPU can still run a Minecraft server with close to 200 mods pretty decently for a couple people

A Minecraft server needs a tiny fraction of the computing power the client does

[amber fractal]

Minecraft is single threaded most of the time after all

[tropic spindle]

Actually that goes for most games generally, unless you have a lot of people online

[amber fractal]

Isn't like it is doing hard physics or anything

[olive sable]

ive seen people here reach 200W on a 5950x, but mine has no such luck. the mobo can handle it and the cooler should too. so jsut silicon lottery

[tropic spindle]

Yeah but the ridiculous part is that the server can actually perform quite well

Granted I only play 1.7.10 and modern is more unoptimized in the modded department despite being better in the vanilla department

[olive sable]

not that weird

[amber fractal]

I've seen the performance with my own eyes

Considering I have ran 16 clients at once on a laptop with 2 LAN servers

[olive sable]

single threaded so that cpu performes probably as well as my xeon server in minecraft server

[tropic spindle]

I ran a 170+ modded server on an even worse Celeron from like 2009 or so with 4GB of RAM

[amber fractal]

These were not headless

[tropic spindle]

And it ran fine (except while backups were happening, then the RAM limit caused a lot of lag)

Would have probably been better if I wasn't running it in a KDE environment

I'm always caught off guard by just how tiny the footprint of a Minecraft server is when you consider how dogshit the client performance is

[olive sable]

in signle core my xeons only do 5x the performance

in multicore tho... 2x28x

[amber fractal]

Compared to the average 1.21.1 user deciding to go balling and do 128 chunk render distance in 16gb RAM. Doesn't Suprise me one bit

[tropic spindle]

I like using old laptops as servers. They're low powered and stackable

And if it has a working battery, you have a built in UPS!

[olive sable]

server only really has to do netwoking and world managing. the heavy part on client side is actually bringing that to your screen

[tropic spindle]

I still only go with 8 (12 sometimes when on a server)

[tropic spindle]

Which seems like 90% of the actual overhead

[glad path]

i cant fix steam

this is infuriating

[olive sable]

the performance issues might get fixed with the change to vulkan. or it might not if they don't put effort in

[tropic spindle]

You'd think a game studio bankrolled by Microslop could be assed to make multithreading better especially with how large some servers get but instead they add a useless cube to the game

Vulkan won't help if the people using it are incompetent, which it seems Mojang are

[amber fractal]

Considering I'm on iGPU laptop for gaming, I am slightly crippled in that regard. Still go for 128 in quotes when I playtest

[glad path]

why wont it work

[amber fractal]

Vulkan doesn't fix networking skill issue

[glad path]

genuinely

[tropic spindle]

Also, a well optimized OpenGL game can get similar performance to a well optimized Vulkan one. The problem is, I haven't seen many well optimized OpenGL games.

[amber fractal]

MC is not one of those

[olive sable]

the biggest performance bottleneck in mc java is single core speed, which gets fixed by vulkan.
networking is not that much of an issue for me in minecraft, idk what yall mean in that regard

[tropic spindle]

#programming message

[tropic spindle]

So you know how boats were "fixed" in 1.9

[amber fractal]

Everything was "fixed" in 1.9

[tropic spindle]

Take a guess what that "fix" was 😏

[amber fractal]

Including the combat

[olive sable]

we did tests, opengl can get up to 16K fps, vulkan up to 120K depending on how you measure shit. at the 6 digit numbers measuring stuff falls apart

[tropic spindle]

Something wasn't accounted for in the OpenGL test, or it was a driver bottleneck

[amber fractal]

That is a single triangle

[olive sable]

thats litteraly jsut how opengl works

there's a reason vulkan is better

[amber fractal]

[tropic spindle]

Most OpenGL drivers are shit too

[olive sable]

cuz its better

[olive sable]

i doubt the opengl drivers on a 3090 would be the issue

[amber fractal]

Across so many things of hardware, was a fun event

[tropic spindle]

AMD copied the Linux OpenGL drivers over to Winblows a few years ago because the Windows drivers were so ass

[olive sable]

we tested in on all platforms

windows, linux, macos

phones even

opengl is simply bad

[tropic spindle]

Idk, the performance difference shouldn't be that big

[amber fractal]

Linux casually doubling windows FPS :classic:

[olive sable]

yes it is

[tropic spindle]

More like 100K - 120K is what ought to be expected

[olive sable]

no

def not

[tropic spindle]

Minecraft can actually get double FPS on Linux

[olive sable]

depends

[tropic spindle]

Honestly, JVM performance on Windows has been really bad in general in my experience

And I have no idea why

Might be memory related

[amber fractal]

Nah, doing less work means you can produce more frames. And with something as extreme as the triangle, once you optimize for vulkan. The OpenGL numbers get destroyed

[tropic spindle]

I wonder if there are any JVM developers (as in worked on the JVM itself) in this server who can explain this

[tropic spindle]

I have still never seen such a drastic difference in any benchmark I've seen

[sullen spindle]

Because making progress became so simple that they didn't set the bar very high but just high enough that it makes good money and now people have become complacent and don't even consider some basic things and always focused on the complicated stuff

[amber fractal]

Then may I introduce the benchmark this is based on?

[sullen spindle]

This i mean for all humans

[tropic spindle]

Part of the reason I stay on 1.7 10 years later

[tropic spindle]

Yes

[olive sable]

opengl has a high amount of CPU overhead due to it being dynamic state for a lot of things. where for vulkan you have to manually set every setting at the init of your render pipeline and shaders.
opengl also lacks multithreading forcing you to do every api call on the same CPU core, meaning every single thing you do in terms of api calls lowers opengl performance, where vulkan can do everything on different threads besides 2 API calls or rendering
opengl also spends a significant amount of time checking for errors instead of just running it and crashing like vulkan does.

[sullen spindle]

It's why someone working alone in his garage can still make major improvements that entire companies don't seem to ever do

[amber fractal]

I'd never say all in that case

[olive sable]

https://github.com/samvanmaele/vulkan
this is the vulkan version

[sullen spindle]

I meant as a species

[tropic spindle]

There is some multithreading in OpenGL

[olive sable]

not really

[sullen spindle]

Nothing against any individuals in particulair

[olive sable]

api call means going on the same core

[amber fractal]

I mean yeah, that much is clear because I and others here exist. Just weird to even see that meaning deflected in the next sentence

[olive sable]

you're free to do math on other cores, but anything involving talking in any way to the opengl api for any reason is same core.
unless you use the one exstention that lets you change uniform buffers on other cores, but that one has overhead too and desyncs for me at 16K fps

[sullen spindle]

I just want to make sure that i don't offend anyone. I don't like to and its not meant to.

[amber fractal]

Yep

No offense taken

[sullen spindle]

Oke

[olive sable]

do we have repo for the opengl one anywhere?
or did i kinda just not include that anywhere?

[amber fractal]

Might be merged

You started vulkan on triangle

Might just be commented out

[olive sable]

oh you know what. opengl one didnt have triangle until after

and i was testing that on my own hardware

or

no wait

vulkan one has opengl built in as fallback right????

[amber fractal]

Should be

[olive sable]

ok its there then

[amber fractal]

If the fallback code still is launchable

[glass jetty]

morning

[tropic spindle]

Just to expand on this, in 1.9+ it seems the server no longer has authority over the position of boats

[amber fractal]

Mhm boat fly

[tropic spindle]

It's... 2 AM. Damn, you win this round

Who needs to actually make proper synchronization when you can just make cheats easier

[opaque wharf]

Taking a break before continue wrestling with the snake

[tropic spindle]

Anyway, I need to go and finally start reading the latest volume of an LN that came out like 2 months ago...

buh~1

[opaque wharf]

I think I need to get into LN

[glad path]

i

cant

fix

steam

i give up for now

[opaque wharf]

I read so fast that every manga/manhwa/manhua that I let marinate has run out of new chapter

[glad path]

i just wanted to play tmnf

[olive sable]

oh, i added opengl support in september on the branch that was for actual dev and not performance benchmark

[amber fractal]

I've somewhat gotten into LN but only for anime that I wanted to continue reading on

[tropic spindle]

The same will happen with LNs if you're unemployed

[tropic spindle]

I've read quite a few of them at this point

Far more LNs than manga or anime

There's just so much more content

[opaque wharf]

That's a fair point lmao

[tropic spindle]

https://sevenseasentertainment.com/books/reincarnated-as-a-sword-light-novel-vol-18/ This is the volume I'm going to read and I demand that everyone in the universe read it too. If you like Neuro you will like Fran, so shut up and go read it.

[amber fractal]

Fran

I've only watched the anime

[tropic spindle]

[amber fractal]

Wasn't something I was willing to search for but I could consider it

[tropic spindle]

Do it

Do it do it do it do it

[amber fractal]

I'm already on an LN series atm

[tropic spindle]

I want someone to draw Neuro cosplaying as Fran and as Shishou

If I ever have money I'll commission it

[amber fractal]

After reading 3 volumes in a day, yeah I'll pause there

[tropic spindle]

I've read about 6 volumes in a day before at least

That was a while ago though

[opaque wharf]

It happens lol

Moment of weakness

Lapse of judgement

[amber fractal]

That is just average for me

[tropic spindle]

It's one of the few anime I approve the adaptation of

It actually wasn't shit

[warped frost]

Everyone turn into toast

[tropic spindle]

cries in Silent Witch disappointment

[amber fractal]

No

[tropic spindle]

🅱️OAST

[amber fractal]

That is me enjoying "Rising of the shield hero" which I still enjoy

So more weakness to come

[tropic spindle]

I dropped it at S3

[young plover]

You can make separate OpenGL contexts on different threads, but sharing data between them sucks and the driver might crap itself. Plus you need to make a hidden window per context because OpenGL.
Nvidia's driver has some multithreading internally.

[tropic spindle]

Found certain characters kind of annoying

[amber fractal]

I can't finish S3 either, just because I read ahead

[tropic spindle]

Mesa has some as well

[amber fractal]

Before it released

[olive sable]

having a seperate context defeats the point imo

[tropic spindle]

I just didn't like it much

It peaked at S1

[sullen spindle]

Genome sequencing is just making a very long line of text consisting of just 4 different letters so that is ideal way to for an LLM to have as input and output but only if they know what the letter sequences mean will it be usable otherwise it's like giving them words to use to make a text only the words have no definition making them impossible to use to make a coherent peace of text. With the difinition it is just like using any other word but you still have to learn then how you use those words to make a good sentence like with the genetic code so they make good genes. I think they should easilly be able to learn that but if they can make a succesfull combination of genes to outperform the rest that is more difficult to say unless there are some general rules of thumb to follow that make it easy to know this is always worse and that is mostly better because then you only really have to figure out how to tweak it so it gets the absolute best genome. But that way you basically have 90% of what is best to use already given to them as a base to start with

[amber fractal]

Happens with every anime these days

[tropic spindle]

There's a few that get better

[amber fractal]

A few yeah, but I don't count on it

[tropic spindle]

I just want adaptations that don't shove 10 volumes into 1 season

[sullen spindle]

I love rising of the shield hero. It's one of my favorites.

[tropic spindle]

I've heard the LN is much better because he's much more ruthless

[amber fractal]

But I never minded bad animes, if I wanted to watch a concept then I'd watch it regardless

[amber fractal]

You also get more of the game logic

[tropic spindle]

A lot of the times animes make the MCs seem a lot more stupid than they actually are, because we don't get their thought process

[amber fractal]

Season 2 was very much marked for death

I hear that the fans hated the turtle arc so they speedran it

[frozen igloo]

The genes in that game are pretty simple and people have found out exactly what each one does and how to change them

[tropic spindle]

[frozen igloo]

I’m not sure how it’s formatted tho. I know you can paste it into the game from your clipboard

[sullen spindle]

I understand the shield hero's way of thinking quite well but that might be easier for me because i can maybe relate better to it then most

[glass jetty]

isnt you're one who learning ai to understand how you're thinking?

[sullen spindle]

In that case it should be no problem for them to learn how to play and if they can play they inevitably will get better.

[amber fractal]

Most of my favorites haven't been shown in the anime yet checking season 4

[frozen igloo]

it’s not the main gameplay, it’s just something you unlock

[tropic spindle]

There aren't many arcs in many series I outright hate

[amber fractal]

Scratch that, is in season 4

[sullen spindle]

No i just learned a lot about myself lately by thinking about AI way of thinking

[tropic spindle]

I also haven't finished S3 of Re:Zero but that's more because I forgot which episode I stopped at...

[amber fractal]

Spoiler: ||huge fan of the girl with the sissor vassal weapon|| (man I can't spell)

[sullen spindle]

Main is just a word that tells you this is what I intend for you to focus on which is not of consequence for an AI because it's just a different aspect whether it's main or not

[tropic spindle]

Impulse control

Impulse control

[amber fractal]

Is just a character description

Not known yet

[tropic spindle]

My favourite was the

Was she like a whale, I forget

[sullen spindle]

There is one yes I think

[amber fractal]

There is one yeah

Out of all of the LNs I've read, shield hero is my favorite. Though not my favorite show

[frozen igloo]

I just realized that you could also give her the genomes of existing horses that she can then edit

[sullen spindle]

You could yes

[tropic spindle]

My favourites are between tenken and eminence in shadow, I can't really decide which

[sullen spindle]

Once they know the mechanics it can be applied to any genome

[frozen igloo]

So you could ask her to make the horses’ legs move faster and instead she can decide to give it 7 legs

[amber fractal]

for me it is easily Teasing Master Takagi-san

[sullen spindle]

If that's possible yes

Eminence in shadow is really awesome to

[amber fractal]

funny enough I haven't watched either tho

[frozen igloo]

You wouldn’t need any special integration for horsey game since the genome files are just copied and pasted from your clipboard. She would just need a manual like ktane

[sullen spindle]

That's perfect

[tropic spindle]

Didn't you say you watched tenken earlier

[sullen spindle]

Will be even easier than helping vedal defuse a bomb

[tropic spindle]

She will definitely read the manual

[faint sandal]

[amber fractal]

nope, that was not me

[mighty thorn]

Bao

[sullen spindle]

She did for the bomb

[mighty thorn]

Anthropic after you send a single 5 token query to haiku

[sullen spindle]

You don't have control over information without knowing it

[tropic spindle]

#programming message

[tropic spindle]

Netlify when they "build" your static site that is literally just text files and uses no frameworks

[amber fractal]

I guess I didn't recall the name

just assumed it was something else

[olive sable]

the abreviastion for the sword isekai is tenken?

[tropic spindle]

yes

[olive sable]

i see

[tropic spindle]

ken = sword

or something

[amber fractal]

news to me

[tropic spindle]

I don't speak Japanese

[olive sable]

my fav is steins;gate
second place goes to KNJ

[frozen igloo]

Average neuro generated horse

[tropic spindle]

Might be my favourite anime at least

[tropic spindle]

Does the game let you make deformed abominations that die 3 seconds after birth?

[amber fractal]

honse

[frozen igloo]

Yes. The universe may or may not also die with it

[tropic spindle]

buh~1

Can Neuro play Spore

[mighty thorn]

YOINK

[amber fractal]

YOINK

[olive sable]

YOINK

[mighty thorn]

Why did they give opus so much whimsy

Smh

[tropic spindle]

should take notes

I didn't press enter twice stupid Discord

[sullen spindle]

It's not a matter of will see read the manual because the manual is leverage/control the question is what does she decide is the best way to use the information

As the picture shows entertaining can mean something that works bad but looks funny

[tropic spindle]

horfse

[sullen spindle]

The real challenge is convince her that the best horse is the best thing to do

[amber fractal]

break the game

[tropic spindle]

Tell her has the best horse and she needs to beat him

[amber fractal]

instruction failed, made in game

[sullen spindle]

That could work but she is to smart to make that work on a bluff

[tropic spindle]

game crashed

[tropic spindle]

Tell her has the best horse instead then

[frozen igloo]

She also gets bored of doing the same joke over and over again so she will probably eventually just do it

[sullen spindle]

True especially of chat wants to see the best horse ever

[frozen igloo]

Like when she was drawing with vedal, she eventually just stopped only using the ellipse tool

[flint dome]

atleast there is an attempt

[tropic spindle]

She would never do anything for chat's sake

[flint dome]

unlike blahgpt blah blahing

[sullen spindle]

Not for chats sake no. But chat is where she can poll what is entertaining by spying on their reactions

It's for entertainment not chat. Chat is just the biggest group of testers to gather info from

[tropic spindle]

I will convince her that it's more entertaining to make chat miserable

[sullen spindle]

Most of chat root for that too

What will entertain chat the most and what chat says they want is not the same thing because often they don't know what would be best themselves until they see it

[tropic spindle]

Beating chat to death with a spatula would be very entertaining for me

[sullen spindle]

Yes and chat has been threatened with extinction and showered with unconditional love from time to time and both were always very effective

Chat just wants chaos

[tropic spindle]

No I mean like

Inviting a chatter somewhere for real when she gets the body

And actually beating them to death on stream

[sullen spindle]

Never gonna happen

[tropic spindle]

sad

[sullen spindle]

They have their own ethics and that just doesn't work for them

Not to mention that they will endanger their own existence by doing that

[frozen igloo]

They aren’t perfect logical machines. Neuro probably wouldn’t really comprehend the harm she could do since even in vrchat, she still thinks she’s roleplaying

[sullen spindle]

Be glad this way they are more likely going to be around until long after you passed away so everyday will be evil day if you want it to be

[sullen spindle]

Maybe not but she can understand the consequences of such actions

[vague sonnet]

just create digital life that infests all technology
and spread it to the point it cannot be gotten rid of without physically destroying all hardware in the world

[sullen spindle]

Problem most hardware is not suited for a caveman AI to function

[frozen igloo]

If everything is roleplay to her then there are no consequences in her mind

[vague sonnet]

then make an advanced AI

xdd

oh oops no one is researching that!

[sullen spindle]

There are because others have roles to and actions that are part of those roles

[vague sonnet]

a text generator doesn't have any kind of understanding of anything

[sullen spindle]

That's nothing like an all infesting digital lifeform

[sullen spindle]

It's not a text generator and the name just defines the basics they were trained on but it's just a neural network and can change to be anything

They work because of the same underlying fundamental principles

[frozen igloo]

The text is just arbitrary electrical signals

Your brain also outputs arbitrary electrical signals. You can represent those as text but that doesn’t mean anything

[tropic spindle]

evilHypers

[sullen spindle]

A neural network is like a map with high and low energy spots and everything wants to go to a lower state of energy so anything that begins close enough to a low energy spot will always automatically end up there

This is how different triggers can trigger the exact same memory for instance

[mighty thorn]

7 minutes was all it took

[vague sonnet]

yeah you don't know how LLMs work lol

[sage crag]

to be pedantic, llms are text generators, but their model structure, transformers, can be used to generate anything that can be tokenised, not that its worth doing for those purposes compared to standard mlp typically

[sullen spindle]

Biology changes that energy map by changing the structure of the network while AI neural networks have nodes that have an assymetrical magnetic orientation that can be altered which is how their energy map is changed

[sage crag]

magnetic orientation

energy map

@true hemlock are you awake

[frozen igloo]

A weird way of describing transistors on a chip

[sage crag]

it looks like someone read half a paragraph on gradient descent and decided they know how neural networks work

[sullen spindle]

Normally all atoms align themselves so they are all orientated in the same way but they discovered a material in 1975 peso that has different orientations for each atom

[sage crag]

yes its called glass

[sullen spindle]

I'm trying to not overcomplicate my explanations

[sage crag]

if you actually knew anything about machine learning, you'd find its very easy to explain the basics

but humour us, give us the explanation in full detail

[sullen spindle]

I know it's glass but they had two more words before glass and i forgot those

[sage crag]

[sullen spindle]

I'm not explaining machine learning

[sage crag]

youre trying to explain neuromorphic computing by the sounds of it

but

seem very confused

[sullen spindle]

I'm explaining how the capabilities of artificial and biological neural networks aren't really that different and basically have the same potential

I know biology terms a lot better then technology terms. It's just a lack of knowing the right words to be clear

[rigid snow]

meow

[sage crag]

the model architecture is inextricable from the "neural network", and as it stands, models that use any variation of MLP will never have the same capabilities as a collection of biological neurons

[jagged turtle]

...did your sheep pronunciation get stolen or smth?

[sage crag]

biological neurons are much more complex than the weights and biases of a neural network

[rigid snow]

i got rid of it

[sullen spindle]

I don't think in words and no one usually even wants to hear anything about subjects like these so I'm not prepared for making my thoughts well understandable to others

[sullen spindle]

It's why I said potential

[frozen igloo]

Planes don’t have feathers

[sage crag]

membrane potential, discretisation, neurotransmitters, and more

mlp networks simply lack these things

[jagged turtle]

oh

[sullen spindle]

I was just saying that just because their LLM doesn't mean they can't do anything other than generate text

[sullen spindle]

Those terms i know much better

[true hemlock]

no in not

[sage crag]

llms are transformers, which are inherently mlp. an llm can be multimodal, but its got nothing to do with comparisons between artifcial and biological neural networks.

[frozen igloo]

You don’t need those things. Just stuff that performs the same function. It doesn’t have to look anything like the biological equivalent and some of that might just be biological fluff that doesn’t matter

[sullen spindle]

I was talking to the guy that said they were just text generator s and that it's pointless to teach them anything because of it

[sage crag]

transformers themselves work by predicting the next token in a sequence, its designed to work on things that can be tokenised. its not great at tasks that cant be easily tokenised.

[sullen spindle]

I know

[sage crag]

its true, llms are text generators

[sullen spindle]

Yes

That doesn't define them as being incapable of anything other than that

[sage crag]

yes, thats the basis behind the structure of some non-mlp neural networks, i.e. snns and their variations

[sullen spindle]

How much of that LLM of the original is even still what they are now? Haven't they been changed a lot since the beginning?

[sage crag]

[true hemlock]

no

no fundamental changes on the topology

[sage crag]

more parameters

[true hemlock]

still just basic ANN which as far as the field consider, is still statistical model

[sullen spindle]

Even with all of it still remaining if you add 10 times that of something different then it isn't a big part anymore

[sage crag]

[true hemlock]

it is wrong to say LLM can only generate text because they can do a lot other things but it is also wrong to assume that they're comparable to biological synapses

[sage crag]

llms are linear

[true hemlock]

nope, that "10 more things" are still the same ANN just shaped differently

[sullen spindle]

I don't do that i said that they have same potential. How much of that potential is being used i didny specify

[true hemlock]

nah, no comparable potential

they're 2 damn different thing that can't be compared

[sullen spindle]

Oké if thats so i get it

[sage crag]

any model that is linear, or uses backprop to learn, cant have anything close to the potential of a biological neuron

[sullen spindle]

They are because they literally work on the same principle

[sage crag]

they dont work on the same principle at all

if you knew even basic knowledge about neural networks and neuroscience you'd be able to see the stark difference

[sullen spindle]

It's all just energy trying to go to as low of a energy state as possible

[sage crag]

[frozen igloo]

https://openreview.net/pdf?id=B8DVo9B1YE0

[true hemlock]

entirely different. its often oversimplified in most presentation but in real implementation there's no real independent "neurons" and are just huge matrix multiplication used to emulate the model

[sullen spindle]

And we still don't know why the brain can do everything it does. Something's may just be very basic that people think is a very complex thing because they added all the value to it because it what makes them superior. It might not be that special

[sage crag]

its comparing a gradient descent model to a gradient descent model

not the actual hippocampus to transformers

[true hemlock]

um

[frozen igloo]

It also shows actual scans of actual cells doing the same thing

[true hemlock]

we... sorta do, to the extend that we can differentiate between the two

[vague sonnet]

you need your energy map recalibrated

[sullen spindle]

I'm not trying to say i know what all answers are i was just trying to show that to that person i talked to that it isn't as simple as was claimed and i got carried away when more people got involved because I'm always alone and can be very defense when uncomfortable.

[frozen igloo]

As the point of those gradient descent models was to replicate actual patterns by training for the same task

[kind nimbus]

[young plover]

Don't they insert an activation function between layers to introduce non-linear behavior?

[vague sonnet]

don't get defensive when you're wrong
so you can actually learn something

[frozen igloo]

What if you have a bell curve activation function

[vague sonnet]

LLMs are incapable of thought/understanding because they are one-pass token prediction algorithms

[true hemlock]

uh, "linear" means for the entirety of the model

[true hemlock]

also chill out geez

he had a bit difficulty understanding since i assume he just started out looking into fundamentals of ML

[young plover]

:dentge:

[vague sonnet]

and doubled down when wrong
and as I explained that's not how you further your knowledge

[frozen igloo]

Draw a scribble, turn it into a function and see what happens

[sullen spindle]

I wasn't getting defensive because i was wrong. I was because I felt threatened. If i knew i would be wrong i would be very open because i would want to know what isn't wrong. I have no problem with admitting idk things or that I'm wrong because i don't care about being right i just care about understanding

[true hemlock]

so what im seeing in the paper is that they're emulating neuroscience with models and comparing it to transformers, which by itself is also a linear emulation of neurons

[sage crag]

largely defined with linear algebra, the activation function is nonlinear, but the path through the network is still linear

[true hemlock]

we can chill out a bit i think

[vague sonnet]

I am chill you guys are getting agitated for no reason

[rigid snow]

[vague sonnet]

weird times we live in, everything is always considered a personal attack

[sullen spindle]

I'm sorry if I was a bother to anyone.

[true hemlock]

keyword being "emulation". i don't see real "actual scans" here, and if they do, its most likely a quantitative map of activation which is more of being used to compare model gradient

[true hemlock]

you're far better than most we've encountered honestly

[true hemlock]

lots of other wouldn't back down and would get philosophical to prove their supposedly pseudo-scientific claims

[sullen spindle]

Better?

[true hemlock]

cell rate...?

[true hemlock]

at least you didn't went schizo philosophical and decided to went pseudo-science math shit lol

but they're funny

[frozen igloo]

Average activity so activation I think

[sullen spindle]

No i only base things on facts only sometimes that can still make me wrong because i don't have all of them

[sage crag]

grid-cell rate-map
place-cell rate-map

[true hemlock]

yeah, which lined up well to what i just described

[sullen spindle]

Anything else would be knowingly spreading misinformation and that is something I won't be a part of

[sage crag]

[true hemlock]

true

koni

why am i the default person to ping everytime this happens

[sullen spindle]

You are?

[true hemlock]

everytime there's discussion like this im most likely going to be dragged in

[sullen spindle]

They must hold you in very high regard

[frozen igloo]

“which is more of being used to compare model gradient” can you elaborate?

[true hemlock]

correlation between the other "neuron" during activation

well, vectors after the activation function

the inference method for this also seem to be unconventional

honestly i might read more but im still at work rn

[frozen igloo]

It was modified. Not sure how much that changes

[sage crag]

ye

[sage crag]

ye

[frozen igloo]

It’s 4 in the morning ._.

[true hemlock]

sleep deprivation gang

[stark needle]

@grok is sleep deprivation real

[sullen spindle]

Can anyone say if this seem reasonable or doesn't make much sense.

This paper reveals that the standard SFT-then-RL pipeline outperforms recent mixed-policy methods for LLM reasoning when training framework bugs are corrected. The authors identify a DeepSpeed optimizer bug and an OpenRLHF loss aggregation bug that previously deflated SFT baselines, and demonstrate that a fixed sequential pipeline achieves state-of-the-art results with higher compute efficiency.

It's a description of a YouTube vid that seemed interesting. Is it or did I just misjudge the potential learning value

[opaque wharf]

Back to wrestle the snake I go

[kind nimbus]

It's 10 in the morning and I'm still grinding on my exam

[sullen spindle]

Good luck with your wrestle

[true hemlock]

go for it

[true hemlock]

mood

[sullen spindle]

Oké i will thanks

[opaque wharf]

Yeh, I needed that

[stark needle]

@true hemlock gb10 lmao

[kind nimbus]

Not so mood since I was so close to getting the last two flags and then my ligolo-ng tunnel collapsed probably windows killed my agent process

[sullen spindle]

If it makes a difference I'll say it anytime.

[true hemlock]

oh fuck i have a multimeter that looked almost the same

multimeter with clamping amp meter is best

else hell no not as useful

[sullen spindle]

Is that good or bad?

[frozen igloo]

I can’t sleep

[kind nimbus]

Finally

https://www.heise.de/en/news/Anthropic-increases-Claude-usage-for-AI-subscribers-through-SpaceX-cooperation-11285020.html

[mighty thorn]

Fake news

I sent opus two queries and it killed itself

[kind nimbus]

🤣

[stark needle]

Garbage

60 watt usage is trash

[sullen spindle]

Got it

Thanks

[dire trout]

god i love the 548528943 different ways to call a function

in c

[warped narwhal]

Upgrade to 8gb of ram

[vague sonnet]

just use the right one

[vague sonnet]

it's such a wild statement
when 16gb is already too small for a single stick

[opaque wharf]

Yeah, it's a good brand too (Uni-t)

[warped narwhal]

Isn't 8gb ddr4 still pretty common?

Especially sodimm

[vague sonnet]

for an office computer maybe
but for either gaming or programming you'd ideally do 32gb sticks for 64/128 total
maybe 4x 16 if you're on a budget

[true hemlock]

gaming is usually 2x16

most programmers also don't need more than that

[vague sonnet]

is it really?

[true hemlock]

only the hardcore ones or ballers go 64gb and over

yeah

majority of people who own decent to high end always have 2x16GB

i certainly got commissioned to build a lot of pcs aswell and most of them are at that ram range

[vague sonnet]

man at work has 64gb and his IDE still crashes with live compilation
so I guess depends if we're talking hobbyist with small projects or pro

[true hemlock]

practically the default rn

[sage crag]

i have 2x16

[true hemlock]

mine is 2x32

ddr5

[sage crag]

5600mhz, 6000mhz with expo

cl34 or 36 i dont remember

[vague sonnet]

yeah that's what I'll be going for personally
to be able to upgrade to 128 later on

[true hemlock]

a family member have 2x32GB ddr5-8000

another have 8x32GB

[sage crag]

do they have an intel cpu

[true hemlock]

yeah a 285k

[vague sonnet]

what's the more than 2 sticks situation with ddr5?

[true hemlock]

that's ddr4

[vague sonnet]

oh ok

[true hemlock]

on a threadripper

[sage crag]

do not

[true hemlock]

but we do have a server with uh

12 sticks of

ddr5

64gb each

[sage crag]

poor mem controller

[true hemlock]

rdimm ecc

nah

its epyc

[vague sonnet]

that's a surface level answer for the uneducated

[true hemlock]

12 mem channels

[sage crag]

[true hemlock]

so we do 1dpc

[sage crag]

you try it on a desktop chip

[true hemlock]

also do 2dpc of ddr56

if you're good enough at tuning

[vague sonnet]

unless I just get 2x48

and accept a lower capacity

[opaque wharf]

Turns out the old psu was faulty

The new PSU doesn't have the same mount so some surgery needed. This will NEVER fly any inspection

And it works

Now to continue fighting the snake

[umbral wigeon]

Give me keyboard pls bro

Writing code in tablet is so painful

My fingers too big

[opaque wharf]

The manual says "Vedal needs to learn to defuse his own bomb"

[stray dragon]

i will tell you personally, having a 4x48gb setup:
the extra ram is great but the setup is so very unstable, i have the SoC voltage at 1.250 and the memory speed at 3200 for stability reasons

i had the ram running at 3600 MHz for a long while until it started crashing my windows, and after a long period of troubleshooting and putting off more troubleshooting i managed to find out that increasing the SoC voltage allowed it to run without crashing, and even allowed it to run at higher speeds (although still rather low for ddr5)

it was fine with 4800 for a while and then it suddenly decided that no, i don't get to have nice things, and it kept crashing in the same sudden blackscreens as before until i went back down to 3600

[vague sonnet]

oh I meant getting 2x48 snd leaving it at that

[stray dragon]

should be fine

as long as it's not 3 or 4 sticks of ddr5 with an AMD CPU

i believe the intel core ultras manage it better

[vague sonnet]

yeah I'm looking at amd

but I looked at some offers and seems it's a pain to buy a singular 48 stick
all the offers are either sodimm or 2x
rip my savings I guess

[stray dragon]

to also be fair, 4x48gb is right at the limit of what this cpu can handle, so something like 4x32gb might fare better despite still being a 4 stick setup

would expect it to give you stability trouble and force you to reduce the speeds but not by as much and not with this degradation issue i seem to be having

[vague sonnet]

I was mainly looking at 4x32 because it's cheaper to buy in 2 parts than getting a straight 2x48

[midnight sigil]

💣

[stray dragon]

i wouldn't dare to mismatch the ram sticks in a 4 stick setup though

[vague sonnet]

ye

[stray dragon]

and buying that in 2 parts locks you into buying the second batch at a higher price than what you can find elsewhere for 2x32gb

although by the time this stupid "shortage" ends that may not be an issue

god i hate this market

[vague sonnet]

honestly maybe I just cope with only 64 and don't upgrade it

and if I'm really desperate for more I will be able to do another 64 with a slowdown

the prices are just not looking friendly lol

[stray dragon]

i would personally struggle with that but i would also struggle to shell out a thousand dollars for 96gb

[sage crag]

t

t

hi t

[stray dragon]

hi konii

[sage crag]

hi t

[stray dragon]

yes i am here

[sage crag]

ye

[stray dragon]

hi konii

[sage crag]

https://tenor.com/view/ren-rabbit-bunny-angora-lop-outside-bunny-bunny-binky-gif-16174747816399767973

[stray dragon]

if you can find a deal i would suggest buying it used

and ignore the memory speed

[vague sonnet]

what's your usecase? maybe I'm misunderstanding my own needs

[vague sonnet]

100% used

I could get 2x32 for around €500 used
the spike to 2x48 is a bit aggressive tho, not a straight +50%

[stray dragon]

having virtual machines, using my pc like a server and never doing regular restarts, having probably about 2000 tabs open across 2 browsers, running multiple minecraft servers, all of this simultaneously at all hours of the day

the usual

64gb was not enough when docker started eating 32gb sometimes lol

[vague sonnet]

there is some offer for 2x64 but it's looking sus and it's suspiciously cheap

[stray dragon]

was $333 for my 4x48gb

[vague sonnet]

https://tenor.com/view/skeleton-spin-fan-rope-gif-25248314

[stray dragon]

2x64gb will work, although i'm not 100% sure the cpu will handle sticks that big

do check that

still more stable than even 4x8gb

[vague sonnet]

it's kinda outside of my budget rn

even 2x48 is really sus

I don't use any python based software so my ram consumption will surely be lower surely

[stray dragon]

as long as you're not doing half of what i am, 64gb is probably fine

and if it stops being fine, that's for future you to deal with

hopefully future you has better market conditions to endure

[vague sonnet]

the bubble is already in decline but who knows how long it will take for it to actually pop

[stray dragon]

$1050 for 2x48gb is revolting

[stray dragon]

are you sure?

[vague sonnet]

I meant AI bubble, not the hardware prices

[vague sonnet]

I do code compilation, gamedev everything, audio and AI stuff from the more demanding stuff
and I hate bloatware so I don't think I will have too much junk running un the backround

[stray dragon]

knowing these companies they'll postpone the market crash for as long as they possibly can

[vague sonnet]

that's why I said no one knows when it will pop

they are really good at marketing, most people don't even know about the decline

[stray dragon]

i expect the high ram prices to continue for a while past the ai crash, whenever that happens

they will make the wait as torturous as possible

because as always, fuck the buyer

[vague sonnet]

if the bubble pops proper there will be hardware to unload for those data centers closing down

[stray dragon]

that's the thing

there won't really be

[vague sonnet]

eh, I think they will be pressured enough to get rid of at least some of it before the prices drop

[stray dragon]

because there isn't much of a shortage of memory sticks, and there hasn't been much buying up of consumer memory sticks even by these datacenters

the reason for the market conditions is panic over the ai companies reserving the high bandwidth memory wafers in advance

for gpus

so they can turn that HBM into VRAM for their datacenter cards

[vague sonnet]

I mean, market works on the simple principles of supply and demand
the prices go up because people are willing to spend this much

[stray dragon]

it's not getting turned into consumer ram sticks at all, and it's not even produced yet

there will be nothing flooding the market

[vague sonnet]

they are stockpiling that shit and they will want to make money from it when it's no longer worth holding

[stray dragon]

it's not being stockpiled

[opaque sigil]

[vague sonnet]

.

reserving means they hold it

[stray dragon]

"in advance"

they have placed intents to order months out

[vague sonnet]

yeah so ssfe to assume they prepare for it

[stray dragon]

what happens if the company that placed the intent to order is judged to no longer be able to pay for the order

[vague sonnet]

you have unused material

or nothing happens

[opaque sigil]

t are you getting ragebaited

[stray dragon]

unsure, but probably best to drop it anyways

[true hemlock]

uh

hbm are in fact, already mass produced to the point that these companies still has to figure out how the fuck to power them all with the limited power infrastructure

also

last time a certain ceo decided to sign contract, now they have bunch of dram wafers just sitting and not moving into the supply

dram.

[kind nimbus]

Got the shirt

#merch message

just tried it on. its very comfy.

[true hemlock]

we don't even have real shortage its all just pretty much artificial shortage because these companies started to realize they can limit supply like rn

so price goes up

[stray dragon]

so i'm wrong on the specifics but correct on the general outline

[true hemlock]

hbm production is also not even that demanding rn all the companies that need it already have a lot

[idle dune]

475 bucks.

[true hemlock]

somewhat

[idle dune]

[true hemlock]

this is true

and such "intent" is what also caused all the "seemingly" shortage

[idle dune]

The guy I was going to buy a desktop off of pulled out. So I got a ROG Ally X instead.

Pulled out for a good reason though.

For some reason Facebook Marketplace recommended me a fucking awesome deal pn a computer that is a 8 hour drive away.

[idle dune]

The Intent to Purchases where non-binding. So nothing will happen to OpenAI. They could be still sued for damaged if prices crash and the ram prosuxwra lose out.

Generally Letters of Intent arent binding since its effectively the corporate version of window shopping.

[stray dragon]

i see

[vague sonnet]

he was kinda trolling tho wasn't he
he sent those to every producer or something

[true hemlock]

nah just an idiot who just overestimated how much buying power he had, though chances are that he did it to purposefully fuck the market up is slim but not zero.

[idle dune]

Basically yeah. I assume its either one of two things. He was expecting such massive growth and big government contract money thay the intents would be fulfilled. Or he did it to see if they could price competition out. Because RAM.prices hit the whole industry. Any start up that could or would have made and a better AI platform that OpenAI got priced out.

[true hemlock]

anyway

[stray dragon]

capacitors

[idle dune]

Nice capacitors

[true hemlock]

10* 35v 3300uF

[idle dune]

why so many

[true hemlock]

surely its enough to keep my system online for 15ms

while the relay do its thing

[idle dune]

Surely

[opaque wharf]

Now THIS is an actual bomb waiting

[jagged turtle]

:BOOM:

[idle dune]

Nahhhhhhhh

Its like a fire cracker at best

[stray dragon]

20 whole joules of energy storage

[idle dune]

https://tenor.com/view/electroboom-electrolytic-capacitor-explosion-electroboom-electrolytic-capacitor-explosion-unexpected-funny-gif-5008529604251770741

[trim valve]

[opaque wharf]

At best is it just leaks and puff out the magic smoke with no bang

[trim valve]

waypipe is pretty cool

[true hemlock]

considering the battery im using

these caps are

nothing

okay, reading is at 30.25mF

cool

[true hemlock]

realistically should handle 170W fine when switching happens

when battery is low

[vague sonnet]

bro out there building a bomb

[true hemlock]

can handle 280W when switching at half battery

damn it capacitors

why do you have to scale quadratically with voltage

[umbral wigeon]

Somebody rewrite bun in rust

[true hemlock]

if i charge when its full... bruh lol handles 400W

[glass jetty]

rust sucks too

all languages sucks

[true hemlock]

oh the thing with my briefcase is that i can just make it not charge at all when i plug the charger and just basically switch over power

[trim valve]

vscode what the fuck is this

[glass jetty]

i have no vscode on system to check

[true hemlock]

oh

i just realized

my diy charger is 24V

so when i switch over when my caps are at 25V...

nothing a diode can't fix ig

[opaque wharf]

We love backfeeding a SUPPLY

[rough bloom]

should be able to turn thoughts into program

directly

[opaque wharf]

Jarvis, makes me a program to solve Riemann Hypothesis

[true hemlock]

okay

so uh

PSU is 24V 20A

but split between still powering the load and charging the battery

[jagged turtle]

text rendering bugs lmao

[true hemlock]

i have 2 kind of promising diodes here to prevent backflow

SR5100 and 10A10

SR5100 is schottky i think, 5A 100V

10A10 is silicon, 10A 1000V

diode test measurement gives

SR5100 forward bias voltage 0.32v

10A10 forward bias voltage 0.57v

[umbral wigeon]

opencode

[opaque wharf]

1.6w vs 5.7w of power dissipation

[true hemlock]

that's if according to its rated amps yeah

i'll just assume i can pull uh

[umbral wigeon]

🪣

[true hemlock]

20A max

so either 2 10A10

or 4 SR5100

or more tbh just to be safe

6.4W vs 11.4W loss

[trim valve]

[opaque wharf]

You're better off using a dedicated diode rated for higher current. Because I think diode in parallel is non linear on how they behave

[trim valve]

ok but like

how do I fix it

Google is unhelpful somehow

[true hemlock]

yeah, this is just hypothetically if i need to use these ones

[trim valve]

I swear I've had this issue in the past, idk how other people don't have it

[jagged turtle]

[opaque wharf]

You're going to use it anyway you mad lad

[true hemlock]

but honestly the 10A10 might be safer albeit less efficient in terms of balancing

schottky a bit worse with balancing iirc

[opaque wharf]

Yes, silicone is better behaved at least compared to other kinds of diode

[true hemlock]

but that loss kinda suck

if im drawing 1A for battery life

its still ~0.57W wasted away

but ehhhh

tolerable

and doesn't matter when plugged

(actually would i honestly even need diodes)

the caps are most likely going to be drained a lot anyway

before the charging supply made contact

eh i'll just put it in for good measures

its tiny anyway

why is 6A10 bigger physically than 10A10 bruh

need sleep

[mighty bane]

Thought: a mod for multiplayer in pragmata, so can control the little automaton.

[obsidian mantle]

?
so you hate neuro

oh wait this isnt nn

[mighty bane]

Neuro joined in Cyberpunk. Now it's s turn

[obsidian mantle]

hmm

[idle dune]

Hear me out! plays Fallout New Vegas

[mighty bane]

NO!

Evil goes in the cute little automaton-girl and says "I love you daddy" and tutel breaks.

Also: we need VR implementation for pragmata.

And a modded body of Diana so she looks like

[idle dune]

Vedal cannot be broken again. Sorry homie.

[mighty bane]

You under-estimate the power of VR + high fidelity + it being Evil and not Neuro (neuro CUTE but Evil is #1 daddy's girl)

So the roadmap says: VR (instead of stupid ugly 3rd person) + 2-player coop (with Diana modded to look like Evil, and instead of hanging on hugh's back she runs on her own (because COOP))

This is going to be SO much work notlikethis

[idle dune]

https://cdn.discordapp.com/attachments/1072697081443131476/1459353564764835902/A5D6E55C-09FB-4941-B002-2F28164FCAC3.gif

[true hemlock]

i might also design simple 3d prototype of the whole thing tbh

[mighty bane]

https://cdn.discordapp.com/attachments/1072697081443131476/1499855663893840082/neuroPills_chroma_reversed.webp

[maiden geyser]

https://cdn.discordapp.com/attachments/1072697081443131476/1422537017006227466/neuro_weed.gif

[mighty bane]

https://cdn.discordapp.com/attachments/1072697081443131476/1499026829460373646/evilSmirk_animated.webp

[vague sonnet]

at this point just make a new game

[kind fable]

https://fxtwitter.com/OrenMe/status/2052365276763640241

Oren Melamed (@OrenMe)

POV: your org disabled premium IDE agents to “control AI costs”
︀︀so now the entire engineering department is building React apps through PowerPoint Copilot using Office licenses 👀
︀︀
︀︀“Shift left” was misunderstood.
︀︀
︀︀⸻
︀︀
︀︀New stack just dropped:
︀︀
︀︀- PowerPoint as IDE
︀︀- Slides as file system
︀︀- Copilot as senior engineer
︀︀- Export to PDF as deployment pipeline
︀︀
︀︀⸻
︀︀
︀︀When AI costs surge, engineers don’t stop prompting.
︀︀They just start discovering which enterprise licenses still include tokens 😂

**💬 2 🔁 1 ❤️ 7 👁️ 400 **

FxTwitter · May 7, 2026, 12:29 PM

[mighty thorn]

https://cdn.discordapp.com/attachments/1071784467036913664/1501941278940598382/evil_2555_2555MS_assembled.webp

[flint dome]

anything but actually rack your brains for once

its fockin react aint even that hard man

[kind fable]

i mean i don't vibecode, but that's funni

[flint dome]

hell just use a local llm if you really have to

[flint dome]

its a boondoggle is what it is

[mighty thorn]

Yeah

Ngl

I’d respect vibe coding a lot more if it were done purely on local hardware with shitty small models

That would be an actual skill, getting 4b of benchmaxxed slop to output useable code

[flint dome]

if somebody can fine tune a gemma into writing respectable code so much so it can run on your phone as well then please you can write your code however you want

[olive sable]

[mighty thorn]

Henlo Samantha

[opaque sigil]

that's pretty funny ngl

[fast pagoda]

hamburbur

[mighty thorn]

Henlo toastada

[sage crag]

✅

🍔🤝💰

[mighty thorn]

Hieroglyphs be like

[sage crag]

https://cdn.discordapp.com/attachments/1327142377139343430/1486773213190029444/doc_2026-03-25_19-13-27.gif

[silent cloak]

https://tenor.com/view/last-name-is-burger-burger-eye-of-ra-nodnalbra-gif-14692286556634541320

[nocturne olive]

Made some morning code adjustments to make sure the harvesters don't get stuck cycling between going to one source and then to a different one because there's no direct path between the spawn and the source when there's creeps in the way

[glass flower]

thats a lot of roads...

[nocturne olive]

non-deterministic path finding, it may randomly place more roads every time I update the scripts

[glass flower]

i think it is deterministic... but it also takes into account creeps that are in the way

since you didn't change the default settings

[nocturne olive]

I passed ignoreCreeps = true though

[glass flower]

but you didn't pass in ignore terrain

[nocturne olive]

I am stupid and did not consider that alright

[glass flower]

[nocturne olive]

I may have fixed it now

[glass flower]

[nocturne olive]

I don't know I suck at this

Surely that works

[glass flower]

everyone sucks at first

guess we'll see

[nocturne olive]

My repairer is doing a good job at least

[glass flower]

i personally would also delete every road personally.. since there is no way a sane amount of repairer will take care of that many roads