🤖 AI Agents - BBS Door Automation

This article introduces 'spree', a project enabling LLMs to interact with BBS door games. It covers the setup for automating gameplay within a SynchronetBBS environment using Docker and vllm.

Key Points:

• Enables LLMs to play classic BBS door games.

• Leverages Docker for easy deployment of SynchronetBBS.

• Integrates vllm for efficient LLM inference.

• Automates gameplay for vintage text-based adventures.

🚀 Implementation:

Spin up SynchronetBBS: Utilize Docker to set up the BBS environment.
Launch vllm instance: Deploy a vllm instance for the LLM.
Configure LLM for gameplay: Connect the LLM to play TW2 Door.

🔗 Resources:

• wightmanr ↗ - Project author on X

• Original Tweet ↗ - Spree project announcement

🤖 AI Agents - Zork Gameplay Automation

This article demonstrates an AI agent successfully playing the interactive fiction game Zork. It highlights the agent's progress through a complex text-based environment, tracing its actions and state.

Key Points:

• Showcases AI agent's capability in complex text adventures.

• Successfully navigates game environments like caves and dams.

• Achieves objectives such as defeating in-game trolls.

• Maintains a trace record of gameplay progression.

🔗 Resources:

• wightmanr ↗ - Project developer

• Original Tweet ↗ - Zork gameplay progress

🤖 3D Reconstruction - Pixel-Aligned Gaussian Splatting

This article introduces PAGaS, a method that refines depth maps for multi-view stereo reconstruction. It details how Gaussian Splatting is constrained along viewing rays to achieve highly detailed results.

Key Points:

• Refines highly detailed depth maps.

• Utilizes pixel-aligned 1DoF Gaussian Splatting.

• Enhances multi-view stereo reconstruction.

• Constrains splatting along viewing rays for precision.

🔗 Resources:

• Almorgand ↗ - Author of the paper on X

• Original Tweet ↗ - PAGaS announcement

🤖 AI Agents - Direct Corpus Interaction for Search

This article explores Direct Corpus Interaction (DCI) for agentic search, which rethinks traditional retrieval methods. It focuses on enabling agents to interact directly with raw data using CLI tools rather than relying on pre-compressed snippets.

Key Points:

• Rethinks retrieval for agentic search.

• Enables direct agent interaction with raw corpora.

• Utilizes CLI tools for robust data access.

• Offers a powerful direction for tool-using search agents.

🔗 Resources:

• lupantech ↗ - Discusses DCI for agentic search

• zhuofengli96475 ↗ - Co-author mentioned in the thread

• Original Tweet ↗ - DCI approach for agentic search

Image

🤖 Biomedical AI - Microbiome Agent Co-pilots

This article announces a collaborative project focused on developing microbiome-specialized AI agent co-pilots. The initiative aims to accelerate biomedical discovery with support from Stanford institutions and leading researchers.

Key Points:

• Develops microbiome-specialized AI agent co-pilots.

• Aims to accelerate biomedical discovery research.

• Supported by StanfordHAI and Stanford Center for Digital Health.

• Collaboration with leading researchers.

🔗 Resources:

• lupantech ↗ - Project contributor on X

• YifanGao15 ↗ - Project collaborator on X

• LabSonnenburg ↗ - Collaborating lab on X

• james_y_zou ↗ - Project collaborator on X

• StanfordHAI ↗ - Supporting institution

• Original Tweet ↗ - Project announcement

🤖 Robotics - Zurich Robotics Event

This article details the upcoming ZurichRobotics #4 event hosted at the ETH AI Center. It will feature discussions on Conway's Law in Robotics and the design principles for robots with varied locomotion and manipulation capabilities.

Key Points:

• Features discussions on Conway's Law in Robotics.

• Covers design principles for advanced robotic systems.

• Hosts speakers from leading robotics organizations.

• Provides networking opportunities at ETH AI Center.

🔗 Resources:

• NandoMetzger ↗ - Event organizer/announcer

• zurichnlp ↗ - Announcing organization on X

• ETH_AI_Center ↗ - Event host

• vateseif ↗ - Speaker on X

• watneyrobotics ↗ - Speaker's company

• RSVP Link ↗ - Event registration

• Original Tweet ↗ - Event announcement details

💡 AI Ethics - Human-in-the-Loop AI Decisions

This article discusses the contrasting applications of AI in practical and judicial contexts. It highlights how the effectiveness and acceptance of AI tools depend significantly on human involvement and oversight in decision-making processes.

Key Points:

• AI effectively assists in production code bug fixes.

• Government use of AI for grant decisions faced judicial block.

• Same AI technology yields different outcomes based on application.

• Emphasizes the critical role of human-in-the-loop for AI adoption.

🔗 Resources:

• xiz25 ↗ - Discussing AI application insights

• Original Tweet ↗ - AI ethics discussion

💡 AI Strategy - Future Interaction Paradigms

This article examines the significant investments by major tech companies into AI infrastructure, contrasting this with the current user interface limitations. It suggests that the future of AI innovation lies in developing more intuitive interaction methods beyond simple text prompts.

Key Points:

• Big Tech invests heavily in AI infrastructure.

• Current AI consumer interface remains largely text-based.

• Future AI innovation requires new interaction paradigms.

• Focus shifts from larger models to better user interaction.

🔗 Resources:

• xiz25 ↗ - Discussing AI market trends

• Original Tweet ↗ - AI investment and future interfaces

✨ Work Environment - Personal Office View

This article provides a glimpse into a professional's afternoon office view. It showcases the environment where academic or technical work is conducted, offering a visual context for a work setting.

Key Points:

• Showcases a typical professional work setting.

• Provides visual context for a work environment.

• Highlights the surroundings of a technical professional.

🔗 Resources:

• CSProfKGD ↗ - Sharing personal work environment

• Original Tweet ↗ - Office view

Image

🤖 LLM Alignment - Reinforcement Learning Methods

This article explains the significance of reinforcement learning algorithms like GRPO, DPO, PPO, and RLHF in aligning large language models. It emphasizes that implementing these algorithms is crucial for understanding how base models evolve into sophisticated AI systems.

Key Points:

• Highlights key algorithms for LLM alignment.

• Explains the transformation of base models into systems like ChatGPT.

• Emphasizes practical implementation for deep understanding.

• Introduces a comprehensive RL track for learning.

🚀 Implementation:

Study GRPO: Understand the Generalized Policy Optimization algorithm.
Learn DPO: Explore Direct Preference Optimization techniques.
Implement PPO: Apply Proximal Policy Optimization principles.
Practice RLHF: Develop Reinforcement Learning from Human Feedback.

🔗 Resources:

• kunaldargan ↗ - Discussing LLM alignment

• TensorTonic ↗ - Platform offering RL track

• Original Tweet ↗ - RL track announcement

Image

⭐️ Support

If you liked reading this report, please star ⭐️ this repository and follow me on Github ↗, 𝕏 (previously known as Twitter) ↗ to help others discover these resources and regular updates.