Google DeepMind vs OpenAI: The Race for AI Video Generation Dominance

Google DeepMind announced Veo 2, a next-gen video AI capable of producing 4K, 2+ minute videos, surpassing OpenAI’s Sora. Currently, videos are limited to 720p, 8 seconds via Google’s experimental VideoFX tool. Improvements include sharper visuals, realistic physics, better camera control, and fluid dynamics, though challenges remain with coherence and character consistency.

Trained on large video datasets, likely including YouTube, Veo 2 raises copyright concerns. DeepMind uses filters and SynthID watermarking to address risks like deepfakes. Alongside Veo 2, Google upgraded Imagen 3 for better image generation in styles like photorealism and anime.

Link to full article

Only 6.1% of U.S. Companies Using AI in Production

Goldman Sachs' research reveals limited AI adoption among U.S. companies, with just 6.1% using AI to produce products or services, a slight rise from 5.9% in Q3. Key highlights include:

1. Adoption by sector:

   • Finance and insurance lead in AI usage.

   • Information, manufacturing, and education saw declines.

2. Productivity impact:

   • Generative AI boosts labor productivity by 23-30%, according to studies and company insights.

3. AI investment trends:

   • Strong growth is projected, especially in the semiconductor industry, with a 37% revenue increase forecast by next year.

4. Company size:

   • Large firms (250+ employees) report 10% adoption and expect further growth in the next 6 months.

5. ROI challenges:

   • PYMNTS data shows limited returns on AI investments, with only 13% of CFOs reporting “very positive” ROI (down from 27% in March).

   • Companies with strong ROI are increasing AI budgets significantly (19%), while those with negligible ROI show modest increases (6.2%).

6. SMBs:

   • Small and medium-sized businesses doubled their AI adoption rate but face concerns like cybersecurity and identifying practical uses for AI.

Overall, while AI adoption is growing, companies face challenges in achieving ROI and fully the technology’s potential.

Link to the full article

OpenAI cofounder Ilya Sutskever says the way AI is built is about to change

Ilya Sutskever, OpenAI’s cofounder and former chief scientist, made a rare public appearance at NeurIPS in Vancouver, where he shared insights into the future of AI. Sutskever predicted that pre-training, the current approach to AI model development using vast amounts of data, will eventually end because the internet’s human-generated data is finite — a concept he compared to fossil fuels.

He foresees a shift toward AI systems that are both agentic (autonomous systems capable of decision-making and task execution) and capable of reasoning. Unlike today’s pattern-matching AI, future systems will be able to process information step-by-step, akin to human-like thinking, making them more unpredictable but also capable of understanding from limited data without confusion.

Drawing parallels to evolutionary biology, Sutskever suggested AI might discover new scaling approaches, similar to how hominid brain evolution diverged from other mammals.

Link to the full article

Grok-2 From Elon Musk is now free for every X user

Since launching Grok-2 in August, the Grok team has been enhancing the model on 𝕏, introducing features like web search, citations, and the Aurora image generator. A new version of Grok-2, three times faster with improved accuracy, instruction-following, and multi-lingual support, is now rolling out to all users for free, with Premium users getting higher usage limits.

Key updates include:

1. Web search and citations: Grok combines posts from 𝕏 and the broader internet for real-time, accurate answers with source citations.

2. Aurora: A photorealistic image generator for creative photos and memes, with the “draw me” feature letting users create personalized images.

3. Grok button: A new feature on 𝕏 posts to provide context and insights on trending topics.

For developers, Grok introduced:

• Two new API models: grok-2-1212 and grok-2-vision-1212, improving accuracy and steerability.

• Reduced pricing: $2 per 1M input tokens, $10 per 1M output tokens, with $25 free credits for new sign-ups.

• Upcoming API access to Aurora for photorealistic image generation.

These updates enhance Grok’s capabilities for users and developers alike.

Link to the full article

The first 8 days of OpenAI 12 days of OpenAI

1. Day 1 - o1 and o1 Pro Mode: OpenAI introduced a premium subscription plan for $200/month, granting access to the powerful o1 pro mode, designed for researchers and engineers needing advanced reasoning models.

2. Day 2 - Reinforcement Fine-Tuning Research Program: This program allows developers to fine-tune models for specific, complex, domain-specific tasks, enhancing their effectiveness in specialized areas.

3. Day 3 - Sora: A video generation model capable of creating detailed scenes with multiple characters, motion, and accurate background depictions, understanding real-world dynamics.

4. Day 4 - Canvas: A new interface for collaborative projects, allowing users to work alongside ChatGPT in writing and coding tasks. It’s built on GPT-4o and is available in beta for Plus and Team users, with broader access planned.

5. Day 5 - ChatGPT on Apple Devices: ChatGPT is integrated into iOS, iPadOS, and macOS, offering expertise and document-understanding capabilities directly within Apple’s ecosystem, eliminating the need for switching between tools.

6. Day 6 - Santa Mode and Advanced Voice: Introduced voice interaction capabilities, allowing users to converse with ChatGPT through spoken input and receive spoken responses, creating a more natural and conversational experience.

7. Day 7 - Project in ChatGPT: A new feature for organizing chats, files, and custom instructions in one place, simplifying management of ongoing work or multiple tasks.

8. Day 8 - ChatGPT Search: A new search functionality that allows ChatGPT to access real-time information by searching the web, providing up-to-date answers.