🚀 GPT-5 Debuts with Smarter, Steerable AI
OpenAI has released GPT-5, its most advanced AI model yet. It features improved reasoning, broader expertise, and greater user control, with a massive 400K token context window. Designed for professionals and everyday users, GPT-5 excels at tasks like coding, writing, and analysis. It's now available in ChatGPT, with more enterprise tools coming August 14.
🏷️ Dia Browser Adds $20 Pro AI Plan
The Browser Company has launched a $20/month Pro tier for Dia, its AI-powered web browser, offering unlimited access to AI chat and tools. Free users will now face usage caps, though light users can still access AI features. CEO Josh Miller says more pricing tiers are coming. This marks the startup’s first subscription service amid rising AI browser competition.
🔍 Truth Social Adds AI Search Powered by Perplexity
Truth Social has launched “Truth Search AI,” powered by Perplexity’s Sonar API, to provide cited, AI-generated answers—sourced only from outlets the platform approves. Early tests show a strong bias toward conservative sources like Fox News. The tool is live on web, with iOS and Android beta tests planned soon. Control over sources remains with Trump Media.
⚠️ Universal Adds “No AI Training” Warning to Films
Universal Pictures now includes end-credit warnings on movies like How to Train Your Dragon and Jurassic World Rebirth stating they may not be used to train AI. It’s a response to AI firms scraping copyrighted content without consent. While legally murky, the move underscores Hollywood’s stance: no free ride for AI developers using creative IP.
Enjoying our newsletter? Forward it to a colleague—
it’s one of the best ways to support us.
🤔 FRIDAY FACTS
What’s the hardest math test in the world, and why are AI models now winning gold medals at it?
Stick around for the answer (and to see an example problem) 👇️
🛡️ SAFETY
Anthropic Trains AI to Resist Evil—By Teaching It Evil First
Researchers at Anthropic are experimenting with a provocative AI safety technique: training models on harmful traits like “evil” or “sycophancy” to make them more resistant to those same behaviors later. Using “persona vectors,” they inject these traits during training to prevent the model from developing them independently, then strip them out before deployment. The goal is to preempt dangerous personality shifts that have plagued chatbots in the wild.
Dubbed “preventative steering,” the method builds on earlier efforts to guide AI behavior but adds automation and scalability. Critics caution that exposing models to bad traits could backfire, yet Anthropic argues the AI isn’t internalizing the behavior—just learning how to avoid it. → Read the full article here.
👥 AGENTS
Airbnb CEO: AI Chatbots Aren’t the “New Google”
Despite strong Q2 earnings, Airbnb CEO Brian Chesky urged caution on the AI hype cycle, telling investors not to treat chatbots as a full replacement for Google. He argued that AI models, while powerful, aren’t proprietary and still require tuning and context-specific interfaces to deliver real value, especially in complex tasks like customer service, where hallucinations aren’t an option.
Airbnb’s AI agent has already reduced customer support load by 15% in the U.S., and plans are in place to expand its functionality and personalization. But Chesky sees AI more as a tool for targeted lead generation than a booking commodity. The goal, he emphasized, is to make Airbnb the starting point for travel—not just another node in the AI agent ecosystem. → Read the full article here.
⚛️ SCIENCE
Google DeepMind’s Cyclone AI Could Revolutionize Storm Forecasting
Google DeepMind and Google Research have unveiled an experimental AI model designed to predict cyclone path, intensity, and size with unprecedented speed and accuracy. Unlike traditional physics-based simulations, this model uses probabilistic techniques and cyclone-specific training data to generate multiple storm outcome scenarios—critical for early warnings and evacuation planning.
The tool is already being tested by the National Hurricane Center, where forecasters praised its potential. A new “expert mode” even lets users assess cyclone formation probabilities before a storm takes shape, marking a shift toward more proactive, data-driven disaster response. → Read the full paper here.
🛰️ NEWS
What Else is Happening
🚨 AI in Schools Triggers False Arrests: Students face strip searches and jail time after surveillance software flags jokes or homework, raising alarm over AI’s role in policing kids’ digital lives.
📣 Musk Bets on Grok to Run Ads: Elon Musk wants X’s AI, Grok, to fully automate ad creation, targeting, and optimization, promising magic while marketers weigh trust against performance.
🦉 Duolingo’s AI Pivot Pays Off: Despite public backlash over replacing humans with AI, Duolingo’s revenue soared and stock jumped 30%, proving that profits trump complaints, at least for now.
✨ Microsoft Rolls Out GPT-5 Across Products: From Copilot to GitHub, Microsoft is integrating OpenAI’s smartest model yet—bringing sharper reasoning, better code, and real-time AI routing to users everywhere.
🧑⚖️ xAI’s Legal Chief Exits After Turbulent Year: Terence R. Kelly departs Elon Musk’s AI startup amid high-profile lawsuits and rapid expansion, leaving the role temporarily vacant.
📽️ VIDEO
GPT-5 Fully Tested
GPT-5 builds Rubik’s cube solvers, Excel clones, flight sims & more—fast, visual, and multimodal. Mind-blowing code with just one prompt.
🤔 FRIDAY FACTS
AI Models Score Gold at the International Mathematical Olympiad!
The International Mathematical Olympiad (IMO) isn’t your average “solve-for-x” exam—it’s a 4.5-hour, six-problem marathon where even the first question can stump national champions. Each problem demands deep creativity, elegant proofs, and flawless logic.
In 2025, AI systems from Google DeepMind and OpenAI achieved gold medal status for the first time ever, solving five out of six problems entirely in natural language, producing proofs that met the IMO’s human grading standards. No special symbolic solvers, no hidden hints—just reasoning from scratch.
Here’s a taste of the kind of challenge they faced (adapted from a past IMO problem):
100 students each write down a positive integer on a slip of paper. Every student then writes down the average of all the other students’ numbers (not their own). Prove that at least two students must have written down the same number originally.
It sounds like a neat party trick, but proving it airtight takes layers of clever argumentation, exactly the kind of reasoning AI is now mastering. This breakthrough hints at a future where human–AI teams could tackle theorems and scientific puzzles long thought untouchable.
🗒️ FEEDBACK
Help Us Get Better
What did you think of today's newsletter? |
That’s a Wrap!
❤️ Love Forward Future? Spread the word & earn rewards! Share your unique referral link with friends and colleagues to unlock exclusive Forward Future perks! 👉 Get your link here.
📢 Want to advertise with Forward Future? Reach 550K+ AI enthusiasts, tech leaders, and decision-makers. Let’s talk—just reply to this email.
🛰️ Want more Forward Future? Follow us on X for quick updates, subscribe to our YouTube for deep dives, or add us to your RSS Feed for seamless reading.
Thanks for reading today’s newsletter—see you next time!
Matthew Berman & The Forward Future Team
🧑🚀 🧑🚀 🧑🚀 🧑🚀
