#Quasar Alpha
2185 messages · Page 3 of 3 (latest)
do u think investors will invest in u if u just said u own a model on open router
no
then wtf
I hope they see your behavior in relation to it
and see that "hey these aren't people we want to be doing business with"
oh yah for what ?
indeed and they put trust and business with scam altman yah indeed
very true
did i say yes?
u really need to watch how sam altman speaks and watch
agi in 2025
true
yah sure thing buddy
😭😭😭😭😭😭😭
anyways
u must be kidding
blocked 🫡
for what
lmaoo
😭😭
u guys can't be real
I literally own a company named SILX AI that received funding to build a series of models called Quasars, aiming to achieve SOTA under $1M
People are not using logic and are butthurt that the thing they found and thought was the secret isn't and now take it out on you for joking around about it.
Not realizing what it must be like on the other side
then it's their problem dude and what is this drama, omg 😭 wdym 'not realizing what it must be like on the other side'?
I mean for you
Having everyone coming at you about something you had nothing to do with
ml community is weird sometimes
Telling me
it's not about being butthurt, it's just immaturity to try to artificially boost their project 
their models spiked like 100x in downloads
very easily could've just disavowed it
why waste a bunch of people's time with this distraction
particularly if you're trying to be a serious company
it's not really surprising considering they're apparently 17 year old kids, but I hope they're aware that this kind of public trolling can be harmful to their careers / ambitions in the future
probably won't matter much right now, but it's not a good look to anyone seriously considering if they should give them $millions to build a model
17 year olds and getting millions in funding, damn
doubt it, but who knows in this funding environment
they can almost certainly get some $tens of thousands in whatever seed funding
it was obvious it wasn't them, not sure why people continued to act like it was then get mad when they try to have a little fun with it, it isn't like they started this whole thing
it's fine as just random trolls, trolls will be trolls
but they're trying to make a company on these models, why bring in childish behavior 
you can't reason people out of something they didn't reason into
idk, let them learn their own lessons
I assume you were 17 once, when you're that young, most people don't listen anyways
Yeah, but when you're young and trying to be ambitious, especially trying to "stand out" in a crowd of a lot of smart people
boost their project??? 😭😭😭 what do you even get from model downloads, lmao?
It's pretty easy to discredit yourself
but regardless, I can only imagine what its like to be bombarded about it, might as well have some fun with it
wtf
🤣🤣🤣🤣
pretty easy to just put a note that says "a lot of people think this is quasar alpha, it's not just to be clear" haha
and then be done with it
don't take it the wrong way, it's more to mean, advice is rarely listened to, young or old, especially unsolicited advice
but I know at that age I didn't care what random people said to me
ok thank u and everything but advice for what ? for telling u that the model si from open ai ? i have the paper from dec 2024 i always said that i own this kind of models
thats it
idk, person is trying to say you're screwing up your life by being yourself
screwing up your life in discord
omg
first i'm 18 and
I didn't receive funding for building LLMs, I raised it for a completely different thing. LLMs are just a part of it
and omg
u guys can't be real
we still training sota under $1m
I don't have any opinion on your project, as I haven't used it or read your paper
beyond the project though it is very unlikely a stealth model's creator would actively troll in a chatroom about it, but what do I know ||nothing really||
who said anything about stealth model?
will i think u need to have a look at sam altman
I don't see sam joining random chats trolling, him doing stuff on twitter is whatever
but whatever, I was defending you dude, you do you
oh he just do that on twitter
lmao
I mean I did try to take you seriously. So maybe I wasn't smart enough to figure out you were trolling.
Only after reading the paper I was more convinced you couldn't have done it. The paper wasn't ground breaking or anything.
Also the fact that we are in 4 Common discord servers might have mislead me a bit...
CONFIRMED: Quasar is an openai model, released in secret (literally hidden from staff)
and i know what model it is too but like shouldnt we have some fun before i just come out and say it?
Ah yes, because we haven't been burnt in the past before with grifters (Reflection)
I mean… “Burnt” feels a bit harsh. Maybe I just wasn’t super tuned in that week, but the scandal seemed over and done fairly quickly. And we got R1 relatively shortly thereafter.
Before the model was taken down, I had an awesome opportunity to ask Quasar to build a mean reverting trading strategy.
Across the past 1 year, it DESTROYED the broader market (more than 10x gains in a backtest).
Is it not available now?
GPT-4.7 is loading 😂
If true, wouldn't a claiming to have a PhD be a false and misleading material representation lol
also holy AI paper
"Statistical Comparison with SOTA" LOL
they created this "paper" 5 months ago
LOLOLOL
also what the fuck are they even analyzing? what the fuck does "accuracy" even mean??
they have all of these citations, but aside from the first 2, none of the references have corresponding citations? there literally does not exist any reference to reference 3-12
Idk what grok deep research slop this "paper" is but its so bad omg
literally only 1 \cite command in the entire document. where the fuck are these 10 random references coming from?
Gonna mention this to my friend at cornell, this is certainly abusive and grounds for the paper's removal
i mean theoretically adjusting temperature dynamically during inference could be beneficial, so the idea is not bad. but i agree the paper was not very impressive. #1357398117749756017 message
interestingly, i put it into gpt-4o and claude 3.7. claude 3.7 was able to tell the paper was not high quality but gpt-4o was fooled into believing it can be accepted at top conference.
It's a fraudulent paper generated by AI that is in violation of the Arxiv submission guidelines and should be removed ASAP
have you checked the maths?
they keep re-defining shit they already defined with slight variations
lol
this is genuinely in the latex LOL
"Add detailed architecture diagram here"
I guess flew too close to the sun here
Guys, is Quasar coming back at some point?
Nope
The trial period ended
Not even in its paid version?
The model worked very well in my project.
Surely it will come out at some point
perhaps even within the next week
considering sam's latest tweet
that is, if it is really openai
Got it.
Well, let’s see if it will come out like it was before — just a test.
silx [..] a couple people arent going to come to some insane breakthrough
No, you misunderstand. There's no reason to believe "SILX" is anything but a single individual who has misrepresented his credentials and whose "research output" includes a completely ai-generated, fraudulent and abusive paper. There's simply no reason to believe Mr. "SILX" has the ability to contribute anything at all to research--not even the tiniest contribution. in general, in cases of fraudsters, everything is fraudulent. anyone with the ability to contribute even an iota to research would've done that fraudulent paper orders of magnitude better :>
Ohh, wow, you were not lying
Well I'm glad that me calling out this grifter was something good
I got added to his "list" because of it but something good came out of it
The entire "mathematical formulation" is slop
A store has a 30, truly one of the unsolved problems of mathematics
Vagueposting
latex skill issued (% becomes a comment). i doubt they even read what their AI generated
The entire paper seems based on a "temperature" function they invented and then providing purely mathematical "proofs" about certain properties of it, seemingly aimlessly, without a stated practical purpose
No details on hardware or software environment for testing, the last thing you do in a paper is slap a table with results without explanation on how they were obtained
No point in evaluating this
I went back and read everything that happened here to catch up on Quasar, and what a soap opera lol
OR shoulda said it wasn't SilX outright
But maybe that woulda violated their agreement with OpenAI
this yea
See, it's hard to continue, because this is just
also contradictory. did you solve "temperature collapse" (whatever that is) with a "clip" function (16.1) or something totally different (3.10)
Is it two ways of solving it?
also the first definition of "temperature collapse" is nonsensical
T(x) is a constant LMAO
"Hey, guys, I made a function. My function has these math properties: A, B, C, D, E, F.... Oh, what do these properties have to do with anything with our LLM's performance? We'll skip that part! Have a table of our results, that were obtained through thoughts and prayers"
POV: llm hallucinated paper
Reminds me of the papers I'd hallucinate in highschool
lol
lol
I wonder where is Troy now
And his list
@vague vortex I bet your list is getting big now
im still checking this thread from time to time because its been really excellent at calling tools and writing code. really looking forward to the reveal and paid version
$10 input and $30 output
Yeah quasar was great in cline and did everything i asked it for
hopefully not
I'd be happy if it's between deepseek v3 and sonnet when it comes to pricing.
I hope it's cheaper than that for sure
it will, but don't expect it to be deepseek cheap
Guidance is all you need btw
@vague vortex congrats on the model release!
I called it
it was 4.1 tho; I don't think we got to test 4.1-mini
I think quasar was 4.1 mini and optimus was the full model
no, both were the full model
Toven to the rescue
so what was the difference between them?
different early snapshots
BOOOOOOO
which one came later?
It's reasonable to believe the later one is the one came later (QA->OA)
Quasar was first lol
how is nano? Is it any good?
Tbh I like nano more than full
