Week 4 of 2024- W042024
Let the dogs out
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
“Nobody cares, work harder” -Cameron Hanes
“Who let the dogs out? Who, who, who, who.” -The Baha Men
“Release the hounds!” -Montgomery Burns
News
AIML
We Asked A.I. to Create the Joker. It Generated a Copyrighted Image.
Yann LeCun: ‘Human-level artificial intelligence is going to take a long time’
New York City Passed an AI Hiring Law. So Far, Few Companies Are Following It.
OpenAI: New embedding models and API updates
Language models might be able to self-correct biases—if you ask them
GARField: Group Anything with Radiance Fields
The Evolution of Retrieval Augmented Generation
Nightshade: An offensive tool for artists against AI art generators
Demystifying Mixtral of Experts
Professor warns against STEM studies amid AI’s rise
Painting
Head of a Dog, Edvard Munch, (painted 1930), Munch Museum, Oslo, Norway
Angry Dog, Edvard Munch, (ca 1938–43), Munch Museum, Oslo, Norway
Crypto
AI and crypto mining are driving up data centers’ energy use
Cyber
23andMe is reportedly turning the blame back on its customers
Russian is sanctioned over the hacking release of Australian health insurer client data
Markets
Hedge funds rake in huge profits betting on catastrophe risks
World’s largest hedge funds record bumper year of profits
All the Major Tech Layoffs in 2024 So Far
No, office mandates don’t help companies make more money
Remote Work Doesn’t Seem to Affect Productivity, Fed Study Finds
The return-to-office wars are over
To Slash Carbon Emissions, Colleges Are Digging Really Deep
In the Ozempic Age, Has ‘Craveable’ Lost Its Selling Power?
Misc
Hit Man (2001)- basis for new Linklater film
Why is machine learning 'hard'?
Disney Unveils the HoloTile Floor
Inside the Weird, Shady World of Click Farms
Why Holes at the Bottom of the Ocean Disappear and Reappear
How a baker survived the Titanic by getting spectacularly drunk (2022)
The insides of pro bowling balls (2020)
Papers
Self-Rewarding Language Models
Concept: Self-Rewarding Language Models self-improve by generating and evaluating their own training data.
Key Process: Iterative self-instruction creation and instruction following, utilizing AI Feedback for data augmentation.
Implementation: Uses Direct Preference Optimization (DPO) with LLM-as-a-Judge prompting.
Results: Improved performance in instruction following and reward modeling, outperforming systems like Claude 2, Gemini Pro, and GPT-4.
Setup: Based on Llama 2 70B model, fine-tuned with human-authored seed data.
Observations: Iterative training showed marked improvements in successive iterations.
Potential Impact: Demonstrates the capability of models to continually self-improve.
Limitations: Results are preliminary, with further evaluations needed for safety and to explore limits of iterative training.
Documentary Series
For All Mankind- What if the Soviets were the first to land on the moon?
Book
The Worst Hard Time: The Untold Story of Those Who Survived the Great American Dust Bowl thanks EK for the rec!