weekend ai reads for 2024-07-12

July 12, 2024

📰 ABOVE THE FOLD: ENERGY

  • related, Google Is No Longer Claiming to Be Carbon Neutral — The tech giant, which has seen its planet-warming emissions rise because of artificial intelligence, has stopped buying cheap offsets behind the neutrality claim. The company now aims to reach net-zero carbon by 2030. / Bloomberg (7 minute read)

“Let’s not go overboard on this,” Gates said. “Datacentres are, in the most extreme case, a 6% addition [in energy demand] but probably only 2% to 2.5%. The question is, will AI accelerate a more than 6% reduction? And the answer is: certainly.”

…

Some expert estimates have claimed that an increase in the number of AI datacentres could cause electricity demand to rise by up to 10% in developed countries, after years of declining energy due to greater efficiency.

The Potential Impact of AI on the Public-Sector Workforce [PDF] / Tony Blair Institute for Global Change (33 minute read)

In addition, we assume each public-sector worker that uses the technology incurs a small cost related to the use of computer hardware (including the cost of energy used to power the hardware), as well as data-storage and transfer costs. We conservatively estimate the annual cost is around £30 per user or £155 million per year across all applicable workers.

 

📻 QUOTE OF THE WEEK

But I’m worried that this focus on skill and being meritocratic—dubbed MEI (merit, excellence and intelligence) by one founder—misses what it really means to be meritocratic. If you want to find the most skilled talent period, you have to work very, very hard to find it across populations that are underserved. You can’t find it by picking the most qualified person who comes in your door.

Jessica Lessen (source: Why Women Are Disappearing from Tech) ($)

 

🏗️ FOUNDATIONS & CULTURE

Pop Culture / Where’s Your Ed At (18 minute read)

One particular myth Covello dispels is comparing generative AI “to the early days of the internet,” noting that “even in its infancy, the internet was a low-cost technology solution that enabled e-commerce to replace costly incumbent solutions,” and that “AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn't designed to do.”

I Switched to Claude 3.5 for ChatGPT, here's why — Speculations on the role of RLHF and why I love the model for people who pay attention. / Interconnects, Substack (sorry) (8 minute read)

The Evaluate feature, now available directly in the console, allows users to test prompt quality against real input before deploying to production. New test cases can be added manually, imported from a CSV file, or automatically generated by Claude.

The survey finds that 52% of likely voters say they are not interested in using these services with new AI features included. While Democrats are split on this issue, 57% of both Independents and Republicans say they are not interested in using AI features being introduced in big tech companies’ consumer products.

We will repeal Joe Biden’s dangerous Executive Order that hinders AI Innovation, and imposes Radical Leftwing ideas on the development of this technology. In its place, Republicans support AI Development rooted in Free Speech and Human Flourishing.

Each bot’s profile had an identity or “soul” based on an archetype or persona. They were given biographical information, political ideologies, and a location. For example, one account purported to be a Bitcoin enthusiast and free speech absolutist from Minneapolis who was opposed to the New World Order.

  • stop arguing with bots, people

How I Use AI — The jobs for which I have been using AI as a solopreneur. / Jobs for AI, Substack (sorry) (9 minute read)

 

🎓 EDUCATION

Animated AI TAs are coming to Morehouse College — The AI avatars—which can look like students’ professors—are intended to answer course questions 24-7. / Inside Higher Ed (5 minute read)

Professor AI — Supercharge your prep for AP’s with Professor AI

this guide seeks to inform product leads and their teams of innovators, designers, developers, customer-facing staff, and legal teams as they work toward safety, security, and trust while creating AI products and services for use in education. This landscape is broader than those building large language models (LLMs) or deploying chatbots; it includes all the ways existing and emerging AI capabilities can be used to further shared educational goals.

  • the report [PDF] (90 minute read)

  • key recommendations:

  1. Design with teaching and learning and mind

  2. Show how evidence-backed principles were used to create products

  3. Take steps to remove or mitigate bias

  4. Protect student privacy

  5. Be transparent about how products are designed

What aspects of teaching should remain human? — Even techno optimists hesitate to say teaching is best left to the bots, but there’s a debate about where to draw the line / The Hechinger Report (14 minute read)

 

📊 DATA & TECHNOLOGY

via adam m, Running Fabric Locally with Ollama: A Step-by-Step Guide / Bernhard Knasmüller (7 minute read)

  • Daniel Miessler talking through Fabric: Hack your life (with demos) and get Superpowers! / David Bombal, YouTube (1 hour video)

  • source for Fabric — provides a modular framework for solving specific problems using a crowdsourced set of AI prompts that can be used anywhere / danielmiessler, GitHub

  • fascinating; worth playing with if you have time; adam m recommends Obsidian as a backend; we concur

AI Agents That Matter / arXiv (87 minute read)

  • long paper that identifies AI agents’ benchmark and evaluation practice shortcomings and introduces steps “for addressing these shortcomings [to] spur the development of agents that are useful in the real world and not just accurate on benchmarks.”

  • worth reading if benchmarking is anywhere near your set of responsibilities or interests

Documentation: Private-Sector AI Indicators — The Private-Sector AI Indicators dataset includes a diverse range of indicators of AI-related activity for hundreds of companies worldwide, from startups to multinationals / Emerging Technology Observatory

 

🎉 FUN and/or PRACTICAL THINGS

ChatGPT is bull / Ethics and Information Technology, Springer (39 minute read)

Many popular discussions of ChatGPT call its false statements ‘hallucinations’. One also might think of these untruths as lies. However, we argue that this isn’t the right way to think about it. We will argue that these falsehoods aren’t hallucinations later. For now, we’ll discuss why these untruths aren’t lies but instead are bull.

  • describe a color and AI will find that color; somehow, “cartwheel” is exactly the color it should be

  • related, AI color palette generator — Inspirational color palettes tailored to your vision

  • unrelated (1), ElevenLabs Reader — Convert text into natural, expressive speech. Ideal for articles, ePubs, PDFs, or any text. / iOS App Store

  • unrelated (2), Free Online TTS Maker — Quickly convert text to speech with multiple languages and natural voices.

In quick summary, it has delivered about a 2:1 win/loss ratio advantage using just one day holds, significantly improving returns and Sharpe Ratios relative to the “buy and hold” strategy on the S&P 500 index.

FoxVox: one click to alter reality — How AI could subtly manipulate the content you consume / Palisade Research

  • Chrome extension; interesting examples

Live Portrait — add mimics and lip sync to your static portrait driven by a video / KwaiVGI, HuggingFace

CVPR 2024 AI Art — This gallery was curated by Luba Elliott for CVPR 2024 from 346 submissions received in the open call for art that uses or looks at computer vision, including techniques such as generative models, object and facial recognition.

  • much better display of the diversity of AI art than the more common Midjourney/Stable Diffusion ones

Not Quite Past — make your own Delftware with AI

  • and order physical tiles, ones you’ve designed or others’

makes proposed legislation in the UK more accessible and understandable for engaged citizens and advocacy groups who want to stay informed but don't have time to read full legal texts.

  • just great; someone do this for the US

 

🧿 AI-ADJACENT

 

⋄