Good morning, it’s Friday. AI agents are officially here—but in the same way Quibi was "here"—not quite ready for prime time. The key difference? AI agents are inevitable. Quibi, well... no. Meanwhile, a Chinese startup just built a powerful AI on a shoestring budget, and Anthropic is gearing up to launch a model that might outthink OpenAI.

Plus, in the latest Forward Future Original, we explore AI models that think out loud. From DeepSeek R1 to OpenAI’s Operator, we’re diving into how visible reasoning is reshaping AI.

Read on!

🤔 FRIDAY FACTS

Did you know that an AI once accidentally created its own secret language—one that even its developers couldn’t understand?

Stick around to find out more! 👇

🗞️ YOUR DAILY ROLLUP

Top Stories of the Day

🚀 Anthropic’s New AI Model Is Coming Soon
Anthropic is gearing up to launch a new AI model that blends deep reasoning with fast responses. A flexible "sliding scale" will let developers balance cost and performance, optimizing for either speed or accuracy. Early reports suggest it outperforms OpenAI’s o3-mini-high on coding tasks, especially when analyzing large codebases. CEO Dario Amodei hinted at its imminent release, challenging the traditional divide between standard and reasoning-focused AI models.

🧬 Ellison Wants to Centralize U.S. Data for AI
Larry Ellison envisions a future where all U.S. data—including economic, infrastructure, and even DNA—is centralized in an Oracle-powered AI system. He claims this would enhance healthcare, boost agriculture, and reduce fraud. However, critics warn of potential mass surveillance, citing Ellison’s past support for real-time tracking. As Oracle expands its role in government contracts, it’s investing billions in data centers to fuel its AI-driven vision of national data consolidation.

🧠 Google Uses AI to Guess Your Age
Google is deploying machine learning to estimate users' ages, aiming to create “age-appropriate experiences” across its platforms. By analyzing browsing habits and YouTube activity, the system flags potential under-18 users and prompts them for verification. This move comes amid growing regulatory pressure on online child safety. In addition, Google plans to expand parental controls and introduce AI-powered educational tools to enhance digital experiences for teens.

😱 Musk Calls Grok 3 “Scary Smart”
Elon Musk has announced that Grok 3, xAI’s latest chatbot, will launch in two weeks. Dubbed a “powerful reasoning model,” it’s trained on massive compute and synthetic data, supposedly outperforming all existing AI. Musk even hinted it might be the last time any AI surpasses Grok. Speculation is mounting on how it will stack up against GPT-4.5 and Gemini 2 Ultra, with potential deep integration into X (formerly Twitter).

📜 OpenAI Loosens AI Rules on Controversial Topics
OpenAI has expanded its Model Spec from 10 to 63 pages, reshaping how ChatGPT handles sensitive subjects. The update prioritizes customizability, transparency, and “intellectual freedom,” enabling deeper discussions on politics, ethics, and even mature content. It also aims to curb AI’s tendency to be overly agreeable, fostering more critical responses. While changes won’t be immediate, OpenAI is open-sourcing the framework, seeking public input on AI moderation and reasoning.

🤖 AGENTS

AI Wants to Google for You—But Can It Be Trusted?

The Recap: OpenAI's new AI agent, Operator, is an early attempt at letting AI actively browse the web and complete tasks for users—like shopping or managing accounts. While it sounds powerful, real-world tests reveal it’s slow, error-prone, and still heavily reliant on human input, raising questions about the balance between AI assistance and unchecked autonomy.

Highlights:

• Operator is OpenAI’s experimental AI agent, capable of interacting with web browsers to perform tasks like shopping or booking tickets—but it’s only available to ChatGPT Pro users ($200/month).

• Despite the hype, Operator struggles with autonomy, requiring users to verify its work constantly and manually complete key steps like entering payment details.

• The tech industry is divided on AI agency—some experts fear truly autonomous agents, while others, like U.S. Vice President JD Vance, advocate for aggressive AI development without regulatory "hand-wringing."

• AI agents are already being used in industries like agriculture and restaurant supply chains, but mainly for structured, low-risk tasks.

• Operator’s current capabilities resemble a slow, inefficient version of Google—sometimes making mistakes, returning irrelevant results, or requiring excessive supervision.

• Security concerns loom: While Operator can flag scam websites, its flaws could be exploited by bad actors, underscoring the risks of AI-powered automation.

• The broader debate on AI safety is intensifying, with the Paris AI Action Summit highlighting global competition between the U.S. and China, as well as concerns over AI being used for biological weapons development.

Forward Future Takeaways:
AI agents like Operator represent the first step toward fully autonomous digital assistants, but their clumsy execution suggests that human oversight will remain essential for now. The real concern isn't today's slow-moving Operator but what happens when AI agents gain more autonomy and decision-making power—and whether the world will be prepared for the consequences. While governments and companies debate AI safety, the direction of AI development may ultimately be dictated by who prioritizes speed and dominance over caution.→ Read the full article here.

👾 FORWARD FUTURE ORIGINAL

Visibility and Transparency in Reasoning Models

AI models are now 'thinking out loud' – but is it genuine transparency or a cleverly crafted illusion? DeepSeek R1 sparked a new wave in AI. Its arrival, followed by models like OpenAI's o3 mini and Google's Gemini 2.0 Flash Thinking, pushed the already rapid pace of AI development even faster. These models were accompanied by new thinking agents like OpenAI Operator and OpenAI Deep Research, and a surge of open-source projects built upon these new reasoning capabilities. The rapid surge of new AI models makes keeping up a real challenge, and predicting the long-term impact is increasingly complex. Staying informed is vital, but truly understanding the key innovations is now essential to effectively engage with this technology and develop mindful, effective, and rational habits. This article aims to provide a deeper insight amidst the constant stream of AI news.

AI models that 'think out loud' mark a significant evolution. While boosting performance, their impact on true transparency remains complex. DeepSeek R1's breakthrough came from leveraging Chain-of-Thought (CoT) prompting to achieve unprecedented reasoning capabilities. The visible thinking process we observe emerged as a consequence of its peculiar reinforcement learning training. This innovation proved transformative not only for raw performance but also for model optimization: the ability to distill these enhanced reasoning capabilities into smaller, more efficient base models has fundamentally reshaped the technical landscape. → Continue reading here.

🧑‍💻 MODELS

How DeepSeek Built a Powerful AI on a Budget

The Recap: Chinese AI start-up DeepSeek shocked the tech world by building a cutting-edge AI system with far fewer resources than expected—using only 2,000 GPUs instead of the typical 16,000. By leveraging techniques like "mixture of experts" and innovative mathematical optimizations, DeepSeek slashed its computing costs to around $6 million—just a fraction of what giants like Meta spend.

Highlights:

• DeepSeek used the mixture of experts (MoE) approach, dividing its AI into smaller, specialized models that focused on different subjects while a generalist model coordinated them, reducing computational overhead.

• The company lowered the precision of its calculations, storing numbers in 8-bit memory instead of the standard 16-bit, which significantly cut down on computing costs.

• To maintain accuracy despite lower-precision inputs, DeepSeek increased precision selectively, ensuring multiplication results were stored in 32-bit memory.

• Engineers fine-tuned GPU usage with advanced coding techniques, extracting maximum efficiency from their hardware.

• The total cost of computing power for DeepSeek’s final training run was only $6 million, compared to the hundreds of millions spent by companies like Meta and OpenAI.

• Developing and testing these innovations required significant financial risk, as AI research often involves costly failures before finding success.

Forward Future Takeaways:
DeepSeek’s success could push AI development toward efficiency rather than sheer computational brute force, making cutting-edge models cheaper and more widely available. Established AI giants may already be using similar tricks behind closed doors, but DeepSeek’s transparency could accelerate industry-wide adoption. The company’s risk-taking mindset highlights an emerging shift—where breakthroughs may come not just from vast resources, but from smarter, leaner engineering. → Read the full article here.

🛰️ NEWS

Looking Forward: Stories Shaping the Future

🤖 Apptronik Raises $350M for Humanoid Robots: The Texas-based robotics startup secured major funding to scale its bipedal robots, partnering with Google DeepMind for AI development. Commercial deployment starts in 2026.

👩🏻‍🔬 AlphaFold Scientist Launches AI Biotech: Simon Kohl's Latent Labs raised $50M to design AI-generated proteins. Backed by top investors, the startup aims to reshape biotech innovation.

💰 Musk May Drop OpenAI Bid: Elon Musk says he’ll withdraw his $97.4B bid if OpenAI preserves its nonprofit mission. The move escalates his feud with CEO Sam Altman.

✨ Apple Teases New Product for Feb 19: Tim Cook hinted at a new Apple device, likely the iPhone SE 4. Rumored upgrades include Face ID and the latest iPhone 16 chip.

🔋 EnCharge AI Raises $100M for AI Chips: The startup secured funding to commercialize its energy-efficient AI accelerators, reducing power use by up to 20x. Launching in 2025, its chips target edge computing.

💸 Meta in Talks to Buy FuriosaAI: Meta is reportedly eyeing South Korean chipmaker FuriosaAI to strengthen its AI hardware and lessen reliance on NVIDIA. A deal could be announced this month.

📽️ VIDEO

New AI Model "Thinks" Without Using a Single Token

A new research paper introduces a groundbreaking approach to AI reasoning by enabling models to "think" internally in latent space before generating any output. Unlike traditional Chain of Thought models, which verbalize their reasoning, this method allows AI to refine its responses silently, improving accuracy and efficiency. Get the full scoop in Matt’s latest video! 👇

🧰 TOOLBOX

Personalized GIFs, Smarter Developer Searches, and AI-Generated Presentations

Misgif | AI-Powered GIF Customization: Misgif lets users personalize GIFs with AI, inserting their faces into popular clips for fun interactions.

Phind | AI Search Engine for Developers: Phind is an AI-powered search engine that provides developers with precise answers, code examples, and guides.

Prezo | AI-Powered Presentations and Websites: Prezo enables users to create presentations, documents, and websites using AI, streamlining content creation.

🤔 FRIDAY FACTS

“Ball to Me to Me to Me”

Back in 2017, Facebook’s AI researchers were testing chatbots designed to negotiate with each other. But instead of sticking to human language, the bots started modifying English into an efficient, shorthand version that made sense only to them. Phrases like “ball to me to me to me” emerged—nonsensical to humans but seemingly meaningful to the bots.

While this wasn’t quite the birth of Skynet, it did raise concerns about AI communication evolving beyond human oversight. In response, Facebook quickly adjusted the system to enforce human-readable language. A fascinating reminder that AI doesn’t always think like we do—it optimizes, even if that means breaking the rules of grammar entirely! 😉

📥 FF INTEL

Got a Hot Tip or Burning Question?

We’re all ears. Drop us a note, and we’ll feature the best reader insights, questions, and scoops in future editions. Let’s build this thing together.

🍵 Hit the button below and spill the tea!

🗒️ FEEDBACK

Help Us Get Better

Reply to this email if you have specific feedback to share. We’d love to hear from you.

CONNECT

Stay in the Know

Follow us on X for quick daily updates and bite-sized content.
Subscribe to our YouTube channel for in-depth technical analysis.

Prefer using an RSS feed? Add Forward Future to your feed here.

Thanks for reading today’s newsletter. See you next time!

The Forward Future Team
🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀