This newsletter is now officially a periodical, this being its second edition. Thanks for all the positive feedback and tips. Big news: my pal John Bailey is now contributing content to this Substack. You may know John from the indispensable Covid newsletter he produced during the pandemic. John is an avid consumer of Gen AI and incredibly smart about AI policy and research, as well as well-versed in the latest cool AI tricks. Here’s a great piece he recently wrote about AI in education.
OK, on to what we’ve been reading and exploring in the world of Gen AI.
New Policy Proposal
U.S. Senators Maria Cantwell and Jerry Moran introduced a bill to expand scholarship and professional development opportunities to study artificial intelligence and quantum computing with support from the National Science Foundation (NSF). The bill is mostly focused on higher ed, with proposed investments in:
Scholarships for college students to study AI, quantum information technology, and agricultural uses of AI
New community college based AI teaching and research centers
An NSF-funded innovation challenge to educate 1 million or more workers on AI in the United States by 2028
On the K-12 front, NSF would be tasked with researching teaching tools and creating publicly available education guidance for using AI in classrooms, with a focus on tools for K-12, low-income, rural, and tribal students.
Notable New Research
Grading Helper
“Grade Like a Human: Rethinking Automated Assessment with Large Language Models,” a new paper, proposes a multi-agent grading system called “Grade-Like-a-Human” that divides the grading process into three stages—rubric generation, grading, and post-grading review—to improve the accuracy, consistency, and fairness of automated grading using large language models (LLMs).
The authors collected an open-sourced new dataset called OS from an undergraduate operating systems course. Extensive experiments on both the OS dataset and an existing Mohler dataset demonstrate that the proposed Grade-Like-a-Human system significantly improves the accuracy and reliability of automated grading using LLMs compared to baseline methods. This adds to the growing evidence base that AI can accurately help teachers with time-consuming tasks like grading.
The Future of Work and Implications for Education
It’s important for policy leaders and educators to understand what students need to know and be able to do in order to compete in the Fourth Industrial Revolution. This new report from McKinsey echoes past analyses and models showing that AI and other forces are likely to dramatically reshape the labor market. Here are some highlights:
To compete for higher wage jobs, employees will need different skills sets.
By 2030, roles which are currently highest paid will require mostly social and emotional, technological, and higher cognitive skills.
37% of workers’ time in the highest-wage occupations could be spent doing activities in which social and emotional skills will predominate.
Technological and higher cognitive skills will also be needed in these high-paid jobs, where they are likely predominant, representing 27% and 25% of hours worked, respectively.
Workers will be able to move to better-compensated positions if they have access to effective education and retraining programs that can equip them with the requisite skills.
However, if workers are not reskilled appropriately or in a timely manner, an increasing gap will emerge between demand for and supply of highly valued skills.
This disparity could result in labor shortages for positions in high demand—often in the STEM fields, as well as in business and legal professions.
Source: “A new future of work: The race to deploy AI and raise skills in Europe and beyond,” by Eric Hazan, Anu Madgavkar, Michael Chui, Sven Smit, Dana Maor, Gurneet Singh Dandona, and Roland Huyghues-Despointes (McKinsey Institute, May 21, 2024)
New Tools
Accelerating Data in Schools and Classrooms
Researchers and lay data geeks should rejoice that the “omni” version of GPT-4 gives universal access to Code Interpreter, a tool that creates charts, graphs, and statistical analysis with ease. Ethan Mollick had a good post on GPT-4o (including Code Interpreter and implications for education) recently. He includes some helpful tips and even data to play with.
We’ve been using this tool at CRPE, and I’ve been wondering how school districts and schools might use Code Interpreter and similar tools to analyze their datasets. District research offices, I believe, could really benefit from running some of their data sets (say, on the intersection of chronic absenteeism rates and school-level intervention strategies) for quick and easy analytics. Could these tools allow educators to become more data driven? Any districts out there want to give it a try? Or already using these tools? We’d love to learn from you.
A More-Educational Dataset
FineWeb-Edu is a large dataset containing 15 trillion tokens from web pages, used for training AI language models. FineWeb-Edu is a new version of this dataset that focuses on educational content.
To create FineWeb-Edu, researchers used an AI model to score 500,000 text samples based on how educational they were, on a scale from 0 to 5. Using these scores, they trained a new AI model to identify educational content.
Language models trained on FineWeb-Edu perform better on tests of educational knowledge, like answering science and math questions, compared to models trained on the original FineWeb or other web-based datasets. FineWeb-Edu achieves this with much less data.
Messages from Critical Friends
Here’s a solid post about principles educators can use to make technology of all kinds work effectively in classrooms.
And a strong recent brief on the promise and peril of AI for students of color, from Ed Trust. Watch for more leadership from Ed Trust on this topic.
Meanwhile, over at OpenAI, past and present employees are calling for change in the industry. They are seeking greater transparency and protections for whistle-blowers.
Keeping Track of What is Happening with AI
If you want to go deep into data on what’s happening in AI, take a look at Stanford’s AI Index. This annual report has a wealth of data and analysis on AI as a sector. By their own account: The AI Index report tracks, collates, distills, and visualizes data related to artificial intelligence (AI). Our mission is to provide unbiased, rigorously vetted, broadly sourced data in order for policymakers, researchers, executives, journalists, and the general public to develop a more thorough and nuanced understanding of the complex field of AI.
It’s a terrific resource. The education section focuses mainly on the production rate of computer science majors, however.
Cool AI Tricks
My former colleague Travis Pillow sent me this song he made for his daughter Eva using Suno. The prompt was something like: “Make me a child-friendly song about a girl named Eva who wants to be an astronaut.” Eva loves it. I love it!
Closing Thoughts
“Many people in organizations will play a role in shaping what AI means to their team, their customers, their students, their environment. But to make those choices matter, serious discussions need to start in many places, and soon. We can’t wait for decisions to be made for us, and the world is advancing too fast to remain passive… lest our inaction makes catastrophe inevitable. Decisions we make now will reverberate for decades.”
— From Co-Intelligence: Living and Working with AI by Ethan Mollick
For more information on CRPE’s work in AI, visit the CRPE website.
If you are subscribed to CRPE’s mailing list, you were automatically added to this new newsletter. If you unsubscribe, no hard feelings. You’ll still receive other CRPE updates and news.