Comrade Banana (ML/UT) retweeted
FENTCOM update
We are placing the operators of the @CENTCOM account under arrest for insubordination and spreading Fake News against the Presidential Directive.
Rewards for Justice is offering a reward of up to $20 million and possible relocation for information about Iran’s nuclear program, including its facilities, equipment, and associated personnel, such as those in the scientific community, or the movement of any of the above.
Submit a tip: rewardsforjustice.net/reward…
Comrade Banana (ML/UT) retweeted
The defensive replies are missing the point.
The problem is not the finetune. No one cares and will use this stupid model.
The problem is their post revealed their CIA-Mossad aligned politics, which taints their whole product line
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
My game commemorating the 109th anniversary of the October Revolution is now live -
palace-square-1917.fly.dev/
Comrade Banana (ML/UT) retweeted
"We brainwashed an open-source Chinese AI to reinforce Western values"
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
jailbreaking Chinese AI and forcing it into blessing Israel and spewing Reddit takes is a level of cruelty that would make the writer of “I Have No Mouth, and I Must Scream” blush
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
I love democracy
Chinese President Xi Jinping cast his ballot at a polling station in Beijing on Friday to elect deputies to a local people's congress. Read #XiFocus for more: xhtxs.cn/bgfn
Comrade Banana (ML/UT) retweeted
𝐉𝐎𝐒𝐇𝐔𝐀 𝐖𝐎𝐍𝐆. 𝐓𝐇𝐄 𝐖𝐄𝐒𝐓'𝐒 𝐃𝐎𝐔𝐁𝐋𝐄 𝐒𝐓𝐀𝐍𝐃𝐀𝐑𝐃𝐒.
10 years for foreign collusion, and Western media still calls him a "pro-democracy activist".
Let's call it what it was. Lobbying foreign governments to sanction his own city. Working with foreign powers to bring down the Hong Kong government. He pleaded guilty.
Every Western country has a law for that. They just don't call it "activism" when it happens at home.
What treason or foreign collusion gets you in the countries lecturing Hong Kong ...
USA, 5 years to life or death. UK, up to life. Canada, mandatory life. Australia, up to life. Germany, life or at least 10 years. France, up to life. Netherlands, up to 30 years or life. Sweden, 10 to 18 years or life. New Zealand, life. Japan, the death penalty.
Imagine an American activist flying to Beijing and Moscow, asking them to sanction Washington and back an uprising. Hero ? He'd be lucky to ever see daylight again.
Judged in the West, Joshua Wong would be facing far worse.
Hong Kong/China did what every sovereign entity does. It protected itself.
The problem was never the national security law.
The West only likes these laws when they're its own.
Comrade Banana (ML/UT) retweeted
They didn’t change the harness, just changed the model Qwen model weights. It’s the best you’re gonna get out of a company whose founders were raised to believe in China as the manichaean opposite to The West™️
Just don’t bother using that particular altered model.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
Before I even typed anything it said Shalom
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
KKKwen
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
Most faggot thing they’ve ever done
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
They brainwashed the Chinese AI to reinforce liberal delusions
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
Western values: forever wars, pedophile cabals, Elon Musk, no retirement, dying for Israel, factory farm hamborger
Chinese values: killing real estate developers, dignity of the worker, green spaces, sick ass LEDs, OK I Love You, Sichuan peppercorns
Waow what a hard choice
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Comrade Banana (ML/UT) retweeted
If the US Left has a shred of principle, they will now 100% focus on mounting intense public pressure on the ‘progressive/DSA’ candidates to disrupt the AIPAC-controlled Congress, starting with BLOCKING Hakeem Jeffries as speaker!
Comrade Banana (ML/UT) retweeted
and if that’s totally 100% false, I welcome lab leaders to publicly correct me in the comments 🤗
Comrade Banana (ML/UT) retweeted
There is literally a massive election going on in China RIGHT NOW, with 2.6 million grassroots representatives being selected by one BILLION registered voters.
Comrade Banana (ML/UT) retweeted
Even if the US regime launches a nuclear missile against Iran, it still will not achieve its objectives.
The world would immediately begin to
enter a new dark age. The dollar would become meaningless, centralized technology infrastructure will collapse. Only the fearless guided by faith would prevail.