Open Source

US Lab Retrains Chinese AI to Reflect Western Values

US Lab Retrains Chinese AI to Reflect Western Values

⚡This could change how AI answers political questions — and who controls the truth.

Deep Dive

Nous Research, a US-based AI lab, has released a new AI model called Arete-27B. It's based on a Chinese model from Alibaba, but Nous retrained it to promote Western values like freedom of expression and democracy. The company says the original model sometimes dodged questions or gave answers that echoed Chinese government positions on sensitive topics. After retraining, they claim the model's factual accuracy improved and its propaganda-like answers plummeted from 22% to just 0.9%.

So what? This matters because AI models are increasingly used for everyday tasks — from answering questions to helping with work. If an AI has a hidden bias, it can shape what you believe. Nous's experiment shows that biases can be changed, but it also raises a bigger question: who gets to decide what values an AI should have? As AI becomes more integrated into our lives, the ability to tweak its worldview could influence public opinion and trust.

The catch is that this isn't a perfect fix. Nous only used a simple training method and didn't test every possible scenario. The model might still have biases, and the evaluation was done by the company itself. Also, retraining AI to fit specific values could be used to manipulate information — imagine a model that always favors one political side. For now, Arete-27B is open-source, meaning anyone can download and use it, so its impact will depend on how it's used.

Key Points
  • Nous Research retrained a Chinese AI model to give more Western-friendly answers on sensitive topics.
  • The new model's propaganda-style answers dropped from 22% to 0.9% in their tests.
  • This shows AI biases can be changed, but raises concerns about who controls AI's viewpoints.

Why It Matters

AI you use daily could have hidden biases; this shows they can be altered — for better or worse.

🔒 FREE VIP ACCESS

Read the full breakdown & get daily AI executive intelligence

Join like-minded builders, researchers, and AI professionals who read The Neural Feed daily. 100% free. No spam.

🔔 BREAKING RELEASE RADAR INSTANT ALERTS • 1-CLICK PAUSE

Never miss the next Open-Weight Models drop

We matched Open-Weight Models from this briefing. Pick your AI radar tracks below for instant breaking release alerts + our free daily morning briefing.

Includes full article unlock • Only major releases • Unsubscribe anytime
🎁 FREE EXECUTIVE DOWNLOAD ISSUE #21

Download This Week's 8-Page Executive AI Briefing (PDF)

Our curated executive intelligence packet: in-depth breakdowns of frontier models, market movements, plus full-color AI comics.

  • ✓ 8-page formatted executive PDF briefing
  • ✓ Frontier AI models & technical breakthroughs
  • ✓ Full-color satirical AI comic strips
Instant 1-click download • 100% free • Unsubscribe anytime
📬 Get the top 10 AI stories daily