🗞 Google Drops AI Weapons Ban, ByteDance's Human AI, and Altman's India Vision

🔎 The Latest on the AI Frontier:

• Google Removes AI Weapons Ban: Shifts defense stance, prioritizes responsible innovation

• ByteDance's OmniHuman-1: New AI system creates realistic human videos

• Altman Bullish on India's AI: Sees rapid growth amid declining costs

• Google Search's AI Evolution: Search giant transitions to AI assistant

• Figure AI Goes Solo: Ends OpenAI partnership, promises humanoid breakthrough

• Google's Gemini Update: Code hints at 2.0 Pro or Flash 2.0 release

• Anthropic Boosts Claude Security: New system blocks 96% of jailbreaks

• Other news you might find interesting

⚠️ Google removes its previous pledge to not develop AI for weapons and surveillance

• Google quietly updated its AI principles page, eliminating the "applications we will not pursue" section that prohibited AI weapons development.

• The company's new stance emphasizes "responsible AI" and collaboration with governments on national security, while maintaining alignment with international law and human rights principles.

• This policy shift comes amid ongoing employee protests over Google's military contracts with the U.S. and Israeli forces, while the Pentagon confirms some AI models are accelerating military operations.

🤖 ByteDance unveils OmniHuman-1 for AI-driven human video generation

• New framework creates realistic videos from single images and audio/video inputs, supporting any aspect ratio and body proportion.

• System uses multimodality motion conditioning for mixed training, overcoming data scarcity limitations of previous approaches.

• Currently unavailable for download, with researchers warning against fraudulent service offerings.

🌏 OpenAI CEO Sam Altman sees major AI opportunities in India

• OpenAI's second-largest market after US has tripled users in last year, with Altman supporting India's plans to build its own AI models.

• Predicts 10x reduction in AI intelligence costs every 12 months, making frontier models more accessible despite high infrastructure needs.

• Comments on DeepSeek, saying he was "extremely skeptical" of claimed low costs but acknowledges it's a "good model."

🚀 Google Search is evolving into an AI assistant that could change how billions interact with the web

• Google remains dominant in GenAI search with 16.5 billion U.S. visits in December 2024 - 30x more than ChatGPT's 517 million visits, though ChatGPT's numbers doubled year-over-year.

• The company is integrating new AI systems like Project Astra (for live video processing) and Project Mariner (an AI agent that interacts with websites on users' behalf), moving beyond just information retrieval.

• This shift raises concerns for businesses relying on Google traffic, as AI-driven search could reduce the need for users to visit websites directly.

✌️ Figure AI ends OpenAI partnership to launch its own proprietary AI system for humanoid robots

• Figure AI, valued at $2.6 billion with $1.5 billion in total funding, terminates its collaboration with OpenAI following a claimed major breakthrough in robot AI technology.

• CEO Brett Adcock states the company will demonstrate unprecedented humanoid capabilities within 30 days, citing the need for integrated hardware-software development as the key reason for the split.

• The decision comes as Figure's robots are already being deployed in BMW's South Carolina factory, while OpenAI signals its own humanoid ambitions through a recent trademark filing.

📱 Google expected to update Gemini today with potential new model release

• Internal code suggests two possibilities: either the launch of Gemini 2.0 Pro or a new Flash 2.0 version with thinking process capabilities.

• Evidence includes a pricing FAQ update mentioning 2.0 Pro (later reverted), and Sundar Pichai's announcement of model updates during Q4 earnings call.

• The release follows January's Gemini 2.0 Flash web launch, with new changelog placeholders suggesting further updates planned through February 10.

🛡️ Anthropic launches Constitutional Classifiers to protect Claude against jailbreaks

• New security system reduces jailbreak success rates from 86% to 4.4%, while increasing compute costs by 23.7% and benign query refusal rates by just 0.38%.

• System uses synthetic data to distinguish harmful from harmless requests, focusing on filtering CBRN-related content.

• Public demo invites users to test Claude's safeguards, aiming to gather real-world data for further improvements.

🖨 More news you might find interesting:

• OpenAI unveils subtle brand refresh with new logo and custom typeface.

• ChatGPT expands WhatsApp integration with voice and image capabilities.

• Meta releases Frontier AI Framework focusing on open source and risk mitigation

• Hugging Face develops open-source alternative to OpenAI's deep research tool in under 24 hours.

• AMD reports strong Q4 2024 with $7.66B revenue, driven by AI and data center growth.

• China launches antitrust probe into Google amid escalating trade tensions

• Ex-Google engineer faces new charges for alleged AI trade secret theft to China.

• Workday cuts 1,750 jobs (8.5% of workforce) citing AI pivot.

• Alphabet reports strong Q4 2024 with $96.47B revenue, driven by AI initiatives.

• xAI plans greywater plant in Memphis faces Council pushback.

Have any feedback? Send us an email