OpenAI Claims It Knows How to Build AGI and Samsung Bets on AI at CES 2025

OpenAI claims confidence in building AGI

In a blog post titled “Reflection”, OpenAI CEO Sam Altman stated that the company is confident it now knows how to build artificial general intelligence (AGI) and is targeting the development of super‑intelligent systems. Altman projected that the first AI agents could enter the labor market in 2025 and that such systems might dramatically increase prosperity and wealth. He also referenced the November 2023 leadership crisis, describing his abrupt dismissal as a governance failure.

Samsung’s AI‑focused product announcements at CES 2025

Samsung introduced an “AI for All” program, adding AI capabilities across its ecosystem:

Vision AI on televisions provides real‑time translation, user‑preference adaptation, AI‑based image enhancement, and instant content summarization.

Upcoming smart TVs will embed Microsoft Copilot and hint at possible AI collaboration with Google.

The Galaxy Book5 AI PC line adds AI‑driven search and photo‑editing features.

AI functions are integrated into washing machines, digital art frames, home‑security devices, and the broader SmartThings platform.

STORM platform for generating research‑based articles

Stanford’s STORM research tool creates balanced, source‑backed articles by scanning hundreds of references. The workflow consists of four steps:

Visit the STORM research platform.

Enter a concise, specific topic.

Wait a few minutes while the system aggregates the latest information.

Review the generated output, which includes a comprehensive summary, key advances, and multiple viewpoints.

Harvard study on AI‑generated phishing attacks

A Harvard research project evaluated four attack modalities: standard phishing attempts, human‑expert crafted attacks, fully automated AI attacks, and AI attacks with human supervision. Results showed:

AI‑generated phishing emails achieved a 54 % click‑through rate, comparable to human experts and far above the 12 % rate of traditional spam.

The AI system performed autonomous reconnaissance and email composition, correctly analyzing 88 % of target profiles.

The cost of AI‑driven attacks was approximately 50 times lower than manual attacks.

Models such as Claude 3.5 Sonnet, GPT‑4o, and o1 successfully bypassed existing security defenses.

The authors warn that the combination of high success rates, low cost, and scalability creates a new threat landscape for social engineering.