weekend ai reads for 2024-08-16

August 16, 2024

📰 ABOVE THE FOLD: RETAIL

This report is based on a survey of 2,500 senior leaders of global enterprises ($10M+ revenue), conducted by Google Cloud and National Research Group. It provides a comprehensive benchmark of the impact of gen AI on business and financial performance.

4 Proven AI Adoption Approaches for “Normal” Businesses / Tobias Zwingmann, The Augmented Advantage (9 minute read)

  • by “normal”, they mean non-big tech and non-startup; so probably your business

COSMOPlat can make unexpected connections, such as tapping a refrigerator insulation specialist to create a vibration-reducing material for washing machines. COSMOPlat even integrates its member companies’ networks to manage distribution and logistics, based on their strengths in each territory.

  • more interesting is the transcript of Walmart’s Q2 2025 earnings call where the CEO talks about AI in his opening statements, and then answers a few questions that refer to AI / The Motley Fool (57 minute read, but about 8 minutes of AI content)

And one of the interesting things that’s happening with generative AI is that cross-category search is more effective, which serves up more general merchandise items, and it helps drive e-commerce profitability, as you were asking about earlier. So, we're in this situation where we’ve got the best e-commerce food offer, and new tools are helping us connect impulse items that are general merchandise in some cases, which helps us improve both sales and profitability.

The company has partnered with Microsoft to develop an Electronic Shelving Label (ESL) system known as Enhanced Display for Grocery Environment (EDGE), using a digital tag to display prices in stores so that employees can change prices throughout the day with the click of a button.

 

📻 QUOTE OF THE WEEK

The marginal product of LLMs is when they are interacting with well-prepared, intricately cooperating humans at their peak, not when you pose them random queries for fun.

Tyler Cowen (source)

 

🏗️ FOUNDATIONS & CULTURE

Actually, we don't want an AI to “do what I want”.

We want an AI to “do what I would have wanted if I knew the outcomes in advance.”

…

What’s worse, by design the most popular machine learning techniques can only find correlations in data, not actual cause-and-effect. Which means AIs will discriminate on traits by default!

There is a difficult question (and one that is painful for me): Is freely shared knowledge always a globally good thing? If we had the DNA sequence of an extremely dangerous virus, would it be best to share it publicly or not?

  • both are long and important; Nicky Case’s has more comics

The AI scams infiltrating the knitting and crochet world - and why it matters for everyone — AI can create unexpected problems for makers. Here's what to watch out for, and why all creators, hobbyists, crafters, and bakers should pay attention. / Zdnet (11 minute read)

How Culture Shapes What People Want from AI — Stanford researchers explore how to build culturally inclusive and equitable AI by offering initial empirical evidence on cultural variations in people’s ideal preferences about AI. / Human-Centered Artificial Intelligence, Stanford University (11 minute read)

Nephew says that she was driven to make the LLM-free label for a number of reasons: She doesn’t want to promote tools that could take people’s jobs, she’s not convinced LLMs are reliable as a business solution, and her early days working in a start-up in the first dot-com boom taught her that, ultimately, clients want sensible tools whose output they understand.

Here’s how people are actually using AI / MIT Technology Review (7 minute read)

Mahari was part of a group of researchers that analyzed a million ChatGPT interaction logs and found that the second most popular use of AI was sexual role-playing. Aside from that, the overwhelmingly most popular use case for the chatbot was creative composition.

xAI’s new Grok image generator floods X with controversial AI fakes — xAI’s latest Grok feature is exactly as chaotic as you might expect. / The Verge (6 minute read)

How does ChatGPT work? Explained by Deep-Fake Ryan Gosling. / How to Fly, YouTube (8 minute video)

In this video, a deep-fake version of Ryan Gosling (created with ElevenLabs and SyncLabs) explains the high-level workings of an LLM like ChatGPT.

  • surprisingly decent deep-fake and explainer

  • related, more niche explainer, Multimodal RAG — Modern RAG for modern models. / Daniel Warfield, Intuitively and Exhaustively Explained, Substack (15 minute read)

A few hours later, my panic turned to bewilderment. When I finally reached my insurance broker, he told me the reason Travelers revoked my policy: AI-powered aerial surveillance. My finances were imperiled, it seemed, by a bad piece of code.

…

Instead, as my broker revealed, the ominous threat that canceled my insurance was nothing more than moss.

 

🎓 EDUCATION

AI’s Impact on Black Americans — A new white paper for the Congressional Black Caucus addresses risks and opportunities of AI for Black communities. / Human-Centered Artificial Intelligence, Stanford University (7 minute read)

  • from the paper [PDF]: “AI-powered video analytics and behavioral biometrics could perpetuate inequalities by performing worse for darker- skinned people and acting as surveillance tools.”

A primary goal of the program is to bring Nvidia resources into community colleges, including curriculum, certifications, software, and bootcamps, so that students can learn how to use AI to help get in-demand jobs. By introducing workshops and labs to students who wouldn't otherwise have access, the program aims to open up new career pathways that could benefit not only students, but California as a whole.

LG, Samsung eye South Korea’s AI textbooks as edtech springboard — Tech companies look to tap foreign markets for robots, digital boards and apps / Nikkei Asia (4 minute read)

Justin Wolfers Webinar: Assigning Homework in a World with ChatGPT / Macmillan Learning, YouTube (68 minute video)

Workshop Resources for Student Supports / AI + Education = Simplified, Substack (sorry) (9 minute read)

 

📊 DATA & TECHNOLOGY

A new public database lists all the ways AI could go wrong — Its creators hope their work could lead to further research to determine which risks to take more seriously. / MIT Technology Review (4 minute read)

  • the database, requires Google account: AI Risk Repository — A comprehensive living database of over 700 AI risks categorized by their cause and risk domain.

  • related, AI Can’t Save Us By Itself / Civic Innovations (4 minute read)

Most importantly, this all suggests that the “human-in-the-loop” approach to implementing new AI solutions in government may not be enough of a check on the tendency for humans to allow algorithms and AI models to supplant their judgement.

In addition to red teaming at the base model level, we also red team specific features. Red teaming projects at the feature level use feature-specific guidelines with attack vectors informed by the feature’s safety policy and engineering concerns. These projects can provide in-depth probing of known risks for that particular feature and also adversarially probe for unknown vulnerabilities.

I’m excited to share that we've built the world’s most capable AI software engineer, achieving 30.08% on SWE-Bench – ahead of Amazon and Cognition. This model is so much more than a benchmark score: it was trained from the start to think and behave like a human SWE. / Alistair Pullen, Twitter (sorry) (3 minute video)

Why I bet on DSPy / Isaac Miller’s Blog (11 minute read)

If you don’t know what DSPy is, it is an open-source framework that helps you compose multiple LLM calls together in a principled manner to solve a problem.

A team of researchers from the University of the Republic in Uruguay has developed an AI-based system that can intercept and reconstruct unintended electromagnetic emissions from HDMI cables.

Launching a Startup in the Age of Generative AI / Vasco Magellan, Altar, Medium (10 minute read)

In contrast, the last two layers — Domain Specific Context and Application Layer — are more universally applicable across various businesses and offer broader opportunities for engagement and innovation.

Let’s focus on those last two, as they are transversal to all modern entrepreneurs — technical or not.

 

🎉 FUN and/or PRACTICAL THINGS

ToonTalk — Struggling to captivate your child with stories or answer their endless “whys”? Wish for a little helper during those busy moments? / Apple App Store

  • Chrome plugin

After installing AI Eraser, a green “Redact” button will appear in the message bar whenever you visit ChatGPT.com. Click the button to redact and click it again to undo.

Renovate AI — Plan your home renovation with AI

Deep Live Cam — real time face swap and one-click video deepfake with only a single image / hacksider, GitHub

The girls are using ChatGPT to see if men are lying about their height on dating apps. Upload 4 pictures, it uses proportions and surroundings to estimate height.I tested it on 10 friends & family members - all estimates were within 1 inch of their real height / Justine Moore, Twitter (sorry)

Rate Loaf — Rate how well your cat sits like bread

  • useless, but interesting into the insight into how it was built

People say Artificial Intelligence will mold every aspect of our lives and yet large amounts of the internet is still manually rating cat loaves. Every day the rate of cat pictures is far surpassing the number of qualified loaf judgers and this discrepancy is only going to get more crumby as time goes on.

 

🧿 AI-ADJACENT

The Lumen Prize — 2024 Finalists

Intent-driven User Interfaces / Luke W (3 minute read)

While these kinds of interactions won’t immediately replace conventional graphical user interface controls, it’s pretty clear they enable a new way of control software with hundreds of features... just tell it what you want to do.

 

⋄