Good morning, it’s Tuesday. If this week were a stock ticker, it’d flash “📉” as China’s DeepSeek sends Wall Street into a spiral, NVIDIA’s valuation teeters on the edge, and doubts about U.S. AI dominance intensify. Thankfully, Matt has put together a video breakdown to help connect the dots—don’t miss it below.

Meanwhile, in our latest Forward Future Original article, we recap the standout moments from the World Economic Forum 2025, where global leaders discussed AI, AI, and yes, even more AI.

Read on!

🖼️ DeepSeek’s Janus-Pro AI Aims to Outdo DALL-E
DeepSeek’s new Janus-Pro AI models promise to outshine DALL-E 3 with superior image generation and analysis capabilities. Ranging from 1B to 7B parameters, these compact models excel in benchmarks and are available on Hugging Face under an MIT license. Despite a resolution limit of 384x384, their flexibility makes them formidable competitors.

🏗 Stargate AI Project to Exclusively Serve OpenAI
Stargate, a $100 billion AI infrastructure project backed by OpenAI and SoftBank, will focus solely on powering OpenAI's operations. With endorsements from President Trump and key partners like Oracle, the initiative plans to expand its budget to $500 billion over four years. A Texas data center is underway, but funding remains incomplete, relying on SoftBank and OpenAI to raise $15 billion each. Stargate aims to meet OpenAI’s skyrocketing data and computing demands.

🛡️ Cyberattack Hits DeepSeek Amid AI Boom
DeepSeek, the rapidly rising Chinese AI startup, suffered a cyberattack that temporarily halted new user registrations as its app soared to the top of Apple’s U.S. App Store. Renowned for its cost-efficient DeepSeek-V3 model, the company has directly challenged U.S. AI dominance and export controls. By leveraging NVIDIA's H800 chips, DeepSeek trained its model for under $6 million—raising questions about the impact of U.S. tech sanctions and the extent of China’s AI capabilities. (More on this below.)

💬 Meta AI Taps Instagram & Facebook for Personalization
Meta AI now uses data from Facebook, Messenger, and Instagram to deliver personalized responses, including insights from user activity and location. The chatbot’s new memory feature tailors interactions by recalling preferences, though memories are limited to single chats and can be deleted. While CEO Mark Zuckerberg touts the updates as enhancing usability, Meta's rocky data privacy history fuels skepticism. The feature is rolling out across the U.S. and Canada.

The Recap: A surge in investor anxiety over Chinese A.I. advancements, driven by DeepSeek, triggered a sharp sell-off in tech stocks. NVIDIA and other major players faced steep declines as the markets reevaluated A.I. dominance and competition dynamics.

Highlights:

Forward Future Takeaways:
DeepSeek’s emergence underscores the rising global competition in A.I., posing a serious challenge to U.S. tech dominance. As the market recalibrates expectations, companies will need to demonstrate both innovation and efficiency to maintain investor confidence. With valuations under scrutiny, the A.I. sector may face increased pressure to balance high growth ambitions with financial prudence. → Read the full article here.

The World Economic Forum (WEF) is an international organization based in Geneva that was founded in 1971. Every year, leading figures from politics, business, science and society gather in Davos, Switzerland, to discuss global challenges and future issues. The declared aim of the WEF is to improve the state of the world through dialog and cooperation.

This year, discussions focused on artificial intelligence. Technological advances and their impact on various areas of life were discussed in detail by renowned experts and leading CEOs. 

The announcement of the “Stargate” project, which came from the White House just one day before the Davos meeting, was particularly explosive. This ambitious program provides for investments of 500 billion US dollars to establish the USA strategically as the world's leading nation in the field of AI.

As part of the “Stargate” project, significant partnerships were forged, including with Sam Altman, CEO of OpenAI, Larry Ellison, co-founder and CTO of Oracle, and Masayoshi Son, founder of SoftBank. These figures played a central role in the strategic plans presented at a high-level meeting at the White House. → Continue reading here.

🤖 Grok 3 Spotted in Beta Testing by xAI Users: Musk's latest AI model teased over the weekend, impressing with code generation and riddle-solving. Grok 3 is trained with "10x" more compute than its predecessor.

💰 Mercor Hits $2 Billion Valuation in AI Talent Boom: The startup connects experts in law, medicine, and beyond with AI firms like OpenAI, skyrocketing from $250M to $2B in months.

🎗️ Reid Hoffman Launches AI Cancer Startup: Partnering with Siddhartha Mukherjee, Manas AI will use generative AI to accelerate drug discovery, starting with breast cancer, prostate cancer, and lymphoma.

🚫 Google DeepMind Unveils MONA to Tackle Reward Hacking: The MONA framework prevents AI agents from exploiting multi-step reward systems, using short-term optimization and human-approved long-term evaluations.

🦾 Uber PM Automates 30% of Job with AI: Nimisha Sharath saves 100 minutes daily using AI for meeting notes, research, and strategy prep. This frees her to focus on relationship building and stakeholder management.

🚀 Iran Deploys AI-Enhanced Missiles in Gulf Drills: The IRGC launched Ghaem and Almas missiles, equipped with AI, from Mohajer-6 and Ababil-5 drones, successfully striking simulated targets.

DeepSeek R1, a groundbreaking open-source AI model, challenges the AI status quo with its low $5M training cost. Released by a small Chinese firm, it rivals top models like OpenAI’s GPT-4, sparking debates on U.S. tech dominance, efficiency, and the power of open-source innovation. Get the full scoop in Matt’s latest video! 👇

Aviral Kumar (@aviral_kumar2) shared insights from a new blog by CMU students, highlighting how optimizing test-time compute in LLMs requires solving a meta reinforcement learning (RL) problem. The approach shifts focus from simply generating answers to using additional tokens for reasoning and discovering solutions dynamically at test time.

The blog argues this process enhances generalization by leveraging token-level "information gain" during output generation. It also discusses measuring test-time compute efficiency and designing objectives to train models for this purpose. → Learn more.

MeetMinutes | AI Meeting Notes: MeetMinutes streamlines meetings with transcription, summaries, tracking, and calendar integrations across major platforms.

Anthropic | Citations API Launched: Anthropic’s Citations API boosts Claude’s accuracy with source-based references for trustworthy, grounded AI responses.

Trae | Adaptive AI IDE: Trae accelerates coding with adaptive AI, streamlining workflows and enhancing collaboration for developers.

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

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
🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀