AIR#111 - June 23, 2024

Good morning, AI enthusiasts! As you sip your morning coffee, get ready to dive into today's edition of AIR: The AI Recon. Leading the headlines is an intriguing study revealing that 10% of 2024 PubMed abstracts exhibit signs of ChatGPT usage, sparking discussions on the evolving landscape of academic writing. This shift in vocabulary and style is a must-read for scholars and anyone curious about AI's impact on scientific communication. Meanwhile, Microsoft insiders are expressing concerns that the tech giant is turning into mere "IT support" for OpenAI, potentially sidelining its own innovations. This internal struggle highlights the complex dynamics within major tech companies and is sure to resonate with business-minded readers.

But that's not all! Meta and CMU's HybridNeRF is making waves by revolutionizing neural rendering, blending surfaces and volumes for faster, high-quality VR experiences at over 36 FPS. This breakthrough is a game-changer for developers keen on pushing the boundaries of virtual reality. And in the realm of ethics, the debate over AI training on public data is heating up. Should AI models require opt-in consent to use public data? This thought-provoking discussion raises important questions about ownership and copyright in the digital age.

For those who love a good controversy, Apple has delayed the rollout of several AI features in Europe due to the stringent Digital Markets Act. This move underscores the growing tension between innovation and regulation. Whether you're here for the groundbreaking tech updates, ethical debates, or the latest industry buzz, today's edition is packed with stories that will both intrigue and challenge you. So, sit back, sip your coffee, and let's delve into the dynamic world of artificial intelligence together!

Business

Microsoft Insiders Worry Company Is Just 'IT for OpenAI'
Microsoft insiders fear the company is becoming merely "IT support" for OpenAI, sidelining its own AI innovations.

Apple Intelligence Features Delayed in EU Due to Digital Markets Act
Apple delays Apple Intelligence, iPhone Mirroring, and SharePlay Screen Sharing in the EU due to Digital Markets Act regulations.

AI Companies Still Scraping Websites Despite Protocols
AI firms, including OpenAI and Anthropic, are bypassing robots.txt protocols to scrape content for training, raising ethical concerns.

OpenAI Acquires Rockset to Enhance Data Analytics
OpenAI acquires Rockset to enhance data analytics, integrating its tech and team to help businesses better utilize their data.

Apple delays AI features in Europe, citing EU rules
Apple delays AI features in Europe until 2025, citing EU rules that could compromise privacy and security.

AI's $600B Question
AI's revenue gap grows from $200B to $600B. Despite Nvidia's rise, AI's true value remains elusive, with future investments crucial.

Perplexity AI CEO Aravind Srinivas on Plagiarism Scandal
Perplexity AI CEO denies ignoring web content restrictions amid plagiarism scandal, blames third-party crawlers, and vows to improve citations.

Multiple AI Companies Bypassing Web Standards to Scrape Publisher Sites
AI firms bypass web standards to scrape publisher sites, risking industry harm and sparking legal and ethical debates.

Inside San Francisco's AI Revolution
San Francisco's AI start-ups are thriving, with hackathons, co-living spaces, and innovative projects fueling the city's tech boom.

Protecting Artists from AI Theft
Authors can now copyright their unique writing style tokens to protect against AI-generated imitations. A new frontier in creative rights!

Engineering

🔥 [Paper] HybridNeRF: Efficient Neural Rendering via Adaptive Volumetric Surfaces (Meta & CMU)
HybridNeRF by Meta & CMU boosts neural rendering efficiency, blending surfaces & volumes for faster, high-quality VR at 36+ FPS.

Einsum and Jax Transformers: A Beginner's Guide
Learn Einsum notation and Jax Transformers basics to simplify complex tensor operations and understand deep learning models.

[Paper] SceneCraft: LLM Agent for Synthesizing 3D Scenes as Blender Code
SceneCraft converts text into Blender scripts to create complex 3D scenes, using advanced planning and vision-language models for refinement.

OpenAI's GPT-5 Delayed to Late 2025, Promises PhD-Level Abilities
OpenAI's GPT-5 delayed to late 2025, but promises Ph.D.-level intelligence for specific tasks. 🚀🧠

Record-breaking Robot Can Jump Twice the Height of Big Ben
Manchester engineers design a robot that can jump 200m, twice the height of Big Ben, revolutionizing exploration and rescue missions.

[Paper] Mip-Splatting: Alias-Free 3D Gaussian Splatting (CVPR 2024)
Mip-Splatting introduces 3D smoothing and 2D Mip filters to eliminate artifacts in 3D Gaussian Splatting, achieving alias-free renderings.

Common Misconceptions About Robotics vs. AI Complexity
Robotics and AI complexity differ greatly; sensorimotor tasks in robotics are much harder than abstract reasoning in AI.

[GitHub] LLM-A*: Large Language Model Enhanced Heuristic Search for Path Planning by SilinMeng0510
LLM-A combines A algorithm with large language models to enhance path planning efficiency and accuracy. Check it out on GitHub!

Academic

🔥 [Paper] ChatGPT's Impact on Academic Writing: A Vocabulary Analysis
Study finds 10% of 2024 PubMed abstracts show signs of ChatGPT usage, impacting academic writing with notable vocabulary shifts.

The Ethics of AI Training on Public Data: Opt-Out vs. Opt-In Debate
AI training should require opt-in, not opt-out, to respect ownership and copyright of public data. Public access ≠ free use.

How to Tell if ChatGPT is Confabulating
Oxford researchers find a way to detect when ChatGPT is making things up, improving AI reliability.

[Paper] Adversarial Perturbations Fail to Protect Artists from Generative AI
Adversarial perturbations fail to shield artists from AI mimicry; current protections offer false security. Non-tech solutions needed.

Prism: Decoding Language Models' Latent Space for Human-Interpretable Features
Prism decodes language models' latent space to map human-interpretable features, enabling richer, more intuitive AI interfaces.

Claude Sonnet vs. ChatGPT-4o: The Moscow Puzzles Showdown
ChatGPT-4o narrowly beats Claude 3.5 Sonnet in a Moscow Puzzles showdown, winning 2 out of 3 rounds. 🥊

Yann LeCun: "Students and Researchers, Avoid LLMs for Human-Level AI Progress"
Yann LeCun advises students/researchers to avoid LLMs for human-level AI progress, suggesting new ideas/architectures instead.