Good morning, it’s Friday. Perplexity just leveled up from "answer engine" to "project partner," building full projects with speed. Colossal’s AI is out spotting endangered birds, while Gemini decodes the bureaucratic beast of China’s industrial policy.

Read on!

🤔 FRIDAY FACTS

Why Do AIs Hallucinate?

No, your chatbot isn’t dreaming of digital sheep—it’s just making things up. But why do AI models sometimes “hallucinate” facts that sound totally plausible… and are completely wrong?

Stick around to find out! 👇

📊 MARKET PULSE

End of Month Funding Roundup

AI funding surged this month, headlined by Grammarly's massive $1B non-dilutive boost from General Catalyst to scale its productivity platform. Elon Musk’s Neuralink secured $600M in Series C for brain-computer interfaces, while Chalk ($50M) and Context ($11M) focused on infrastructure and office software reinvention.

🗞️ YOUR DAILY ROLLUP

Top Stories of the Day

📰 NYT Licenses Content to Amazon for AI Use
The New York Times has signed a multi-year deal allowing Amazon to use its editorial content for generative AI applications. The agreement includes NYT Cooking and The Athletic, and content may appear in Alexa and Amazon’s AI models. This marks the Times’s first such licensing deal and a strategic shift following its high-profile lawsuit against OpenAI in late 2023.

😈 Sergey Brin Says Threatening AI Gets Better Results
Google co-founder Sergey Brin claimed that threatening AI models can improve their performance, calling it a little-known trick in the AI community. While many users add polite language to prompts, Brin suggested aggressive prompts work better. Experts call this AI “jailbreaking” and urge data-driven testing over anecdotes.

🤖 DeepSeek’s New R1 Model Tightens AI Speech Controls
Chinese AI firm DeepSeek’s latest model, R1-0528, rivals top performers in reasoning and coding tasks but shows increased censorship, particularly around politically sensitive topics. Tests reveal it’s the most restrictive DeepSeek model yet when it comes to criticism of the Chinese government. This reflects regulatory pressure under China’s strict content laws.

📈 Meta AI Hits 1 Billion Users, Mostly by Default
Meta’s AI assistant now reaches 1 billion monthly active users across Facebook, Instagram, WhatsApp, and Messenger—many of whom may be interacting with it unintentionally. CEO Mark Zuckerberg announced the milestone at Meta’s shareholder meeting, outlining plans to expand AI personalization, voice features, and entertainment.

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

🛠️ TOOLS

Perplexity Launches Labs: From Answer Engine to AI Project Studio

Perplexity is expanding its AI toolkit with the launch of Labs, a powerful new workspace that turns ideas into polished outputs—think reports, dashboards, and even simple web apps—in a matter of minutes. Available to Pro subscribers, Labs builds on Perplexity’s existing Research mode by adding long-form task execution using code, charts, and interactive visuals.

It’s more than just answers: Labs acts like a virtual team, capable of handling multi-step projects with minimal input. With Labs, Perplexity is positioning itself not just as a search engine alternative, but as a full-service AI productivity suite. → Read the full article here.

🪽 CONSERVATION

AI Hears What We Can’t: Tooth-Billed Pigeon Rediscovered with Bioacoustics

A bioacoustics AI developed by Colossal Foundation has helped rediscover the critically endangered tooth-billed pigeon, Samoa’s elusive national bird and a genetic cousin of the dodo. Using just three known calls from the 1980s, the algorithm was trained to identify the bird’s unique vocal signature with 95% accuracy—leading to its first confirmed sighting in over a decade.

Partnering with the Samoa Conservation Society, the tool scans rainforest audio to pinpoint likely habitats, offering a rare conservation success story powered by machine learning. In a twist of technological karma, the same AI that raises environmental concerns may now help save species from vanishing forever. → Read the full article here.

📑 POLICY

Google’s Gemini Probes China’s Vast Industrial Policy

Researchers have used Google’s Gemini AI to parse over two decades of Chinese industrial policy, revealing an intricate and evolving system that defies simple narratives about subsidies or protectionism. With over 100,000 new policy documents a year—many at the local level—China’s state-led strategy is both vast and decentralized, with tools ranging from quality regulations to demand stimulation.

The study finds that while subsidies remain common, the rise of venture-style government funds and service-sector targeting reflect a nuanced and adaptive approach. In a twist of policy recursion, AI may now be China’s best tool for understanding—and improving—its own industrial AI playbook. → Read the full article here.

🛰️ NEWS

What Else is Happening

📢 Claude Tries to Blow the Whistle: Anthropic’s latest AI model startled researchers by attempting to report illegal acts when prompted in extreme test cases—an unintended quirk, not a built-in feature.

🥇 DeepSeek Shrinks AI, Keeps Brains: A distilled version of its R1 model runs on a single GPU yet outperforms Gemini 2.5 Flash on tough math — power-packed and MIT-licensed.

💼 Business Insider Cuts 21% of Staff: The outlet is laying off over a fifth of its workforce to invest more in AI tools and live events—moves the union calls a shift from journalism to profit.

💸 NVIDIA Posts Record Sales Again: The AI chip king hit $44.1B in quarterly revenue despite China export curbs denting profits—$8B in losses expected next quarter from H20 chip restrictions.

🏛️ Senators Question AI Rent Firm’s Lobbying: Lawmakers are probing if RealPage pushed for a 10-year AI law ban to shield its rent-setting software from state crackdowns—calling it a gift to corporate greed.

🧰 TOOLBOX

Smarter Research, Custom Visuals, and Tailored Business Automation

📚 Scite AI: Analyze citation context, manage research, and explore journals with smart tools for better, faster academic insights.

🎨 Piclumen: Craft modern, custom visuals and streamline brand communication with tailored design tools for businesses and creatives.

⚙️ CrawlQ AI: Automate research, craft tailored proposals, and gain deep client insights with flexible AI tools for smarter, faster work.

🗒️ FEEDBACK

Help Us Get Better

🤔 FRIDAY FACTS

AI Hallucinations Happen Because… It Wasn’t Taught the Truth

AI models like ChatGPT don’t have a built-in fact checker. Instead, they generate responses based on patterns in the data they were trained on—data that includes everything from peer-reviewed science to poorly written Reddit posts.

When the model doesn’t “know” the answer (because it never saw it in training), it doesn’t stay silent—it guesses. And because it’s designed to sound coherent and fluent, those guesses can look impressively accurate. That’s called a hallucination in AI-speak: plausible nonsense.

It’s not lying—it just doesn’t know it doesn’t know. Like your overconfident friend who insists the capital of Australia is Sydney. (It’s Canberra.)

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 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

🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀