AIR#84 - May 27, 2024

Good morning, AI enthusiasts! As you sip your morning coffee, get ready to dive into today's edition of AIR: The AI Recon, where the world of artificial intelligence is buzzing with stories that are as fascinating as they are thought-provoking. Leading the headlines is a powerful call from ex-OpenAI board members who argue that AI firms can't self-govern due to profit pressures and urge for government regulation. This plea underscores the critical need for oversight in an industry that's rapidly shaping our future.

But it's not all about regulatory debates. Dive into the human side of AI with a gripping story of a woman misidentified as a shoplifter by facial recognition technology. This incident highlights the pressing concerns over the accuracy and civil liberties implications of AI in everyday life. Meanwhile, Google is making waves by planning to include ads in its AI-generated search overviews, a move that could significantly impact the landscape of online advertising and user experience.

As we explore these stories, let's not forget the broader implications they bring to the table. From SoftBank's ambitious $9B annual AI investment plans aiming to reshape its business, to the ongoing race among scientists to open-source AlphaFold3, today's edition is packed with insights that challenge our understanding of AI's role in society. So, sit back, sip your coffee, and let's delve into the dynamic world of artificial intelligence together.

Business

🔥 I Was Misidentified as a Shoplifter by Facial Recognition Tech
Facial recognition tech misidentified Sara as a shoplifter, highlighting concerns over accuracy and civil liberties.

Google Search's AI Overviews Will Soon Include Ads
Google Search will soon include ads in its AI-generated overviews, extending its dominance in search advertising.

SoftBank Aims for $9B Annual AI Investments, Seeks Mega Deals
SoftBank commits $9B annually to AI investments, aiming for mega deals to reshape its business and support Arm's growth.

Tim Cook's AI Dilemma: OpenAI or Google?
Tim Cook must choose between OpenAI and Google to bring AI to iPhones, each with its own set of challenges.

Everyone Gets an AI Agent
In 2024, everyone will have personal AI agents to filter, analyze, and recommend content, transforming journalism and how we consume information.

Elon Musk Plans xAI Supercomputer to Rival OpenAI and Google
Elon Musk's xAI plans a supercomputer with 100,000 Nvidia GPUs to rival OpenAI and Google by 2025.

Engineering

🔥 [GitHub] DiscoGrad: Automatic Differentiation in C++ Across Conditional Branches
DiscoGrad enables automatic differentiation across conditional branches in C++ programs, enhancing gradient calculations for optimization.

🔥 [GitHub] Self-Hosted Offline Transcription & Diarization with LLM Summary
Self-hosted offline transcription & diarization with LLM summaries now available on GitHub. Easy setup, web interface, and full-text search.

[GitHub] GoalChain: Framework for Goal-Oriented LLM Conversations
GoalChain: A framework for goal-oriented LLM conversations, streamlining human-LLM and LLM-LLM interactions. Now on GitHub! 🌐

Convert Text or Images to UI with Prototyper's React Playground
Convert text or images to UI in minutes with Prototyper's React Playground. Describe your idea, iterate, and publish seamlessly.

[GitHub] ReproModel: Build ML Models with a Form, Boost AI Research Efficiency
ReproModel on GitHub boosts AI research by simplifying model reproduction and comparison, cutting dev time by 40%. Check it out!

[GitHub] YOLOv10: Real-Time End-to-End Object Detection by THU-MIG
YOLOv10 by THU-MIG offers real-time, end-to-end object detection with improved performance and efficiency. Check it out on GitHub!

[GitHub] Microsoft Olive: Hardware-Aware Model Optimization Tool
Microsoft's Olive simplifies model optimization for cloud and edge by automating hardware-specific techniques for efficiency and performance.

[GitHub] Alexplainable: Transparent ImageNet Recognizer Prototype by joy-void-joy
Joy-void-joy's Alexplainable aims to create a transparent ImageNet recognizer, offering insights into neural network operations via explainable AI.

[Paper] LiteVAE: Efficient Autoencoders for Latent Diffusion Models
LiteVAE: A new family of efficient autoencoders boosts latent diffusion models' performance with fewer parameters and faster training.

[PyPI] RAGoon: Boost LLM Retrieval with Dynamic Web-Search via Groq Chips
RAGoon boosts LLM retrieval using dynamic web-search and Groq chips for faster, contextually relevant query generation. 🚀

[Paper] Attention as an RNN by Yoshua Bengio and Others
New paper shows attention as a special RNN, introduces Aaren, an efficient model combining Transformer-like training with RNN-like updates.

Academic

🔥 AI Firms Mustn’t Govern Themselves, Warn Ex-OpenAI Board Members
Ex-OpenAI board members urge government regulation, arguing AI firms can't self-govern due to profit pressures.

[Paper] AI and the Skill Premium: NBER Study
AI reduces the skill premium by substituting high-skill tasks, impacting labor economics and wage disparities, says NBER study.

Who Will Open-Source AlphaFold3? Scientists Race to Crack the AI Model
Scientists race to open-source AlphaFold3 after DeepMind's initial code withholding sparks global efforts to replicate the AI model.

[Paper] How Close Are We to AGI?
New paper explores our progress towards AGI, highlighting current challenges, necessary technologies, and a roadmap for future development.

Another OpenAI Researcher Quits, Issues Ominous Warning
Another OpenAI researcher quits, warning of opaque decision-making and lack of accountability in the AI industry.

Openness on OpenAI: Scott Aaronson Reflects on Resignations and AI Safety
Scott Aaronson reflects on the resignations of key figures at OpenAI, emphasizing continued dedication to AI safety despite organizational changes.

[Paper] Computational Dualism and Objective Superintelligence
Intelligence relies on hardware, not just software. This limits AGI's potential, making it safer but less powerful than expected.

[Paper] DeepSeek-Prover: Boosting Theorem Proving in LLMs with Synthetic Data
DeepSeek-Prover uses synthetic data to boost LLMs in theorem proving, outperforming GPT-4 and setting new benchmarks in mathematical proofs.