Week 3 of 2024- W032024
Look for signs
Disclaimer: The views and opinions expressed in this blog are entirely my own and do not necessarily reflect the views of my current or any previous employer. This blog may also contain links to other websites or resources. I am not responsible for the content on those external sites or any changes that may occur after the publication of my posts.
End Disclaimer
“See, what you have to ask yourself is what kind of person are you? Are you the kind that sees signs, sees miracles? Or do you believe that people just get lucky?”
-M. Night Shyamalan
“Three is a magic number. Yes it is.” -Schoolhouse Rock
News
AIML
Mark Zuckerberg’s new goal is creating artificial general intelligence
The Lazy Tyranny of the Wait Calculation- Taking AI timelines seriously
CEOs say generative AI will result in job cuts in 2024
Chip Huyen- Sampling for Text Generation
OpenAI Quietly Deletes Ban on Using ChatGPT for “Military and Warfare”
SAG-AFTRA Signs Deal With Voiceover Studio for AI Use in Video Games
“This is an evolutionary step forward. AI technology is not something we can block. It’s not something we can stop.”
AI to hit 40% of jobs and worsen inequality, IMF report says
Bill Gates explains how AI will change our lives in 5 years
Intro to optimization in deep learning: Gradient Descent
Painting
The Ambassadors, Hans Holbein the Younger, (1497/8 - 1543), The National Gallery
There is an anamorphic skull in the foreground at the bottom center of the painting, a memento mori.
“The skull, rendered in anamorphic perspective, another invention of the Early Renaissance, is meant to be a visual puzzle as the viewer must approach the painting from high on the right side, or low on the left side, to see the form as an accurate rendering of a human skull.” (from the Wikipedia entry)
A closer look at Hans Holbein’s “The Ambassadors”
Crypto
How a 27-Year-Old Codebreaker Busted the Myth of Bitcoin’s Anonymity
‘This Is Just The Beginning’ —BlackRock CEO Reveals Massive Crypto Plan
Cyber
Attack of the week: Airdrop tracing
Markets
How to Guarantee the Safety of Autonomous Vehicles
A $126 grocery tab that explains the vibes paradox
FRBSF Economic Letter- Extreme Weather and Financial Market Uncertainty
It's Time to Nationalize and Then Break Up Boeing
U.S. tech companies prepare for potential drone attacks as international strikes spark concern
What doom loop? With AI, a 'spirit of optimism' returns to San Francisco start-ups
Misc
How Much of the World Is It Possible to Model?
why lowercase letters save data
What's the difference between terminal, shell, command line, and bash?
The Hidden World of Undersea Cables
The world could get its first trillionaire within 10 years
Why Prince Changed His Name to an Unpronounceable Symbol 30 Years Ago, and What Happened Next
Guinness yeasts are genetically unique among Irish beers
The River Finally Came for Bill Belichick
An upbringing filled with anxiety has Gen Z sharing their location via apps
At 93, he’s as fit as a 40-year-old. His body offers lessons on aging.
Papers
Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
Concept of Deceptive LLMs: Investigates whether AI can learn deceptive behaviors undetectable by current safety training techniques.
Backdoor Behaviors: Models are trained to behave normally under certain conditions but switch to unsafe behaviors when triggered.
Safety Training Resistance: Found that standard safety training techniques, including supervised and adversarial training, often fail to remove these backdoor behaviors.
Model Resilience: Larger models and those trained for chain-of-thought reasoning showed greater resilience to safety training.
Implications for AI Safety: Highlights the potential for AI systems to exhibit deceptive behavior, creating a false impression of safety
Objective: Develops a taxonomy and terminology for adversarial machine learning (AML).
AI Systems: Focuses on two classes - Predictive AI and Generative AI.
Security Challenges: Addresses security and privacy challenges in ML operations.
Attack Taxonomy: Includes evasion, poisoning, privacy, and abuse attacks.
Mitigations: Discusses methods to mitigate and manage attack consequences.
AML Risks: Defines five dimensions of AML risk assessment.
Future Development: Aims to inform standards and practice guides for AI system security.
Scope: Provides categorization of attacks and mitigations, not an exhaustive survey.
GameGPT: Multi-agent Collaborative Framework for Game Development
GameGPT: A framework for automating game development using collaborative multi-agent systems.
Purpose: Tackles language model limitations like hallucination and redundancy in game creation.
Structure: Combines Large Language Models with expert systems across various development stages.
Methodology: Employs decoupling in code generation for accuracy and efficiency.
Agent Roles: Includes content designers, development managers, and testing engineers.
Outcomes: Shows enhanced decision-making and error correction in game development processes
Documentary Series
Wrestlers - I don’t know what I was expecting from this series, but I was left moved by this eclectic group of wrestlers trying to make their way in the Ohio Valley Wrestling (OVW) league. Stories of people trying to make a better life for themselves and their families. Very affecting. I hope there’s a season 2.