Good morning. It's Thursday, August 20, and we're covering how AI is changing learning, the nuclear power push behind data centers, and Cerebras’ challenge to NVIDIA.
Plus: yesterday’s Real or AI verdict and a new poll on AI ads.
OpenAI is previewing Private Safety Processing for select enterprise customers, an automated system that monitors for misuse across multiple sessions while retaining zero customer data. The move targets Anthropic's July policy of retaining 30 days of session data for Mythos-class covered models, which has aggravated enterprises with sensitive workloads.
Cerebras claims its new CS-4 delivers 30x more tokens per second per user than GPUs, challenging NVIDIA in AI inference. The rack-scale system combines three WSE-3 Turbo processors with 4 trillion transistors, using fast SRAM to reduce data movement.
Robot servers and automated kiosks are increasingly asking customers for tips. Tipsy Robot and Artly Coffee say their human staff receive the tips, while some experts warn employers could capture them. At Newark Airport, even an unstaffed kiosk prompted a tip on bottled water.
OpenAI’s second-quarter revenue rises 18%, but losses deepen as investors compare its progress with Anthropic’s faster growth. The ChatGPT maker reports $6.7 billion in Q2 sales, up from $5.7 billion in Q1, while operating margins move further into negative territory.
POWERED BY HYPERAGENT
Agents work better as a team
Hyperagent is where you set up specialist agents with defined jobs, shared context, and clear handoffs. They run on schedules in the cloud, pass work to one another, and bring you in when a decision requires your judgment.
ZEITGEIST POLL
ChatGPT Ads Expanding to 31 European Countries
Our team spent half of the morning meeting arguing about ads. The disagreement wasn't whether they work, it was where they stop being helpful and start being manipulation.
Where’s your line? |
EDUCATION
AI Raises Homework Scores but Can Undercut Exam Performance
New research suggests artificial intelligence can make students more productive without necessarily making them better learners. A study of 27,000 Chinese pupils aged 12–18 found that AI users raised average homework scores by 18% and cut assignment time from 64 to 45 minutes, but scored 20% lower on exams after six months.
The penalty was concentrated among students who used AI to rush through work; those who spent roughly as long on assignments as non-users saw little decline. A separate study found students using chatbots to learn unfamiliar material scored higher on tests even a week later, suggesting the educational effect depends heavily on whether AI serves as a shortcut or a tutor. → Read the full article here. (Paywall)
ENERGY
TerraPower’s Molten-Salt Reactor Could Help Power AI Data Centers
TerraPower plans to announce its first data center project this year, with construction expected to begin in 2027, as nuclear startups target AI’s growing electricity demand. Its 345-megawatt Natrium reactor uses molten sodium to store excess heat, allowing the plant to respond to rapid swings in electricity demand without constantly ramping the reactor itself.
That could be valuable for AI data centers, whose computing loads can rise and fall quickly and put additional strain on power equipment. TerraPower’s first reactor is already under construction in Wyoming, while the company announced in January that Meta agreed to buy eight Natrium plants. → Read the full article here.
AI DETECTION
AI Detection Tools Face a Losing Race Against Evolving Language
AI-generated and human-written language may be converging, making automated detection increasingly difficult. The article points to research suggesting people are adopting linguistic patterns associated with large language models (LLMs), while newer models can learn from increasingly AI-shaped human writing. Detection systems also face adversarial tools: one recent study found an adaptive “humanizer” evaded Pangram’s detector 76% of the time, while another found models fine-tuned on existing authors’ work passed as authentic 97% of the time. → Read the full article here. (Paywall)
NEWS
What Else is Happening
Unitree Shares Surge 460%: Shares jumped 460% in Shanghai, marking mainland China’s first humanoid robot listing.
GOP Warns AI Firms Over Data Centers: A Senate memo says data-center backlash threatens Ohio’s Jon Husted and could deter future projects.
China Pushes Digital Sovereignty: Beijing rejects pressure to choose sides in the US-China AI race, saying nations should pick partners based on needs.
Young Americans Sour on AI (Paywall): A Pew survey finds growing concern that AI will hurt jobs and bring few benefits to daily life.
China Courts Homegrown Tech Investors (Paywall): Beijing is steering AI-focused companies toward domestic capital and away from Wall Street.
POLL RESULTS: REAL OR AI
Can You Still Tell if an Image Is AI?
Yesterday's image was AI. Here's how you voted: 69% correctly identified it as AI, while 31% thought it was real.
 | 🟨🟨⬜️⬜️⬜️⬜️ Real (31%) 🟩🟩🟩🟩🟩🟩 AI (69%) |
