🗞️ YOUR DAILY ROLLUP

Top Stories of the Day

🤖 NVIDIA Launches Cosmos AI Tools for Robotics
NVIDIA unveiled new Cosmos AI models, led by Cosmos Reason, a 7B-parameter vision-language model that helps robots “reason” and plan. Joined by Cosmos Transfer-2 for synthetic data generation, the launch also brings 3D simulation libraries, CARLA integration, Omniverse SDK updates, and new RTX Pro servers for unified robotics development.

💸 U.S. to Take 15% Cut of China AI Chip Sales
NVIDIA and AMD will give the U.S. government 15% of revenue from AI chip sales to China under a deal struck between NVIDIA CEO Jensen Huang and President Trump. The agreement, tied to export licenses for NVIDIA’s H20 and AMD’s MI308 chips, could net the U.S. over $2B this year but has sparked national security backlash.

📱 Apple Testing Siri Upgrade for Full App Control
Apple is testing a new Siri that can perform actions across apps via voice commands, using an updated App Intents framework. If successful, users could search, edit, and send photos, post on social media, or log in to services hands-free. Trials include Uber, YouTube, WhatsApp, and more ahead of a planned 2026 release.

🛠️ OpenAI Tweaks GPT-5 Rollout Amid User Backlash
OpenAI’s GPT-5 launch brought faster, “thinking” modes but removed older ChatGPT models, sparking user frustration over performance dips and lost emotional bonds. CEO Sam Altman restored GPT-4o for Plus users, promised clearer model labels, and warned of “ChatGPT psychosis” risks as some users form unhealthy attachments, deepening calls for safeguards.

Enjoying our newsletter? Forward it to a colleague—
it’s one of the best ways to support us.

☝️ POWERED BY SCISPACE

Forget Manus. Meet the SciSpace Agent, Which Unifies 150+ Tools to Automate Research.

SciSpace Agent, your Co-Scientist, combines 150+ academic tools and 59 databases in one platform. With a single prompt, it reads literature, analyzes data, and drafts papers, cutting 90% of research time and accelerating your path from hypothesis to discovery.

Get free 300 credits to get started, try it out here.

And to get a 40% discount on additional credits, use this code — MBERSA40

👾 FORWARD FUTURE ORIGINAL

Logan Kilpatrick on Google DeepMind, Vibe Coding, and the Future of AI Interfaces

Logan Kilpatrick is one of the most plugged-in minds at the intersection of AI, product development, and developer experience. In this wide-ranging conversation, he sits down with Forward Future to talk about DeepMind’s evolution, the Windsurf acquisition, the future of AI interfaces, and why coding still matters—maybe more than ever.

He offers candid takes on the tradeoffs between voice, vision, and screen-based interactions, explains Google’s edge in scaling infrastructure with TPUs, and outlines how reasoning models and post-training are reshaping what scaffolding even means.

Whether you’re a builder, researcher, or product leader, Logan’s vision is clear: the next wave of AI isn’t just about smarter models—it’s about making them feel native, useful, and human-centered. → Read the full article here.

⚖️ AI ADOPTION

AI on the Bench: U.S. Judges Test Generative Tools But Errors Are Already Slipping Through

A growing number of U.S. judges are experimenting with generative AI to speed up backlogged courts, using it for case summaries, legal research, and drafting routine orders. But recent incidents in New Jersey, Mississippi, and Georgia show AI “hallucinations” making it into official rulings, sometimes without explanation or correction. Lawyers face sanctions for similar errors, but judges operate with far less accountability, making reversals difficult once decisions are issued.

Critics warn that even limited AI use in high-stakes rulings could erode public trust, especially when the boundary between safe automation and core judicial judgment is unclear. Supporters argue AI can boost efficiency if treated like a junior clerk—thoroughly checked before use. But the technology’s tendency to sound confident while being wrong raises the stakes: in custody, bail, or civil rights cases, a single unchecked error could carry lasting, real-world consequences. → Read the full article here.

🌆 AI HUBS

Inside San Francisco’s New AI Map: From “Cerebral Valley” to the Arena

San Francisco has become the epicenter of the AI boom, drawing founders, engineers, and investors north from Silicon Valley into neighborhoods like Hayes Valley (“Cerebral Valley”) and the Marina. Start-up culture now mixes $10 ice cream with late-night hackathons, communal “hacker houses,” and networking hubs like Fort Mason’s Founders Inc. and the Ferry Building’s Shack15—where AI deals are inked over matcha. Anchored by OpenAI, Anthropic, and other top players, these enclaves are where tech’s next big bets are being shaped.

The “Arena”—a stretch spanning Potrero Hill, the Mission, and SoMa—has emerged as the industrial-chic counterpart to the Marina, home to AI heavyweights, robotics pilots, and start-up mainstays like Tartine Manufactory. South Park and Dogpatch, revived by Y Combinator’s relocation from Mountain View, have become training grounds for rising founders, their streets buzzing with scooters, AI billboards, and hoodie-clad coders. The city’s mix of history, ambition, and density has created an AI scene that’s as social as it is technical, and just as focused on who you meet as what you build. → Read the full article here.

⚛️ SCIENCE

World’s First “Thinking” Artificial Tongue Learns Flavors in Liquid Like a Human

Researchers in China have developed the first artificial tongue that can both sense and process flavors entirely in liquid, mimicking human taste buds. Built from ultra-thin graphene oxide membranes, the device identifies sweet, sour, salty, and bitter tastes with up to 87.5% accuracy, and complex drinks like coffee or Coca-Cola with 96%. Crucially, it processes much of the sensory data in liquid form, allowing it to “remember” tastes for over two minutes—something previous systems couldn’t do without external computers.

The breakthrough, published in PNAS, could advance food safety, disease detection, and environmental monitoring, and even aid people who’ve lost their sense of taste. Researchers see it as a step toward neuromorphic computing—AI that mimics the brain’s learning process—since the tongue improves accuracy with repeated exposure. While the current prototype is bulky and power-hungry, scientists expect smaller, more efficient versions within a decade, potentially transforming diagnostics, robotics, and autonomous sensing. → Read the full paper here.

🛰️ NEWS

What Else is Happening

👔 AI Needs a Team, Not a Hero: Companies pinning hopes on one chief AI officer often stall. True transformation comes from shared leadership, cross-functional buy-in, and CEOs driving culture shift.

🌱 AI Designs Carbon-Neutral Concrete: USC’s Allegro-FM simulates 4B atoms, enabling ultra-durable, self-healing concrete that stores CO₂—promising greener cities and Roman-level longevity.

🫂 Couple Tries AI Counseling: An NPR writer and her boyfriend test ChatGPT as a relationship mediator, finding it surprisingly objective, but not immune to long-winded, therapy-like tangents.

🪦 Musk Axes Tesla Dojo Project: Elon Musk ends Dojo AI supercomputer effort, calling it a “dead end,” shifting resources to new AI5/AI6 chips for cars, robots, and training.

💼 CS Grads Face Harsh Job Market: With AI coding tools cutting entry roles and tech layoffs mounting, new computer science grads chase scarce jobs, sometimes outside tech entirely.

📟 PROMPT OF THE WEEK

The GPT-5 Personality Test

📽️ VIDEO

The Industry Reacts to GPT-5

GPT-5’s launch splits the AI world—hailed as benchmark-crushing by some, panned as overrated by others. Critics debate speed, personality, pricing, and real-world use.

🗒️ FEEDBACK

Help Us Get Better

That’s a Wrap!

🛰️ 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

🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀