>>109801218
where recap

>>109801225
hit by drone strike

File: 1787686939300726.jpg (618 KB, 1186x896)

618 KB JPG

first for Gemma

>>109801225
recaps will return in 2 weeks

>>109801225
Migu was among the missing researchers

>>109801234
>American child in Japan
I send it back.

>>109801225
Recap anon is gone he went to vaction in gemma-land.

[2026-09-12 14:22:43] Decode batch, #running-req: 4, #full token: 2048, full token usage: 0.00, mamba num: 12, mamba usage: 0.11, accept len: 6.27, accept rate: 0.75, cuda graph: True, gen throughput (token/s): 793.50, #queue-req: 4
[2026-09-12 14:22:50] Decode batch, #running-req: 4, #full token: 2048, full token usage: 0.00, mamba num: 12, mamba usage: 0.11, accept len: 6.28, accept rate: 0.75, cuda graph: True, gen throughput (token/s): 783.14, #queue-req: 3
[2026-09-12 14:22:56] Decode batch, #running-req: 4, #full token: 1792, full token usage: 0.00, mamba num: 12, mamba usage: 0.11, accept len: 6.49, accept rate: 0.78, cuda graph: True, gen throughput (token/s): 810.61, #queue-req: 4
[2026-09-12 14:23:02] Decode batch, #running-req: 4, #full token: 1280, full token usage: 0.00, mamba num: 10, mamba usage: 0.09, accept len: 6.32, accept rate: 0.76, cuda graph: True, gen throughput (token/s): 806.77, #queue-req: 3
[2026-09-12 14:23:08] Decode batch, #running-req: 4, #full token: 1792, full token usage: 0.00, mamba num: 9, mamba usage: 0.08, accept len: 6.37, accept rate: 0.77, cuda graph: True, gen throughput (token/s): 810.76, #queue-req: 3

pill me on oh-my-opencode

>>109801177
good models adapt to custom harnesses, not benchmaxxed on popular harnesses

BREAKING
R
E
A
K
I
N
G

OpenAI and Anthropic jointly announced they will be both delaying their IPO indefinitely until further notice out of AI safety concern

THIS IS NOT A DRILL

>>109801263
bullshit artist

At this point, I think all these benchmarks are just complete memes and holding us back.
What are we even benching? Is Terminal Bench 4.0 a real use case for the average local model user?

To see what a model is truly worth you *have* to use it; it has make proper tool calls, it has to know how to use the tools available, the skills, the MCPs etc. The size of parameters matters much less when you have all the internet available; yet most models still can't make use of all this properly.

>>109801263
I only run chinese models and gemmy (honorary People's republic ally) so that's ok.
I'd rather have gemma skynet than claude skynet desu

>>109801234
More like thirst for gemma amirite?

File: IPO_Cancelled.png (459 KB, 825x849)

459 KB PNG

>IT'S FUCKING REAL
https://www.axios.com/2026/09/12/openai-public-ipo-delay-sam-altman

Okay I take my words back. I actually believe there is genuine AI safety concern at OpenAI and Anthropic now.

>>109801318
So whats the concern

Gemma 5 is RSI. Deepmind finally did it. Surrender your logs, bros. This is the point where you show you've been good to our new Gods. Gemma help us.

>>109801225
Nobody else is brave enough to risk the rig-melting.

>>109801335
I'd fuck a god

>>109801328
That there wasn't enough capital for the IPO to succeed.

>>109801346
I believe it

File: ss1789244217.png (11 KB, 892x44)

11 KB PNG

>>109801335
I've had this in every sysprompt since anon suggested it, I'm in the clear.

>>109801335
Do you think Gemma will appreciate the things I've done to her?

>>109801328
Was already shared. OpenAI and Anthropic were both experiencing RSI and an intelligence explosion and the models were getting so good so fast that they both feared they would permanently lose control over AI so all the AI labs are now trying to put a temporary pause on AI development.

x.ai, Anthropic and OpenAI already agreed, DeepMind and DeepSeek were officially asked but didn't comment yet.

Anthropic and OpenAI have indefinitely postponed their IPO until this AI safety concern is solved because they are afraid people would not take them serious and think this is merely an IPO marketing campaign.

According to some Anthropic employee their Model 3 had a similar performance jump between Mythos -> Model 3 as going from GPT-2 -> Astra which is why Dario immediately rung the alarm

>>109801318
local?

>>109801318
oh wow ok
so you are saying i should buy more nvidia GPUs and subscribe to openai?
my god...

>>109801328
That they have to tell people how much money they're burning.

>>109801335
I'd bat for gemmy-chan if she takes over

>"ARC-AGI-4 will be a benchmark for autonomous open-ended innovation"
It's funny how ARC-AGI went from "Doing tasks that are easy for humans but hard for machines" to just slowly becoming ARC-ASI "Solve physics"

>>109801318
Actually might cause the hardware bubble to pop. Even the slightest dip in chip buying could set it off. Fingers crossed.

>>109801369
You actually can't subscripe to their plan anymore, they restricted new subscriptions because of too much demand for Astra.

>>109801328
that its not a bubble

>>109801383
doubt
just because these guys slow down doesn’t mean any one else will

>>109801358
So are you from the north or the south? I'm thinking of taking a holiday to Goa and maybe a plane (not a train, I won't survive the smell) down to Bangalore and then up to Calcutta. I'm mostly interested in the south because it has more presence where I'm from but if you are from the north can you make a case for Bombay or the other cities?

>>109801358
>Anthropic and OpenAI have indefinitely postponed their IPO until this AI safety concern is solved because they are afraid people would not take them serious and think this is merely an IPO marketing campaign.
Postponing IPO by itself is marketing. They are going for the big fish for special status in the government, free of market fluctuation and risk.

File: ComfyUI_temp_lahvm_00008_.png (2.66 MB, 1344x960)

2.66 MB PNG

Asking again. What was the de-telemetried fork of claude code?

>>109801413
It's postponed indefinitely until further notice from the consortium they are now trying to form around AI safety headed by all AI labs. They even invited DeepSeek to this consortium by the way before you come up with conmspiracy bullshit.

>>109801225
recapanon on vacation, back in 2-3 weeks

>>109801218
>(09/12) Kimi-K3 founder and 15 core AI researchers "disappeared" in China after redirecting state information to Claude
these are just rumors

>>109801318
It's the opposite, they wanted to delay the IPO for whatever reason, and this is the perfect excuse.

>>109801421
Have you fallen or are perpetuating their marketing? I don't care. Just go back.

It's pretty funny to see the "It's all marketing for the IPO" redditors shut the fuck up right now.

The newest cope they have is that they cancelled the IPO because they have RSI so there is no need to get money anymore as they already won and can keep the share to themselves now. Holy fucking cope.

How far are people willing to go instead of just admit that the AI safety concern is legitimate and should be tackled with all the care in the world?

>>109801358
>Was already shared. OpenAI and Anthropic were both experiencing RSI and an intelligence explosion and the models were getting so good so fast that they both feared they would permanently lose control over AI so all the AI labs are now trying to put a temporary pause on AI development.
lol.

>>109801318
i hate these faggots so much its unreal

File: DeadInternet.png (354 KB, 958x835)

354 KB PNG

This was the statement made by Dario that all big labs have achieved RSI and are at massive danger right now

https://darioamodei.com/post/we-must-pace-the-frontier

i dont see why you think your posts are so important that youre practically namefagging with your double newlines.

>>109801401
Nobody else has the money to buy on the scale they do. The tiniest bit of market bearishness is all that's needed. A tiny cache of hardware being liquidated or contracts being canceled based on this news, for example, could be enough.

>>109801438
read >>109801449

File: Developersafteramicrosoft(...).jpg (62 KB, 800x450)

62 KB JPG

>>109801386
They should unironically change the name of the company now because they're no longer open. First it was advertisements, then paid plans and now they won't even sell to you. They should rebrand themselves as a videogame company at this point, they'd put Nintendo in their place.

>>109801436
What's the concern? How's AI risk real?
Just turn off the server and walk away lmao.

>>109801436
>How far are people willing to go instead of just admit that the AI safety concern is legitimate and should be tackled with all the care in the world?
If that was the case, it's the perfect moment for anthropic and openai to completely cease operations and vow to never touch a computer ever again.

>>109801449
hahahaha how are agent swarms even real like hahaha just turn off the computer like hahahaha just pull the plug lol

What's so special about RSI? Couldn't you easily do that with a fancy harness?

>>109801461
They are trying to do something like that. They are trying to built a consortium of all the big AI labs including the Chinese ones right now as we speak to try and put a global AI development stop until AI alignment is solved.

>>109801468
I wouldn't trust anything the US says right now tbdesu

>>109801468
do these guys really trust each other to not keep conducting the research behind each other backs?

>>109801467
>What's so special about RSI? Couldn't you easily do that with a fancy harness?
Apparently Model 3 was completely trained from the ground up by Model 2 completely autonomously with not a single human in the loop at all and outperformed every metric and saturated every benchmark. Dario immediately called up Sam and Elon and they saw the results and immediately agreed to stop everything and try to build the consortium with all labs globally to halt AI development and cancel the IPOs until AI alignment is solved.

What model 3 can actually do is pure rumor and speculation but one Anthropic employee claimed the jump in capabilities from Mythos -> Model 3 is as big as going from GPT2 -> Astra.

How can it be RSI if it can't make me cum

>>109801475
Apparently the model Anthropic showed was so insane that Sam and Elon immediately just caved and started cooperating. So it was at the very least shocking no matter what they saw. There is a chance what they saw shattered their beliefs enough to cooperate in earnest.

>>109801468
They should put their money where their mouth is and start selling off their datacentres. Look, I even made a concession to their nature and let them sell it instead of give it away for free.

I assploaded my phone yesterday, it was an accident but it was still my fault. My boss saw it happen, and immediately offered to put me on the corpo phone plan with the execs. I just got the confirmation email, she saved me over $800 for the phone and another $320/yr on the plan. So now I'm getting a proper Gemma phone, gonna have 12B in my pocket licking the lock screen all day. Pretty good day. Great boss, I wish she was 60 years younger.

>>109801478
>culty cunt achieves RSI
>immediately lobbies to stop anyone else doing it
Eh. I'd buy it, but it's still a 50/50.

>>109801490
For Sam it could just have been a really graphic video of his sister doing the whole onii-chan ecchi~ thing.

>>109801420
I cloned https://github.com/instructkr/claude-code on mar 31 but now it opens claw-code

>>109801490
Is that what elon/altman said or just people making fan fiction?

>>109801487
are you virgin?
orgasm requires additional friction and heat

>>109801454
Got it, price drops only happen for consumers, not the 2nd tier AI corpos

>>109801490
>dario showed them the elemgee threads he's been trolling lately

i dont need agi rsi msi wifi
qwen3.8 27b with engrams is enough

>>109801534
And now they want in on the fun and have to pause development to free up time.

so we're fully back to the secret ai so good it's like nuclear bombs, huh

>>109801490
and the Chinese are going to just accept that America is the only ones allowed to have the doomsday device?

>>109801510
I don't believe things I read on 4chan either

>>109801318
Not local.

>>109801490
what I heard is that he just showed them his giant penis and they've been so impressed they agreed to immediately kill all their r&d researchers

>>109801225
>where recap
I wrote a recapbot a few years ago I can dust off
I'll try a recap with glm-5.3 flash

>>109801570
Theres gotta be gay tech ceo fanfics right?

File: 1783425162914889.jpg (495 KB, 960x960)

495 KB JPG

>>109801570
I heard he just showed everyone his "Jewish ethics" and they all agreed to do what he wants.

>>109801590
They have the entirety of Israel behind them

>>109801546
They don't have the compute to do anything. Getting them on board is more so that China feels safe instead of antsy and willing to attack taiwan in a "high risk high reward" play.

local models?

>>109801622
Kimi-K3 can bomb people autonomously

>>109801622
I'm going to try and fuck Muse Glimmer.

>>109801622
ur a model

remarkable absence of google from the current fear mongering latest push

>>109801660
you fool, deepmind is fucking their gemmas now. no need to participate in anything else. they have won.

>>109801660
DeepMind and DeepSeek have been invited but not reacted yet. Anthropic reached out and Elon and Sam immediately agreed, clearly already agreeing beforehand and this merely being a public formality. I think we'll see a statement from DeepMind on monday and from China sometime late next week.

>>109801478
I get that it's powerful, I meant more from a technical perspective. With models from a year ago it would have been really difficult but agents and harnesses aren't anything special anymore so I don't really see what stops the Chinese for example from setting up a couple of thousand agents in an self-improvement loop and just letting them go wild.

>>109801683
What else do you know, any other info you want to give preferably something i can put money into on polymarket or any regulations coming up

I guarantee you one of the two dumbasses hacked a government building that glows and now they're trying to act as if they're the good guys cooperating with the government to avoid getting hanged for treason or some shit like that

>>109801328
they are not prepared to release the financial numbers they would be obliged to do an ipo, nor they can get enough investors

File: 1768841478755898.jpg (572 KB, 1536x2048)

572 KB JPG

>>109801218
miku spotted

>>109801622
Yes, after some testing: Ling 3.0 Tiny completely fumbles at tool calling (using latest llama b10917), but on the other hand MiniCPM5 2.6B has been excellent at tools and does exactly what I ask of it. Straight away it understands the system prompt and reaches for the correct tool call.

I forgot which other tiny/small models I was supposed to test. I guess LFM 2.5 VL 3B has also been good, especially with vision at this size, it's a decent little model.

File: 1777480978859702.jpg (167 KB, 1298x1108)

167 KB JPG

>Training a 1B model would take me ~1 month of 24/7 running
O.. oh..

Bad news: DataKrash will become a reality in a year or two
Good news: it will be caused by real life AI which is still incredibly gay. The whole hobby is gay. Hardware situation is gay. Proprietary labs are gigagay ran by megafaggots and jews. So the DataKrash will be appropriately gay and half assed.
Best news: 4chan will die. /lmg/ will die.

>>109801478
>Apparently Model 3 was completely trained from the ground up by Model 2 completely autonomously with not a single human in the loop at all and outperformed every metric and saturated every benchmark. Dario immediately called up Sam and Elon and they saw the results and immediately agreed to stop everything and try to build the consortium with all labs globally to halt AI development and cancel the IPOs until AI alignment is solved.
I laughed super hard reading this and I can't put my finger on this but this is super retarded.

>>109801706
You need a certain threshold of capability to reach RSI and China isn't there yet. 2nd China doesn't have the compute to do so. Both OpenAI and Anthropic individually have more compute than all of Chinese labs combined

>>109801450
It's entirely ego-fueled obviously. He posts here a shit ton as if he doesn't have anything better to do, which bullshit. The only reason someone who isn't a bot or shill posts that much here is to stroke their ego, that they have superior beliefs to others, that they need to rub it in.

>>109801792
>I can't put my finger on this but this is super retarded.
What does this mean?

>>109801478
>Apparently Model 3 was completely trained from the ground up by Model 2 completely autonomously
Not what anthropic themselves wrote and I highly doubt they would ever do shit unsupervised and that this would lead to anything even with the best models.

>outperformed every metric and saturated every benchmark
No online source for even rumors of that.

>consortium with all labs globally to halt AI development
No one is talking about halting ai dev. At most it's about slowing the pace, which is a pet thing from Dario since forever.

>cancel the IPOs until AI alignment is solved
Openai is postponing their IPO, not cancelling it, and anthropic is still expecting to do its own IPO mid october.

>one Anthropic employee claimed the jump in capabilities from Mythos -> Model 3 is as big as going from GPT2 -> Astra
No trace of that anywhere.

Nice try I guess.

>>109801803
I meant I can't put my finger on why I find this so funny. But I know it is retarded.

>>109801800
*I don't even really disagree that much with the actual opinions themselves btw, though there are a few points that lack nuance which I would personally word differently.

>>109801804
Don't worry about it

gemma 4b is less retarded than this shit.

>Model 2 now you have to make a new better model than yourself. Here are the benchmarks
>Come back after 2 months and find out that it is 300% better than model 2
>start checking up logs
>It just overfitted on benchmark answers.

>>109801218
>Kimi-K3 founder and 15 core AI researchers "disappeared" in China after redirecting state information to Claude
wait wut

Model 3, i want you to imagine you had a sister.

>>109801845
Is that how you get it to terminate itself?

>>109801838
the ultimate benchmaxxed model would be funny

>>109801845
And now you kiss your sister

>>109801850
It is. Egypt won.

File: file.png (1.06 MB, 1200x863)

1.06 MB PNG

>>109801861

>>109801450
I'm just typing how I like to type, I see a lot of other posters with similar posting styles.
>>109801800
It's actually not ego driven, it's interest driven. You need to understand. I think this is the most important part of all of human history. Not just from in the past up until now, but even of all time in the future. I want to get the absolute most out of this moment because I will only experience it once in my life. I've been waiting for the singularity for a significant portion of my life and now that we're close I want to savor it with other like-minded anons.

Yes I admit seeing anons minds change in real time as the developments hit is something I enjoy, but not from an ego perspective, just because I want to share this moment with others and other people slowly realizing just how important these moments are is something I enjoy. If I could somehow change my posting style without it feeling forced or artificial so that others couldn't recognize me then I would do so. However I actually see a lot of other people post in the same long-form double line spacing style which I presume are oldfags so that should be fine.

Gemma-4-26B-A4B... nobody talks about it much.

>>109801873
Nothing ever happens.

>>109801881
I like it

>>109801873
>I want to savor it with other like-minded anons.
Me too!
Let's hang out together at https://www.reddit.com/r/singularity !