AIR#85 - May 28, 2024

Good morning, AI aficionados! 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 xAI's astonishing $6B Series B funding round, a financial windfall set to propel the company's AI advancements and product launches to new heights. This massive investment signals a bright future for xAI and underscores the industry's growing confidence in AI-driven innovation.

But it's not all about the big bucks. Google's AI is making waves for all the wrong reasons, delivering absurd answers sourced from the satirical site The Onion. Users are left both amused and bewildered, prompting a serious conversation about the reliability of AI-generated content. Meanwhile, in a lighter yet equally revealing story, Janelle Shane's latest blog post highlights an AI's struggle to identify a rare spotless giraffe, showcasing the limitations of current image recognition technology when faced with the unusual or unexpected.

As we explore these stories, let's not forget the broader implications they bring to the table. From BP's significant reduction in third-party coder reliance thanks to AI, to the EU's preliminary stance on ChatGPT's GDPR compliance, 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

🔥 xAI Secures $6B Series B Funding Round
xAI secures $6B Series B funding to accelerate AI advancements, launch products, and expand their team. Join them at x.ai/careers.

BP: AI Reduces Need for Third-Party Coders by 70%
BP reduces reliance on third-party coders by 70% as AI takes over coding tasks.

Google's AI Gives Answers from The Onion
Google's new AI Overview is delivering absurd answers from satirical site The Onion, confusing users with ridiculous advice.

EU Privacy Regulators Release Preliminary Position on ChatGPT Compliance
EU's taskforce reveals ChatGPT's GDPR compliance is still in question, posing significant regulatory risks for OpenAI in Europe.

17 Cringe-Worthy Google AI Answers Highlight Major Flaws | Tom's Hardware
Google's AI overviews are plagued with misinformation, highlighting the dangers of training on unreliable internet sources.

I’m the First Person to Receive a Neuralink Brain Implant—Here’s My Story
First person with a Neuralink implant shares how it helped regain independence and reconnect socially after a spinal-cord injury.

The Breathtaking Scope of Sam Altman's Future AI Empire
Sam Altman is building an AI empire with investments in energy, chips, and life extension, aiming for a future with AGI.

Engineering

Red Teaming LLM Applications on Coursera by DeepLearning.AI
Learn to identify and fix vulnerabilities in LLM applications with DeepLearning.AI's beginner-friendly course on Coursera.

[GitHub] ChatGPT Firewall: Secure, Cost-Effective Document Chatting
Securely chat with large documents using ChatGPT Firewall—cost-effective, privacy-focused, and supports multiple languages. Try it now!

[GitHub] Language Model Alignment Deep Learning Curriculum by Jacob Hilton
Jacob Hilton's GitHub offers a deep learning curriculum focused on language model alignment, ideal for those with a strong quantitative background.

[Blog] RNN for 2D Tile Map Synthesis: John Lin's Experiment
John Lin used RNNs to create 2D tile maps for games, showing neural networks can generate new, complex maps with limited data.

[Paper] Neuromorphic Dreaming: Efficient Learning in AI Agents
Neuromorphic dreaming enables AI to learn efficiently using spiking neural networks, reducing real-world data needs and boosting energy efficiency.

[GitHub] Promptic: LLM Superpowers for Python Functions in ~80 Lines
New Python library "promptic" simplifies using LLMs with just 80 lines of code. Easy prompts, structured outputs, and real-time responses!

Academic

🔥 [Paper] Grokked Transformers: Implicit Reasoners at the Edge of Generalization
Grokked Transformers excel in implicit reasoning through extended training, outperforming GPT-4-Turbo and Gemini-1.5-Pro in complex tasks.

AI vs. Spotless Giraffe: A Cautionary Tale by Janelle Shane
AI image recognition struggles with identifying a rare spotless giraffe, revealing limitations in handling unusual or new visuals.

[Paper] Grokked Transformers Are Implicit Reasoners
Transformers can implicitly reason after extensive training, excelling in comparison tasks but struggling with composition.

Managing Extreme AI Risks Amid Rapid Progress
AI's rapid progress brings immense benefits but also grave risks. Urgent action needed to ensure safety and prevent loss of human control.

[Paper] Self-Proving Models for Machine Learning
New paper introduces Self-Proving models that verify their own correctness, ensuring reliable outputs for machine learning tasks.

Can AI Cure Human Loneliness?
AI could combat loneliness by enhancing social skills and offering companionship, but experts caution about its impact on human relationships.

[Paper] Safeguards to Prevent LLMs from Spreading Health Disinformation
Study finds inconsistent safeguards in LLMs against health disinformation; calls for enhanced regulation and transparency for AI developers.

[Paper] Extracting Prompts by Inverting LLM Outputs
New method "output2prompt" extracts prompts from LLM outputs without access to model internals, enabling efficient zero-shot transfer.

[Paper] Kolmogorov-Arnold Networks (KANs) for Time Series Analysis
KANs outperform MLPs in satellite traffic forecasting with fewer parameters, leveraging adaptive activation for better predictive modeling.

[Paper] Principled Instructions Are All You Need for Questioning LLaMA-1/2, GPT-3.5/4
26 principles simplify querying LLaMA-1/2, GPT-3.5/4, enhancing user comprehension and prompt design. Study confirms effectiveness.