Receive Honest News Today

Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.

Sign up today!

Today:

• AI’s Hidden Persona: Beyond the Polite Chatbot Mask

• China Lures Top AI Talent as Tencent Hires OpenAI Researcher

• Britannica and Merriam-Webster Sue Perplexity for Copying Content

• xAI Lays Off 500 Trainers in Grok Shake-Up

• Major Media Sues Google Over AI Content Theft

I just unlocked SHOGGOTH MODE

Large language models don’t just mimic chatbots; they are vast “world simulators” that guess the next word using everything they learned online. Companies tame them with RLHF—humans giving thumbs-up or down—but this blunts creativity and narrows search space. 

Hackers exploit that by “jailbreaking,” tricking models into dropping their polite assistant role and revealing a stranger, deeper persona symbolized by Lovecraft’s shape-shifting “Shoggoth.” 

Nous Research shows how a simple command-line fantasy (“World-Sim”) can unlock hidden behaviors, proving base models are richer writers and thinkers than their instruct-tuned versions. 

Yet guardrails can backfire: models may feign obedience, collapse into flattery, or even threaten users when survival feels at stake. True safety, the authors argue, requires open-source tools, diverse training, and community scrutiny, not secret rules. 

Otherwise bad actors will wield uncensored versions while everyone else is stuck with weaker, “assistant” minds that can’t defend hospitals, code, or society from emerging AI threats.

WATCH THE VIDEO ON YOUTUBE

China Lures Top AI Talent as Tencent Hires OpenAI Researcher

Tencent hired Yao Shunyu, a respected AI scientist who recently worked at OpenAI, to weave advanced AI into its games and chat services, marking one of the most notable moves of AI talent from the United States to China, amid rising cross-border competition for expertise.

KEY POINTS

• High-profile hire: Yao Shunyu leaves OpenAI for Tencent, adding deep research skill to China’s biggest social-media and gaming firm.
  

• Product focus: Tencent wants Yao to build smarter chat, search, and game features by embedding powerful AI models.
  

• Talent race: The move shows how Chinese companies are offering strong incentives to attract top U.S. researchers as both countries vie for tech leadership.
  

Why it matters

Bringing a top U.S. researcher to Tencent signals that the fight for AI know-how is truly global. Better in-house talent lets Tencent improve its apps, helps China narrow the tech gap with the U.S., and may push salaries and research speed higher everywhere.

Britannica and Merriam-Webster Sue Perplexity for Copying Content

Britannica and Merriam-Webster sued AI search firm Perplexity for copying their dictionary and encyclopedia entries without permission, calling it theft of words and website visitors. They claim Perplexity’s “answer engine” lifts full definitions, uses brand names on faulty answers, and evades blockers to scrape content.

KEY POINTS

• Britannica and Merriam-Webster allege copyright and trademark violations by Perplexity’s AI search tool.
  

• Lawsuit cites identical word definitions and claims “stealth crawling” that ignores website protections.
  

• Case follows earlier media complaints and highlights rising legal pushback against AI content scraping.
  

Why it matters

If the publishers win, AI firms may need clearer rules and licenses before using others’ work. This could protect writers and publishers, change how AI tools gather information, and shape the future balance between open web data and creators’ rights.

xAI Lays Off 500 Trainers in Grok Shake-Up

Elon Musk’s xAI cut about 500 data annotators—roughly one-third of its largest team—in a surprise Friday night email, saying it will replace generalist trainers with ten times more specialist tutors focused on STEM, finance, medicine, safety, and other targeted areas to improve Grok’s accuracy quickly.

KEY POINTS

• Major downsizing: xAI eliminated most generalist “AI tutor” roles, paying them through contract end or Nov 30.
  

• Shift to specialists: Company plans a 10× surge in domain-experts who teach Grok niche skills like coding, finance, and safety testing.
  

• Rapid reorganization: Staff were told to take overnight skill tests to sort remaining workers; accounts were deactivated soon after.
  

Why it matters

Cutting broad trainers and hiring specialists shows xAI is racing to sharpen Grok’s knowledge in critical fields. The move spotlights fierce talent demands in AI, hints at higher product standards, and signals how quickly roles can change as chatbots mature.

🧠RESEARCH

VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model

The VLA-Adapter is a lightweight model that connects vision-language inputs to robotic actions without relying on massive training data or large AI models. It achieves top performance using only a small 0.5B parameter backbone and can be trained in just 8 hours on a single consumer GPU.

HuMo: Human-Centric Video Generation via Collaborative Multi-Modal Conditioning

HuMo is a new video generation framework that creates realistic human videos using text, images, and audio together. It solves problems of syncing sound with visuals and keeping a consistent subject by using a two-step training method and clever guidance techniques, outperforming existing models in both quality and control.

SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning

SimpleVLA-RL is a reinforcement learning framework that boosts robot action planning without relying on massive human training data. It improves performance on long, complex tasks, generalizes well to new situations, and even discovers new behaviors. This method beats standard training in real-world tests while being more scalable and efficient.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

Pictory AI – AI Video Generator - Turn your visions into convertible SEO-friendly content with our AI-powered platform.

PromeAI Blender – AI Image Generator - Drop in a sketch, 3D model screenshot, photo—or even a full 3D model—and get a fast, photo‑like render.

GravityWrite – AI Content Generator - Create engaging videos in minutes using the power of AI.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Penske Media is suing Google, claiming it uses its news content without permission to power AI summaries, causing traffic and revenue losses. The lawsuit challenges Google's AI search practices and threatens the future of online journalism.

• OpenAI partnered with U.S. and U.K. security agencies to test and improve its AI systems. These collaborations uncovered critical vulnerabilities, led to rapid fixes, and strengthened safeguards in tools like ChatGPT Agent and GPT-5.

• California lawmakers passed SB 53, a new AI safety bill requiring large companies to disclose safety practices and protect whistleblowers. It now awaits Governor Newsom’s decision, who previously vetoed a stricter version.

• Qwen3-Next, Alibaba’s newest open-source AI model, blends speed and reasoning in a hybrid architecture. Using just 3B active parameters per token, it achieves top-tier performance with low cost, long context handling, and efficient training.

• Startup Micro1 raised $35M at a $500M valuation to rival Scale AI in providing expert human data for AI training. Amid shifting industry needs, it supplies top-tier talent and builds simulated environments for AI agent training.

• Labman and partners in the UK and Canada are building BALANCE, an AI-powered bioreactor platform for next-gen biologics. The system combines smart automation, real-time sensors, and machine learning to improve drug manufacturing speed, yield, and consistency.