SunBrief#26: Can AI feel? Anthropic explores

Welcome to the SunBrief

Today in SunBrief 🌞

• Anthropic’s new research explores AI welfare

• The Smartest AI Library on the Internet

• Gemini’s Veo2 Challenges ChatGPT’s Sora

• AI Highlights of the Week

• Is China making “Human vs Robots”?

• Last Week’s Poll Result

Find out why 1M+ professionals read Superhuman AI daily.

In 2 years, you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

1. Join the Superhuman AI newsletter – read by 1M+ people at top companies

2. Master AI tools, tutorials, and news in just 3 minutes a day

3. Become 10X more productive using AI

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

Anthropic’s new research explores AI welfare

​Anthropic Launches Research Initiative on AI Model Welfare

Anthropic has announced a new research program focused on exploring the potential consciousness and welfare of advanced AI models, such as its Claude series.

Research Focus:

• Assessing AI Experiences: The program investigates the potential for AI models to have experiences, preferences, or signs of distress, and what moral obligations might arise from such capabilities.

• Interdisciplinary Approach: This initiative intersects with Anthropic's existing efforts in alignment science, safeguards, character development, and interpretability, aiming to address the complex questions surrounding AI consciousness and welfare.

• Collaborative Insights: Anthropic's research is informed by contributions from leading experts, including philosopher David Chalmers, who have highlighted the near-term possibility of consciousness and agency in AI systems.

This endeavor reflects Anthropic's commitment to responsible AI development, acknowledging the philosophical and scientific challenges in understanding the inner experiences of AI models.

The Smartest AI Library on the Internet🎉

Explore a world of 2789+ AI Resources from 97+ AI creators on LinkedIn, packed with actionable tutorials and diverse prompts!

Gemini’s Veo2 Challenges ChatGPT’s Sora

Google Expands AI Video Generation with Veo 2 and Whisk Animate

Google has expanded its AI video generation capabilities by integrating the Veo 2 model into the Gemini platform and introducing Whisk Animate, enabling users to create high-quality videos from text prompts or static images.

Key Features:

• Veo 2 Integration: Gemini Advanced and Google One AI Premium subscribers can now generate 8-second, 720p MP4 videos by providing text prompts. Veo 2 understands real-world physics and human motion, resulting in fluid character movements and lifelike visuals. ​

• Whisk Animate: This tool allows users to animate static images into short videos, leveraging Veo 2's capabilities. It's accessible through Google Labs for eligible subscribers. ​

• Content Sharing and Watermarking: Generated videos can be directly shared on platforms like TikTok and YouTube. Each video includes a SynthID watermark to indicate AI generation. ​

These advancements aim to make content creation more accessible, allowing users to produce engaging videos without the need for traditional filming equipment or extensive editing skills.​

AI Highlights of the Week

Meta Cracks Down on Facebook Spam

Accounts using excessive hashtags or unrelated captions will have their reach limited to followers and be blocked from monetization.

Fake engagement is also under fire: coordinated spam comments will be hidden, and users will soon be able to report irrelevant comments.

The goal is simple — more authentic content, better user experience.

Perplexity browser will track everything users do online for 'hyper personalized' ads

Perplexity CEO Aravind Srinivas has announced Comet, a new browser that will track users' online activity beyond the app. The aim: gather detailed data on browsing habits, purchases, and preferences to deliver highly personalized ads.

Srinivas believes users will value the more relevant advertising this approach enables.

Comet is set to launch in May, alongside a deal to have the Perplexity app pre-installed on Motorola’s Razr phones.

Yahoo, OpenAI and others wants to buy Google Chrome

During the DOJ’s antitrust trial against Google, Yahoo said it would consider buying Chrome if forced divestiture happens.

Yahoo’s Search GM highlighted that 60% of search queries come from browsers, making Chrome key to boosting Yahoo’s 3% market share.

OpenAI and Perplexity also showed interest.

OpenAI, struggling with Bing’s search data and slow progress on its own index, sees Chrome as a way to directly reach users and improve search quality.

DuckDuckGo cited financial limits, making a bid unlikely.

Perplexity debuts iOS voice assistant

Perplexity has launched a conversational AI voice assistant for iOS, allowing users to set reminders, write emails, and book rides through third-party apps like Uber. The assistant works inside the Perplexity app and continues running in the background.

It integrates with apps like Apple Music and Mail, but can’t access system functions like setting alarms — still limited to Siri. The iOS launch follows its earlier debut on Android.

OpenAI Launches GPT-Image-1 API for Developers

OpenAI has released GPT-Image-1, a multimodal image generation model now available via API. Previously limited to ChatGPT, it lets developers create and edit high-quality images with diverse styles and accurate text rendering.

It’s being integrated into platforms like Adobe Firefly, Figma, and Wix, streamlining creative workflows. Pricing ranges from $0.02 to $0.19 per image, depending on quality and resolution.

Is China making “Human vs Robots”?

Humanoid Robots Compete Alongside Humans in Beijing Half-Marathon

On April 19, 2025, Beijing hosted a groundbreaking event where 21 humanoid robots ran alongside thousands of human participants in the Yizhuang half-marathon, marking the first instance of such direct competition over a 21-kilometer course.

Event Highlights:

• Diverse Robotic Participants: The robots, developed by Chinese companies like DroidVP and Noetix Robotics, varied in size and design, with some featuring human-like appearances and capabilities such as winking and smiling.

• Human Assistance: Each robot was accompanied by human trainers, some of whom provided physical support during the race. Notably, some robots wore running shoes, and one sported boxing gloves and a headband reading "Bound to Win" in Chinese.

• Performance and Challenges: The winning robot, Tiangong Ultra from the Beijing Innovation Centre of Human Robotics, completed the race in 2 hours and 40 minutes. However, many robots faced difficulties, with some falling or requiring battery replacements during the event.

This event underscores China's commitment to advancing humanoid robotics and integrating them into real-world scenarios, reflecting significant progress in AI and robotics technology.

Last Week’s Poll Result

• Which GPT-4.1 variant aligns best with your requirements?

  GPT-4.1 (full model), 82.35%. GPT-4.1 mini, 11.76%. GPT-4.1 Nano, 5.88%.

  

• Do you believe AI-generated videos can match the quality of human-made content?

  Yes, 80.00%. No, 00.00%. Maybe,20.00%.

• Do you believe humanoid robots will eventually match human athletic performance?

  Yes, 57.14%. Possibly, 28.57%. No, 14.29%.

Feedback

Know someone who may be interested?