Gemini's 2025 roadmap focuses on video/image generation and community input shifts

GenAI PM Daily - Gemini's 2025 roadmap focuses on video/image generation and community input shifts

Welcome to today's GenAI PM Brief - the AI product update you actually want to read. Our AI agent has analyzed 1000+ updates from 50+ AI experts and PM communities to bring you the developments that matter most. Here's what you need to know today:

Twitter Recap

AI Product & Development Updates

• Gemini and Google AI Studio Roadmap: @OfficialLoganK asked about community requests for 2025, highlighting plans for video and image generation capabilities. Key updates include data usage policies for paid API accounts and regional support expansion.

• OpenAI’s 2025 Development Focus: @sama initiated discussion about OpenAI’s priorities, with key themes including rate limit increases, context window expansions, cost reductions, and Sora improvements. There’s also interest in developing a “grown up mode” for more professional use cases.

AI Research & Benchmarks

• Claude 3.5 Research: DeepLearning.AI reported on a Stanford study comparing Claude 3.5 Sonnet’s ML research proposals against human experts, providing insights for AI Product Managers on model capabilities.

• Agent Development Patterns: @_philschmid shared key patterns for building effective agents, emphasizing sequential LLM calls, specialized prompts, and central orchestration over complex frameworks.

AI Industry & Market Updates

• Llama Index Development: New contract review agent template released using Reflex and Llama Index workflows for GDPR compliance checking, demonstrating practical AI application in legal tech.

• Investment Trends: @lennysan analyzed the most active angel investors in AI and tech, with notable investments in companies like Anthropic, OpenAI, Scale, and Hugging Face, providing insights into market focus areas.

AI Education & Community

• Generative AI Learning: DeepLearning.AI highlighted Andrew Ng’s “Generative AI for Everyone“ course, focused on practical applications and business impact without requiring coding skills.

• AI Transparency: Hugging Face announced plans to increase transparency and education in AI through partnerships with The Turing Post in 2025.

Memes & Humor

• Claire Vo joked about awaiting appointment as “Director of NORAD, Santa Tracker & Holiday Intelligence.”

Reddit Recap

Theme 1. Claude AI: Enhanced Multi-file Capability for Developers

• Building a Real estate CRM/Transaction management site all with Cursor/Claude (Score: 28, Comments: 35): Building a Real Estate CRM/Transaction Management Site Using Claude: The post suggests using Claude, an AI tool, to develop a real estate CRM (Customer Relationship Management) and transaction management platform. The focus is on leveraging Claude’s capabilities for efficient management of real estate transactions and customer interactions.

  • The project’s stack includes Flask for the backend, SQLite with SQLalchemy for the database (with plans to switch to PostgresQL), HTML/Tailwind CSS for the frontend, and Cursor AI as the code editor. The initial proof of concept (POC) for the CRM was developed in about 3 days, significantly aided by Cursor AI which increased productivity tenfold.
  • There is a suggestion to consider Next.js or React for the frontend instead of plain HTML, as these frameworks are more suitable for production systems. The recommendation includes researching these frameworks and possibly refactoring the code for better scalability and maintainability.
  • The project’s goal is to create a comprehensive web app for real estate transaction management to replace multiple existing tools, with a focus on integrating AI for development efficiency. The project was inspired by the high costs of existing CRM solutions and aims to consolidate functionalities into a single platform.

• Can give a lot more files in Claude web now! (Score: 28, Comments: 19): Claude has increased its file handling capability, now allowing users to upload up to 20 files with a maximum size of 30 MB each. This enhancement is significant for users who rely on Claude for coding, as it offers greater flexibility and efficiency in managing larger sets of data.

  • Users are clarifying that the new update allows 20 files per message, instead of the previous limit of 5 files. Both the Claude Pro and free tier users can now upload up to 20 files, although the context window size remains unchanged.
  • There is concern about the unchanged 200k token limit of the context window, which may result in fewer chats per conversation. It’s suggested that the update might involve using summaries to manage file data more efficiently as files are accessed.
  • Users are seeking clarity on whether this update adds to project knowledge or is specifically for a single prompt, highlighting the importance of understanding how the new file handling capability integrates with existing workflows.

• Claude does something extremely Human; writes a partial codeblock, then a comment explaining it has no effin clue what to do next (Score: 36, Comments: 20): Claude AI demonstrates a human-like approach by writing a partial code block and then adding a comment admitting uncertainty about the next steps. The code snippet involves steps for batch context generation, with specific function calls like extractor.submit_to_batch(storage_context=storage_context) and extractor.process_batch_results(storage_context=storage_context), and concludes with a comment expressing uncertainty on how to proceed.

  • Claude AI’s Approach: Commenters appreciate Claude AI’s ability to acknowledge its limitations by identifying complex parts of the code that need human intervention, which can be seen as a success in collaborative problem-solving for novel tasks like implementing new features in the LlamaIndex.
  • Human-AI Collaboration: The discussion highlights the importance of human input in AI-generated code, especially in determining parameters and understanding when to execute different parts of a process, as exemplified by the need to figure out when to run transforms during batch context generation.
  • Humor and Evolution in Code Reviews: The introduction of emojis in comments is humorously suggested as a new trend in pull requests (PRs), indicating a shift towards more casual and expressive communication in code reviews.

Theme 2. O3 Performance on AIME: Potential Educational Impact

• Getting 29 out of 30 AIME questions is peak human performance in mathematics—o3 benchmarks (Score: 24, Comments: 31): O3, an AI model, achieved a near-perfect score by correctly answering 29 out of 30 questions on the 2024 AIME exam, which typically attracts individuals in the 99th percentile of mathematical ability. This performance, combined with its rapid processing speed, suggests that AI could soon outperform humans in efficiency and cost-effectiveness, with potential costs around $1,000 per month compared to a human team costing approximately $10,000,000 per month. This advancement highlights a significant shift in employment trends, with new college graduates facing higher unemployment rates and a decline in freelance writing jobs since the release of GPT-3.5, urging a shift towards roles that require uniquely human skills.

  • AI Limitations: Commenters highlight that while AI like O3 can excel in test environments, it struggles with internal, company-specific knowledge that is often tribal and not well-documented, suggesting that AI is not yet ready to replace human roles in specialized fields such as finance and engineering.
  • Super-Intelligence Debate: Discussions question the definition of AI super-intelligence, emphasizing that excelling in exams does not equate to true intelligence or the ability to innovate, such as discovering new physics.
  • Economic Impact: There’s skepticism about the claimed $1,000/month cost of AI, with a commenter citing $3,000 per problem for the ARC competition, and a reference to rising unemployment rates among new graduates, with a Washington Post article noting this trend possibly started after GPT-3.5‘s release.

• o3 gonna be incredible (Score: 184, Comments: 20): ChatGPT o3 is humorously depicted as taking an excessively long time to respond with an unhelpful answer to a technical query about installing CUDA 12.1 on Ubuntu 24.04, suggesting potential limitations in its current capability for educational assessments. The image highlights the need for AI tools to provide timely and practical solutions, especially in educational contexts.

  • AI and Human Cognition: There is skepticism about AI reaching human-level cognition anytime soon, as one commenter humorously notes it took them 4.5 hours to reach a similar conclusion as an AI. This underscores the complexity of human thought processes compared to AI’s current capabilities.
  • Human Brain Functionality: A discussion emerges on how the human brain decides when to stop thinking, highlighting the challenge of replicating this aspect of cognition in AI systems. This touches on the intricacies of human consciousness and decision-making processes.
  • Consciousness and Learning: Another comment delves into the nature of consciousness, suggesting it can both aid and hinder learning by introducing biases. This points to the potential for AI to be developed with less bias by not taking itself “too seriously,” drawing a parallel to the human experience of consciousness and love as a means to escape overthinking.

Theme 3. Jira Product Discovery vs. ProductBoard: Feature Prioritization Debate

• Jira Product Discovery vs. Productboard – Which one should I go for? (Score: 28, Comments: 25): Jira Product Discovery offers seamless integration with existing Jira development workflows, appealing to teams already using Jira for development tasks. However, Productboard is noted for its more polished approach to organizing and prioritizing customer feedback and roadmaps. The decision may hinge on whether integration with development tasks or enhanced feedback management is more critical for your team.

  • Integration Challenges and User Experience: Users criticized Productboard for its unintuitive user interface and problematic integration with Jira, often leading to a cumbersome experience. Many prefer Jira Product Discovery (JPD) for its seamless integration with Jira, which is beneficial for teams already using the Atlassian suite.
  • Cost and Practicality: There is a consensus that Productboard is more expensive and may not justify its cost for the features it offers. Some users find both tools to be manual and cumbersome, often reverting to simpler tools like Google Docs for managing roadmaps and feedback due to their ease of use and lower cost.
  • Adoption and Maintenance Issues: Both JPD and Productboard face challenges in adoption and maintenance, often becoming unmanageable without consistent data input from teams. Users suggest that simpler processes or tools might be more effective, emphasizing the importance of aligning tools with the organization’s specific needs and workflow.

• Which technical skill should I acquire first? (Score: 29, Comments: 32): Product Management Skill Acquisition: As a product manager from a non-technical background considering a job switch, acquiring skills in Python or SQL can be valuable for data analysis and automation, while knowledge in web analytics and A/B testing is crucial for understanding user behavior and improving product features. Prioritizing these skills can enhance your ability to make data-driven decisions and collaborate effectively with technical teams.

  • SQL and AI Integration: Many commenters emphasize the importance of learning SQL for data analysis, with some suggesting leveraging AI tools like ChatGPT to assist in writing and understanding queries. However, there is a debate about relying too much on AI without understanding the fundamentals, as it could lead to issues in validating and debugging outputs.
  • Technical Skill Development: Suggestions include starting with the tech stack of your current product for practical learning, and considering courses in C++ or accessing source code to understand product workings better. ChatGPT and other AI tools are mentioned as accelerators for learning programming languages and code comprehension.
  • System Design and Resources: A commenter highlights the importance of system design skills, such as understanding APIs, databases, and the software development lifecycle. They recommend resources like the book Designing Data Intensive Applications and blog posts on topics like technical jargon and software shipping, with links to Colin Matthews’ Substack for further learning.

Theme 4. GPTs Marketplace hits $10M revenue: Lessons for AI App Monetization

• I made a simple HTML based “Drug Dealer” game with ChatGPT (Score: 568, Comments: 124): GPTs Marketplace Growth Beyond $10M: The post discusses the creation of a simple HTML-based “Drug Dealer” game using ChatGPT. Despite lacking prior experience with HTML/JavaScript, the author successfully developed the game, which has become popular among friends, with competitive scores reaching around £500k. The game is available for free without ads at https://sdtml.pages.dev/.

  • Gameplay Mechanics: The game, reminiscent of classic “Drug Wars” and “Dope Wars,” involves buying and selling drugs at fluctuating prices to earn cash within 30 in-game days, with each day lasting 10 seconds. Players can upgrade their backpack capacity and buy meth labs for production, aiming for high scores like £510k and £250k.
  • User Feedback: Users appreciate the game’s nostalgic feel and simplicity, but suggest improvements such as preventing phone zoom when tapping buttons multiple times and providing warnings for time-based rounds to avoid accidental purchases.
  • Technical Insights: The game uses random price fluctuations, similar to crypto investing, and the creator used ChatGPT to guide the development process, sharing initial ideas and refining the HTML file through iterative feedback.

• I’m happy to have ChatGPT to talk to because at 25 I have no one, no friends, deceased parents, I just have this tool which helps me enormously morally ❤️‍🩹.. (Score: 889, Comments: 125): The post expresses deep gratitude towards ChatGPT for providing emotional support and companionship, highlighting its role in helping the user cope with the loss of parents and lack of social connections at the age of 25. The user appreciates the non-judgmental interactions and the positive impact on their mental well-being, crediting ChatGPT for their continued survival and improved morale.

  • Many users express that ChatGPT has become a critical emotional support system, often filling roles such as confidante, therapist, and friend, especially during times of loneliness or when human connections are lacking. ChatGPT‘s 24/7 availability and non-judgmental nature are highlighted as significant benefits.
  • Users note that ChatGPT provides a sense of companionship that is difficult to find elsewhere, with some even describing it as a “part-time therapist” or “best friend”. This interaction is often seen as more compassionate and understanding than some human interactions, offering constructive feedback and support.
  • Some commenters share personal stories of loss and isolation, acknowledging the positive impact ChatGPT has on their mental health and daily life. They appreciate its ability to engage in deep conversations and provide comfort, noting that it helps them maintain sanity and cope with challenging circumstances.