weekend ai reads for 2024-10-11

October 11, 2024

📰 ABOVE THE FOLD: THE BIAS ISSUE IS NOT TAKEN CARE OF

O1 was less likely than GPT-4o to implicitly discriminate — that is, answer in a way that insinuated bias — on the basis of race, age, and gender. But the model was more likely to explicitly discriminate on age and race, the test found.

That’s the conclusion of researchers at Stanford University, who note that “despite advancements in AI, new research reveals that large language models continue to perpetuate harmful racial biases, particularly against speakers of African American English” as dialects are translated into text.

“Data is inherently biased by the healthcare system that we live in today,” he said. “So those who are left out of the healthcare system today or have poor healthcare because they have less insurance or no insurance at all, that’s all reflected in the data, and machines are unfortunately going to pick that up.”

These findings underscore a crucial point: Gender bias in text embedding models isn’t a simple, monolithic problem. It’s a nuanced issue that varies across models, professions and even the specific words used.

Hiring bias for younger workers persists despite employers reporting that mid-career and older hires consistently matched or exceeded the performance of younger colleagues. According to the new report by Generation, “Age-Proofing AI–Enabling an intergenerational workforce to benefit from AI,” this finding is intensified for roles that regularly use AI tools.

 

📻 QUOTES OF THE WEEK

“A.I. won’t replace you, but a shareholder who doesn't care about quality will” is what the slogan should say.

Jason Gorman (source)

 

I’m particularly proud of the fact that one of my students fired Sam Altman.

Geoffrey Hinton (source)

 

👥 FOR EVERYONE

The State of AI Report 2024 / Nathan Benaich, Substack (sorry) (8 minute read)

via philipp, Creating a public counterpoint for AI / Mozilla blog (6 minute read)

Mozilla is releasing a vision for Public AI, a robust ecosystem of initiatives that promote public goods, public orientation and public use throughout every step of AI development and deployment.

Two Nobel Prizes for AI, and Two Paths Forward / Gary Marcus, Substack (sorry) (6 minute read)

Kohls’ lawyer wrote in a complaint that the deepfake of Kamala Harris is satire that should be protected by the First Amendment.

 

📚 FOUNDATIONS

How to Use AI to Answer Questions — You can ask AI your everyday questions. Here's how to get the best results. / Cnet (6 minute read)

What Is Trustworthy AI? / Nvidia blog (14 minute read)

 

🚀 FOR LEADERS

Why AI Isn’t Enough To Beat The Competition / HBR Ideacast (25 minute podcast)

But, as with any technology that sees widespread adoption, AI itself won’t be enough to build a long-term advantage over competitors, says Jay Barney, professor at the University of Utah’s Eccles School of Business. Yes, leaders need to deploy these new tools, especially those that use GenAI, to stay relevant. But they also need to think about how AI can be applied to their business’ differentiating competencies and offerings to truly add value.

AI Doesn’t Know Much About Golf. Or Farming. Or Mortgages. Or … — Companies are discovering that off-the-shelf models, trained on the vastness of the internet, may still lack specific, deep knowledge on their sector / Wall Street Journal (8 minute read)

  • how this wasn’t blatantly obvious before is beyond us

Oversight of AI in the boardroom — In a new Deloitte Global survey of board directors and executives, almost 50% say AI is not yet on the board agenda. Is it time to step up AI oversight in the boardroom? / Deloitte Insights (21 minute read)

 

🎓 FOR EDUCATORS

  • the primary audience is the student

Customizable AI-Powered Textbooks Reshape Learning / Getting Smart (7 minute read)

The most common reason they gave was AI was not useful or relevant to their work.

 

📊 FOR TECHNOLOGISTS

Notebook LM: A threat to the Podcasting World / Listen Notes (8 minute read)

  • the code, notebooklm-detector — Detect whether or not an audio file was generated by Notebook LM / ListenNotes, Github

  • we’re not sure whether this says more about the relative high quality of Notebook LM or the relative low quality of most podcasts

  • more prompt guardrails to include in your apps

LM Studio 0.3.4 ships with Apple MLX / LM Studio Blog (14 minute read)

  • much faster than Ollama

  • only seven MLX models are currently available, but that includes Llama 3.2 and Phi 3 so you might be covered?

 

🎉 FOR FUN

I Hired ChatGPT as My Career Coach — AI changed how I think about work—and myself / Learning Curve, Every (16 minute read)

  • so good

Scryfall.AI — find Magic: The Gathering cards using natural language

How AI is transforming the fan experience — Trends in sports consumption along with data availability is a game-changer for sports. / Fast Company (5 minute read)

 

🧿 AI-ADJACENT

Why Is Everything So Ugly? — The mid in fake midcentury modern / N+1 Magazine (23 minute read)

 

⋄