weekend ai reads for 2024-10-25

note: we’re trying text fragments in some of our links with quotes today; may not work on Safari

šŸ“° ABOVE THE FOLD: CLAUDE COMPUTER USE

Developing a computer use model / Anthropic (8 minute read)

Claude’s agentic future and the current state of the frontier models — How Claude’s computer use works. Where OpenAI, Anthropic, and Google all have a lead on each other. / Interconnects, Substack (sorry) (12 minute read)

When you give a Claude a mouse / Ethan Mollick, One Useful Thing, Substack (sorry) (10 minute read)

How To Install and Use Claude's New AI Agent / Matt Wolfe, YouTube (14 minute video)

  • starting with installing Docker

  • relatively easy to follow, even for technical naĆÆfs, like us

Claude Computer Use TESTED - This is VERY Promising! / All About AI, YouTube (17 minute video)

  • see it in action

  • he doesn’t smooth over any of the rough edges and there are a lot

 

šŸ“» QUOTE OF THE WEEK

The moment you go down the path of gratitude grievances, you’ll see ungrateful ghosts everywhere.

David Heinemeier Hansson (source)

 

šŸ‘„ FOR EVERYONE

Are We Too Impatient to Be Intelligent? / Rory Sutherland, Behavioral Scientist (19 minute read)

And businesspeople, governments, and politicians aren’t looking to solve problems; they’re looking to win arguments. And the way you win an argument is by pretending that what should be an open-ended question with many possible right answers isn’t one.

The 3 AI Use Cases: Gods, Interns, and Cogs / Drew Breunig (7 minute read)

And while Interns are delivering tremendous value, they are secondary to the experts driving them. How do you improve the output of an AI copilot? Simple: find a better expert.

Character.ai Faces Lawsuit After Teen’s Suicide / New York Times (18 minute read)

She accused it of harvesting teenage users’ data to train its models, using addictive design features to increase engagement and steering users toward intimate and sexual conversations in the hopes of luring them in.

 

šŸ“š FOUNDATIONS

How to get started with Google’s NotebookLM / Google blog (8 minute read)

The meta-prompt (a prompt optimizer) and system prompt (a model conditioner) operate behind the scenes, issuing detailed instructions that dictate how the AI should behave throughout an interaction. When users initiate a conversation, the meta-prompt has already set the stage, guiding the AI on everything from understanding the main objective of a task to structuring its output.

Prompt engineering methods that reduce hallucinations / The Prompt Engineering Substack, Substack (sorry) (7 minute read)

#1: Step-Back Prompting

#2: Chain of Verification

#3: According to…

 

šŸš€ FOR LEADERS

The AI Investment Boom / Apricitas Economics, Substack (sorry) (12 minute read)

with hundreds of billions of dollars going to high-end computers, data center facilities, power plants, and more. Right now, US data center construction is at a record-high rate of $28.6B a year, up 57% from last year and 114% from only two years ago.

TLDR: Don’t buy H100s. The market has flipped from shortage ($8/hr) to oversupplied ($2/hr), because of reserved compute resales, open model finetuning, and decline in new foundation model co’s. Rent instead.

ā€œYou don’t need an AI strategy, you need a business strategy for AIā€ — Microsoft tells us why utilizing AI at work could not only make you more productive — but happier too / Tech Radar (6 minute read)

She notes the recent Microsoft Work Trend Index found three quarters (75%) of respondents are now using AI at work in some way - double the number from six months ago, with the technology helping boost creativity and freeing up time to focus on crucial tasks.

The team pointed out that data being encrypted today could be at risk if adversaries are stealing it with the intention of decrypting it in the future when quantum technology advances.

  • we would guess the target area for state hackers just became much larger; they’re no longer deterred by encrypted data because they can just steal it and wait

  • probably a good nudge to upgrade to AES-256, SPHINCS+, or some other post-quantum encryption protocol

 

šŸŽ“ FOR EDUCATORS

A 2023 study by Stanford University researchers found that AI detectors were ā€œnear-perfectā€ when checking essays written by US-born eighth grade students, yet they flagged more than half of the essays written by nonnative English students as AI-generated.

In all, about 1,000 students got help from about 900 tutors, and students who worked with AI-assisted tutors were four percentage points more likely to master the topic after a given session than those in a control group whose tutors didn’t work with AI.

How Harmful Are AI’s Biases on Diverse Student Populations? — Large language models exhibit alarming magnitudes of bias when generating stories about learners, often reinforcing harmful stereotypes / Human-Centered Artificial Intelligence, Stanford University (7 minute read)

Duolingo CEO Luis von Ahn on AI, gamification, and the power of freemium — The cofounder of the world’s largest education app thinks AI and gamification can supercharge language learning. / The Verge (84 minute audio, with transcript)

 

šŸ“Š FOR TECHNOLOGISTS

Automating a Marketing Team with AI Agents / Alejandro AO, YouTube (63 minute video)

  • detailed and specific walkthrough of creating and setting up agents

OpenAI’s shift toward system two thinking could reshape the competitive landscape for AI, especially in enterprise applications. While most current models are optimized for speed, the deliberate reasoning process behind o1 could offer businesses more accurate insights, particularly in industries like finance and healthcare.

Aria: First Open Multimodal Native MoE Model / Rhymes blog (13 minute read)

Aria processes text, images, video, and code all at once, without needing separate setups for each type, demonstrating the advantages of a multimodal native model.

  • have not yet tested this model

Llama-3.1-Nemotron-70B-Instruct / Nvidia blog (2 minute read)

  • try it at the link above

  • the speed is impressive

 

šŸŽ‰ FOR FUN

Pope Francis and the Vatican just created an ā€œAI Bibleā€ reshaping faith in the Digital Age — The Vatican has signaled its determination to be an influential voice in the rapidly advancing field of artificial intelligence. / The Brighter Side of News (7 minute read)

Thom Yorke and Julianne Moore join thousands of creatives in AI warning — Statement comes as tech firms try to use creative professionals’ work to train AI models / The Guardian (5 minute read)

i h8 ai / Meta, YouTube (3 minute video)

  • well-done and emotionally manipulative ad for Meta AI

 

🧿 AI-ADJACENT

ā€œIt’s not that we need to enhance skepticism, per se. It’s that we need to give them the ability to use that skepticism to their advantage,ā€ Orticio said.

 

ā‹„