weekend ai reads for 2024-07-19

July 19, 2024

📰 ABOVE THE FOLD: ROBOTS

Google says Gemini AI is making its robots smarter — DeepMind is using video tours and Gemini 1.5 Pro to train robots to navigate and complete tasks. / The Verge (3 minute read)

The convenience store chain Re-Up announced that it’s installing “The Wingman,” a robot from Nala Robotics that drops fry baskets into hot oil and rolls chicken wings around in sauce before dumping those things into buckets for your consumption (at least, based on the video below).

“Like a muscle, these soft actuators actually stiffen,” Truby said. “If you have ever twisted the lid off a jar, for example, you know your muscles tighten and get stiffer to transmit force. That's how your muscles help your body do work. This has been an overlooked feature in soft robotics. Many soft actuators get softer when in use, but our flexible actuators get stiffer as they operate.”

Orbisk’s technology uses cameras and scales to detect the types of food hitting the bin — uncovering previously hidden opportunities for waste reduction. For example, in its first two months using Orbisk, Carnival found that its Bonsai Sushi restaurants have been scrapping a considerable amount of cucumber as chefs mainly used the outside of the veggie in dishes.

Watch mini humanoid robots showing off their football skills — These soccer-playing robots can respond faster than ones trained in a standard way because they improved their skills via an artificial intelligence-based technique called deep reinforcement learning / New Scientist (4 minute read)

 

📻 QUOTES OF THE WEEK

In order for me to write poetry that isn’t political

I must listen to the birds

and in order to hear the birds

the warplanes must be silent.

Marwan Makhoul (source)

 

The persistent are attached to the goal. The obstinate are attached to their ideas about how to reach it.

Paul Graham (“The Right Kind of Stubborn”, source)

 

🏗️ FOUNDATIONS & CULTURE

Chum King Express — The man behind the AI gaffes at Sports Illustrated and USA Today has a yearslong history of filling the internet with garbage. / The Verge (23 minute read)

For years, according to former colleagues, internal documents, and court records, a person behind the operation has seemingly used his connections in the media industry to enrich himself: Ben Faw, CEO and cofounder of AdVon. For Faw, AI-generated sludge was just the next tactic to do so.

$3 billion Lattice ‘made history’ by giving its AI ‘digital workers’ the same hiring treatment as humans — A $3 billion software unicorn scrapped its plan to give AI ‘workers’ rights after tech execs said it ‘disrespects the humanity of your real employees’ / Fortune (5 minute read)

  • the popular (and easy) thing to do online was to be angry, incredulous, or dismissive of this

  • but the reality is that as “digital workers” (aka: “apps”) play a more active role in business, there will be a need to onboard, train, upskill, modernize, etc. them; the process won’t be identical but it will have parallels to hiring humans

The Content Origin Protection and Integrity from Edited and Deepfaked Media Act (COPIED Act) would direct the National Institute of Standards and Technology (NIST) to create standards and guidelines that help prove the origin of content and detect synthetic content, like through watermarking. It also directs the agency to create security measures to prevent tampering and requires AI tools for creative or journalistic content to let users attach information about their origin and prohibit that information from being removed. Under the bill, such content also could not be used to train AI models.

You Don’t Need an LLM Agent / Louis Chan, Toward Data Science, Medium (7 minute read)

  • KPMG employee suggests agents may introduce unpredictability into how businesses run their operations; the different proposal is to use LLM Pipelines, which may provide a neat and traceable way to blend in GenAI with the stuff companies are already doing.

  • counterpoint (but it may just be definitional differences), The Future of Knowledge Assistants / Jerry Liu, AI Engineer, YouTube (17 minute video)

Mapping the Landscape of AI-Powered Nonprofits / Stanford Social Innovation Review (12 minute read)

We developed a working landscape to connect the dots, and in the end, four big categories of nonprofit AI use cases emerged: Structuring Data, Advising, Translating, and Platforms. Excitingly, we found that most APNs don’t use AI in just one way. Their solutions tend to consist of a constellation of use cases.

Where to get started with GenAI / Byte Byte Go Newsletter, Substack (sorry) (13 minute read)

  • we sometimes forget the “Foundations” part of this section; thorough summary of a lot of aspects of GenAI to build or maintain that foundation

 

🎓 EDUCATION

Teachers would still design the course material, but they’d be supported by this AI assistant. The startup does not yet appear to have built or tested the efficacy of integrating AI assistants into the classroom. At least one Georgia State University study found that AI teaching assistants helped some students get better grades.

This Teacher + AI symbiosis could run an entire curriculum of courses on a common platform. If we are successful, it will be easy for anyone to learn anything, expanding education in both reach (a large number of people learning something) and extent (any one person learning a large amount of subjects, beyond what may be possible today unassisted).

We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students. Across modules there was an 83.4% chance that the AI submissions on a module would outperform a random selection of the same number of real student submissions.

“The AI in education market size reached US$3.52 billion in 2023. Looking forward, Reports and Insights expects the market to reach US$58.52 billion by 2031, exhibiting a growth rate (CAGR) of 42.1% during 2024-2032.”

Pearson has compared student engagement with the eTextbooks that have the AI tools versus those that don’t and found that students with the AI tools are nearly doubling their number of sessions with their textbooks. The company plans to expand the technology to more and more titles and has decided that every Pearson textbook will be AI-equipped. Hess said he strongly believes in making sure this doesn’t become something “the haves get and the have-nots don’t.”

My trip to the frontier of AI education — First Avenue Elementary School in Newark is pioneering the use of AI tools in the classroom. / Gates Notes (5 minute read)

Bhutan’s first AI startup is seven college kids in a dorm — NoMindBhutan services prominent clients like the Bhutan National Bank and Drukair - Royal Bhutan Airlines. / Rest of World (8 minute read)

 

📊 DATA & TECHNOLOGY

How To Price A Data Asset — Everything you ever wanted to know about data pricing. / Abraham Thomas, Pivotal, Substack (sorry) (28 minute read)

  • comprehensive; no actual dollar figures discussed

A data leader’s technical guide to scaling gen AI / McKinsey Digital (15 minute read)

It starts at the source: Improve your data

Access to proprietary data will be increasingly important. Intellectual property falls into this category. Disney has access to its vast catalog of films; The New York Times has more than 150 years of archived articles; Pfizer has its decades of drug R&D and clinical trials. But for most companies, the biggest source of proprietary information is customer data.

Time Series Are Not That Different for LLMs — Harnessing the power of LLMs for time series modeling / Henry Lai, Towards Data Science, Medium (10 minute read)

How to Finetune Llama-3 and Export to Ollama — Beginner’s Guide for creating a customized personal assistant (like ChatGPT) to run locally on Ollama / Unsloth Docs (16 minute read)

  • 1-2 hour project, depending how much Unsloth you want to learn

  • absolutely worth doing

  • uses free Google Colab for GPU

Exo — Run your own AI cluster at home with everyday devices. / exo, GitHub

  • okay, no one is going to do this; but great proof of concept

  • unless … you work in IT have a bunch of old laptops and phones and tablets lying around

 

🎉 FUN and/or PRACTICAL THINGS

Rufus is designed to help customers save time and make more informed purchase decisions by answering questions on a variety of shopping needs and products right in the Amazon Shopping app—it’s like having a shopping assistant with you any time you’re in our store. We’re pleased to share that Rufus is now available to all U.S. customers in the Amazon Shopping app.

A MidJourney sref Code (style reference) is a specific parameter used on the MidJourney platform to generate unique and consistent styles of images. Each sref code corresponds to a particular effect, making it a powerful tool for artists and creators looking to achieve specific visual outcomes.

Whisper Timestamped / Xenova, Hugging Face

  • transcript from video, with timestamp for each word

Gendo. AI for architecture — Enterprise and Professional grade AI visualisation - built by, and for designers

Rep. Jennifer Wexton has lost her ability to speak, so “AI has allowed me to make a new model of my voice like it was before my PSP.” / Rep. Jennifer Wexton, Twitter (sorry) (2 minute video)

Gemini, Please Analyze My Poem. — One Powerful Way I, as a Poet, Have Embraced AI / Sierra Elman, Medium (5 minute read)

Eno documentary: behind the first generative feature film — Director Gary Hustwit tells us about his living documentary and the technology that changes it for every screening. / The Verge (13 minute read)

 

🧿 AI-ADJACENT

Capybara Affirmations AI — Get customized affirmations to power your positivity. And… There are capybaras.

 

⋄