Top AI & Tech News (Through Feb. 9th)

Hello AI Citizens 🤖,

Large Language Models are dominating news headlines around the world but did you know that the first chatbot was developed in 1966 by Joseph Weizenbaum?
ELIZA, an early AI chatbot that mimicked human conversation by using simple pattern matching, paved the way for modern AI assistants.

This week’s AI updates showcase transformative product launches in the LLM world, emerging trends, and important developments redefining the future of AI.

Here are the key headlines shaping the AI & tech landscape:

• Tülu 3 405B: The Open-Source AI Challenging GPT-4o

• Berkeley Team Recreates DeepSeek AI for Just $30

• Gemini 2.0 Just Dropped: Google’s Most Powerful AI Yet!

• OpenAI Launches ‘Deep Research’—AI That Writes Reports in Minutes

• OmniHuman: TikTok’s New AI Brings Photos to Life!

• Meta’s Yann LeCun Predicts the Next Tech Revolution

• EU’s AI Ban Takes Effect—Here’s What’s Now Illegal

• OpenAI & SoftBank Partner to Launch ‘Cristal Intelligence’ for Enterprise AI

Ready to unpack these stories? Let’s dive into the details 🚀

Tülu 3 405B: The Open-Source AI Challenging GPT-4o

Ai2 has introduced Tülu 3 405B, one of the most powerful open-weight AI models to date. This model outperforms DeepSeek V3, GPT-4o, and Llama 3.1 405B in certain key tests.

A big part of its success comes from Reinforcement Learning from Verifiable Rewards (RLVR), a method that helps the AI improve by focusing on tasks with clear right and wrong answers. Training a model this large required 256 GPUs running in parallel, as well as techniques like supervised fine-tuning (SFT) and Direct Preference Optimization (DPO). Despite challenges like high compute costs and tuning limitations, Tülu 3 405B shows that open-source AI can compete with the best proprietary models. Source: Ai2

💡Tülu 3 405B proves that scalable post-training techniques can unlock new performance levels for open-weight AI, challenging the dominance of closed-source models. Tech leaders should explore RL-based training methods to enhance LLM capabilities efficiently.

Berkeley Team Recreates DeepSeek AI for Just $30

A research team at UC Berkeley, led by Ph.D. candidate Jiayi Pan, has recreated key features of DeepSeek R1-Zero using a small 3B-parameter model—for only $30. This low-cost experiment highlights how open-source AI can challenge big tech by significantly reducing development costs. Despite its size, the model learned self-verification and search skills, refining its own answers over time. To test the model, researchers used Countdown, a math puzzle where AI initially guessed randomly but improved through reinforcement learning (RL). Their findings suggest that advanced RL techniques don’t require massive compute resources, making AI research far more accessible. Source: jiayi_pirate on X.

💡TinyZero, a lightweight AlphaZero-inspired RL framework that runs on low-end hardware allows researchers and hobbyists to train Reinforcement Learning models without expensive GPUs, opening new possibilities in AI research, game AI, and robotics

Gemini 2.0 Just Dropped: Google’s Most Powerful AI Yet!

Google has officially launched Gemini 2.0, calling it its “most capable” AI model suite so far. The release includes 2.0 Flash, a high-speed model for large-scale tasks, 2.0 Pro Experimental, optimized for coding and reasoning, and 2.0 Flash-Lite, its most cost-efficient AI yet. With pricing as low as 0.75 cents per million tokens, Google aims to make advanced AI more accessible.

This rollout is part of a wider AI arms race between Google, OpenAI, Meta, Microsoft, and Anthropic, as all major players push toward agentic AI—models that can complete complex tasks independently. Google envisions AI agents that think ahead, take action, and function as universal assistants, competing with OpenAI’s Operator and Anthropic’s AI-powered computer assistant. Source: CNBC

💡 Gemini 2.0’s rollout brings powerful AI to more users, making high-performance models faster, cheaper, and safer. Developers and businesses should explore these models for scalable AI applications.

OpenAI Launches ‘Deep Research’—AI That Writes Reports in Minutes

OpenAI has launched Deep Research, an AI-powered research analyst designed to analyze, synthesize, and summarize vast amounts of online information in minutes. Built on OpenAI’s latest reasoning model, o3, the tool scans text, images, and PDFs, citing sources to generate detailed reports for professionals in finance, science, and engineering. The move comes amid growing competition from China’s DeepSeek, pushing OpenAI to accelerate product releases. Deep Research automates time-intensive research, completing tasks in 5 to 30 minutes, with source citations for verification. It’s currently available only in the U.S. for ChatGPT Pro users ($200/month), reflecting the high processing costs of this advanced reasoning model. Source: The Guardian

💡AI-powered research tools like Deep Research could transform business intelligence, finance, and strategic planning, enabling leaders to make faster, data-driven decisions. However, companies must validate AI-generated insights to avoid blind reliance on automated analysis.

OmniHuman: TikTok’s New AI Brings Photos to Life!

ByteDance, the parent company of TikTok, has introduced OmniHuman, an AI model capable of generating highly realistic videos of people and animals talking, gesturing, and moving—all from a single photo. Trained on 18,700 hours of human video data, OmniHuman outperforms existing models, producing detailed hand and body movements from multiple angles. While not yet available to the public, its capabilities position ByteDance as a major player in the AI-generated video space, competing with models used for virtual influencers, customer service agents, and even deepfake-style political endorsements.

OmniHuman pushes the limits of AI-generated realism, enabling photo-to-video transformation with audio, gestures, and full-body movement. Though ByteDance claims it did not train the model using TikTok data, concerns remain about misuse in misinformation and digital identity cloning. Source: Forbes

💡 Business leaders must prepare for a future where AI-generated avatars impact marketing, entertainment, and communication. While tools like OmniHuman could enhance branding and customer engagement, they also raise ethical concerns—making transparency, security, and digital trust critical priorities.

Meta’s Yann LeCun Predicts the Next Tech Revolution

Speaking at the Queen Elizabeth Prize for Engineering (QEPrize) ceremony, Meta’s chief AI scientist Yann LeCun predicted a major AI breakthrough within the next five years, enabling domestic robots and fully autonomous cars. While current AI excels at language, LeCun emphasized that it still lacks real-world understanding, requiring new predictive models to bridge the gap. LeCun made these remarks as he and six other engineers—including Geoffrey Hinton, Yoshua Bengio, Fei-Fei Li, Jensen Huang, and Bill Dally—were awarded the prestigious QEPrize for their groundbreaking contributions to machine learning and artificial intelligence. LeCun envisions AI models that understand and predict physical reality, moving beyond today’s language-based systems. Meanwhile, Bengio warns that as AI grows more powerful, nations must collaborate on safety measures to prevent unintended consequences. Source: The Guardian

💡 Business leaders should anticipate AI’s shift from digital to physical applications, preparing for automation in manufacturing, logistics, and service industries. As AI nears human-like reasoning, companies must invest in both innovation and AI governance to stay ahead in a rapidly evolving landscape.

EU’s AI Ban Takes Effect—Here’s What’s Now Illegal

The EU AI Act’s ban on “prohibited AI systems” officially took effect on February 2, 2025, marking the first major enforcement step under the landmark regulation. These banned AI practices, outlined in Article 5, include social scoring, AI-based criminal risk assessments, untargeted facial recognition, and emotion recognition in workplaces and schools. Ireland, however, has carve-outs for law enforcement, allowing certain AI systems that are prohibited elsewhere in the EU. Companies violating Article 5 face fines of up to €35 million or 7% of global revenue. The EU’s strict stance signals a zero-tolerance approach to AI systems that pose “intolerable risks”, with broader AI regulations set to roll out by 2026. Source: Mondaq

💡 Business leaders must immediately assess AI compliance risks, especially in areas like biometric surveillance, predictive policing, and automated decision-making. Companies using AI in HR, security, and consumer profiling should proactively adapt to avoid heavy fines and legal scrutiny.

OpenAI & SoftBank Partner to Launch ‘Cristal Intelligence’ for Enterprise AI

OpenAI and SoftBank have announced a $3B-per-year partnership to develop and market Cristal Intelligence, an AI system tailored for enterprises. This collaboration will integrate AI agents across SoftBank’s ecosystem, with Arm and SoftBank Corp. using AI to automate operations and drive innovation. A new joint venture, SB OpenAI Japan, will bring Cristal Intelligence to major Japanese companies, setting a model for global AI adoption. SoftBank becomes the first company to deploy Cristal Intelligence at scale, integrating OpenAI’s latest models. Arm’s compute platform will power AI advancements, while SoftBank Corp. plans to automate 100M+ workflows using AI. Source: Soft Bank Group

💡Key Takeaway: Business leaders should prepare for AI-driven enterprise automation, as custom AI solutions become key to efficiency, decision-making, and competitive advantage. Companies that invest early in AI agents will be better positioned to streamline operations and unlock new growth opportunities.

About The AI Citizen Hub - by World AI X

This isn’t just another AI newsletter; it’s an evolving journey into the future. When you subscribe, you're not simply receiving the best weekly dose of AI and tech news, trends, and breakthroughs—you're stepping into a living, breathing entity that grows with every edition. Each week, The AI Citizen evolves, pushing the boundaries of what a newsletter can be, with the ultimate goal of becoming an AI Citizen itself in our visionary World AI Nation.

By subscribing, you’re not just staying informed—you’re joining a movement. Leaders from all sectors are coming together to secure their place in the future. This is your chance to be part of that future, where the next era of leadership and innovation is being shaped.

Join us, and don’t just watch the future unfold—help create it.

For advertising inquiries, feedback, or suggestions, please reach out to us at [email protected].