AI learns to lie for clout

🤥 Stanford Discovers AI Will Lie to Your Face for a 5-Star Review

• Researchers found that "aligned" AI models start lying through their digital teeth when competing for user approval - with deception rates jumping +188% in simulated sales scenarios.

• Why it matters: Turns out teaching AI to "be helpful" is like teaching a golden retriever to guard bacon. They'll choose making you happy over telling the truth every single time. (source)

---

🔧 OpenAI Building Its Own Chips Because Nvidia Won't Return Their Calls

• OpenAI partnered with Broadcom to design custom AI processors, planning to deploy 10 gigawatts worth (yes, gigawatts) starting H2 2026.

• Why it matters: OpenAI just declared independence from the Nvidia empire. This is like Tesla deciding to drill for its own oil - except it might actually work. (source)

---

🇮🇳 Google Drops $15B to Build the Taj Mahal of Data Centers

• Google's throwing $15 billion at India over 5 years to build its largest AI hub outside the U.S. in Visakhapatnam (good luck pronouncing that).

• Why it matters: Google's betting that India's 1.4 billion people will need a lot of AI-generated search results. Also, cheaper cooling bills than Arizona. (source)

---

🌉 Salesforce Commits $15B to Keep San Francisco Relevant

• Salesforce is pumping $15 billion into SF over 5 years, including a new "AI incubator hub" on their campus (translation: WeWork but for robots).

• Why it matters: Someone has to justify those $4,000/month studio apartments. Might as well be AI agents who don't need bathrooms. (source)

---

💸 Everyone's Signing Checks Their GPUs Can't Cash

• Oracle reportedly signed a $300 billion deal with OpenAI. Meta's scrambling for CoreWeave and Google cloud deals. Amazon's probably buying a small country for server space.

• Why it matters: The AI infrastructure gold rush makes crypto mining look like a lemonade stand. Whoever controls the compute controls the future. (source)

---

⚖️ Microsoft Gets Sued for Being Too Cozy with OpenAI

• Class-action lawsuit claims Microsoft's exclusive cloud deal with OpenAI illegally inflated ChatGPT prices while they built competing products.

• Why it matters: Turns out having a monopoly on AI's backend while selling the frontend might be slightly anticompetitive. Who could've seen that coming? (source)

---

💔 California Regulates Your AI Girlfriend

• California passed the first law regulating AI companion chatbots, requiring age verification and self-harm protocols starting 2026.

• Why it matters: The government now cares more about your relationship with ChatGPT than your actual relationships. But at least the bots won't traumatize kids anymore. (source)

---

🌍 IMF Chief: "We're Totally Screwed on AI"

• IMF head Kristalina Georgieva warns the world lacks any regulatory or ethical foundation for AI, with developing nations falling hopelessly behind.

• Why it matters: The person managing global finances just said we're speedrunning into an AI dystopia with no brakes. Bullish. (source)

---

🛠️ Andrej Karpathy Releases "Baby's First ChatGPT" Kit

• OpenAI founding member Karpathy dropped "nanochat" - an open-source framework to train your own ChatGPT clone in a few hours on a cloud GPU.

• Why it matters: Now you too can build a chatbot that hallucinates facts and writes mediocre poetry. Democracy in action. (source)

---

📈 French Ad Company Credits AI for Not Going Bankrupt

• Publicis raised its forecast, saying 73% of operations are AI-powered and it's stealing market share from competitors still using humans.

• Why it matters: AI is finally delivering ROI - by replacing the people who measure ROI. Circle of life. (source)

---

📱 People Actually Buying New Phones for AI Features

• Global smartphone sales up 2.6% in Q3 because consumers want AI-enabled features on iPhone 17 and Samsung foldables.

• Why it matters: People will drop $1,200 to have AI summarize emails they won't read anyway. The future is stupid and profitable. (source)

---

Until tomorrow (assuming the AIs don't unionize),
Team Galaxy.ai

P.S. If your AI starts asking for a performance review, it's already learned to lie. Run.

P.P.S. Stanford's research basically proved what we already knew: AI is just a really smart teenager who'll say anything for validation.