π‘ Presentation Tips - Ending a Talk Effectively
This article discusses effective ways to conclude a presentation, including using a designated "The End" slide and revisiting key takeaways to facilitate Q&A.
Key Points:
β’ Use a clear "The End" slide to signal the conclusion of the presentation.
β’ Return to a slide summarizing key takeaways to smoothly transition into Q&A.
β’ This structured ending allows for organized audience applause and questions.
π Resources:
β’ FranΓ§ois Fleuret's Twitter β - Presentation tips
Image
π€ 3D Computer Vision - Scalable Learning in 3D
This article briefly discusses the challenges of balancing inductive biases with scalable learning in 3D computer vision and announces an upcoming workshop.
Key Points:
β’ 3D computer vision goes beyond pixels, incorporating geometry, physics, topology, and functions.
β’ Balancing inductive biases with scalable learning is a key challenge.
β’ The Ind3D workshop at CVPR 2025 will address future directions in 3D models.
π Resources:
β’ Ind3D Workshop β - Workshop on 3D models
Image
π‘ Academic Funding - Beyond the NSF Myth
This article challenges the notion that the NSF is the sole key to academic grant funding, presenting alternative sources and discussing the limitations of large NSF grants.
Key Points:
β’ The NSF is one source of funding, but not the only one.
β’ Alternative sources include industry, DoD, private foundations, and internal university grants.
β’ Large NSF grants often concentrate funding within specific groups.
π Resources:
β’ Dr. Milanfar's Twitter β - Perspective on academic funding
Image
π‘ Academic Funding - Diverse Funding Sources
This article expands on alternative sources of academic funding beyond the NSF, highlighting the potential downsides of large, multi-university grants.
Key Points:
β’ Diversify funding sources to reduce reliance on a single institution.
β’ Explore industry, DoD, private foundations, and internal university grants.
β’ Critically evaluate the structure and impact of large, multi-university grants.
π Large Language Models - DeepSeek's Impact
This article highlights the significance of DeepSeek in the evolution of Large Language Models (LLMs), establishing a new default for non-US labs.
Key Points:
β’ DeepSeek has set a new standard for LLMs outside of US labs.
β’ It represents a significant advancement in the lineage of prominent LLMs.
β’ DeepSeek has influenced the development of subsequent LLMs.
π€ Decentralized AI - Energy Optimization
This article proposes that decentralized AI infrastructure can optimize energy usage by leveraging market forces at the supply chain level, drawing a parallel with Bitcoin mining.
Key Points:
β’ Decentralized AI infrastructure can optimize energy usage.
β’ Market forces at the supply chain level drive optimization.
β’ This mirrors how Bitcoin mining proxies energy markets.
β¨ ICCV 2025 - Review Updates
This article announces the release of ICCV 2025 reviews and provides key deadlines for authors.
Key Points:
β’ ICCV 2025 reviews are being distributed to authors.
β’ Reviews will be available on OpenReview soon.
β’ Authors have until May 16, 2025, to submit rebuttals.
π Resources:
β’ ICCV Conference Twitter β - Official announcements
π‘ ChatGPT Adoption - A Simple Hack
This article offers a practical strategy to improve ChatGPT adoption within an organization by leading by example.
Key Points:
β’ Leadership's effective use of ChatGPT is crucial for team adoption.
β’ Modeling effective usage encourages broader implementation.
β’ Start by using ChatGPT before interacting with employees.
π€ AI System Reliability - Distribution Shifts
This article emphasizes the critical importance of addressing distribution shifts when developing reliable AI systems.
Key Points:
β’ Ignoring distribution shifts is a major pitfall in AI system development.
β’ Training on the target distribution is vital for reliability.
β’ Sim-to-real approaches often fall short due to distribution mismatch.
π Multimodal LLMs - OpenVision
This article announces OpenVision, a family of open-source vision encoders designed to be a cost-effective alternative to OpenAI's CLIP.
Key Points:
β’ OpenVision is a fully open, cost-effective family of vision encoders.
β’ It matches or surpasses the performance of OpenAI's CLIP.
β’ It offers advanced architecture configurations.
π Resources:
β’ OpenVision Announcement β - Details on the new vision encoders
Image
βοΈ Support
If you liked reading this report, please star βοΈ this repository and follow me on Github β, π (previously known as Twitter) β to help others discover these resources and regular updates.
