Welcome AI Explorers

While AI companies are busy flexing their quarterly earnings (spoiler: they're all up), we're focusing on what matters - making sense of AI in education. This week's newsletter dives into ChatGPT's attempt to play hide-and-seek with its IP tracking, new AI search wars, and a fascinating game that simulates the race to artificial general intelligence (spoiler #2: it doesn't end well).
In this edition:

• 🔍 ChatGPT's new search feature has Google sweating

• 🎭 A Broadway play featuring AI and starring Iron Man himself

• 🛠️ A toolkit from the U.S. Department of Education for AI integration

• 📚 Latest research on AI in education (it's complicated)

• 📊 The latest AI benchmarks and testing results

Prompt of the Week 💭

Twitter user @bilawalsidhu used Gemini 1.5 pro to get feedback on a keynote speech

AI App Spotlight

ChatGPT has a new web search feature ChatGPT Search

• Provides real-time web search with source links

• Available today for Plus and Team users. Available to Free users in next few months

• Includes specialized displays for weather, stocks, sports, news, and maps

• Partners with major news publishers including AP, Reuters, FT, and others

Watch Out Google. ChatGPT and Perplexity are both coming for you. These two AI search tools are changing the way we find information online.

• They deliver conversational and direct answers without requiring you to wade through links

• No ads in the results

• Google still provides more comprehensive web-wide results, but ChatGPT and Perplexity are good choices for quick and focused queries

To be fair, Google is has updated their search with AI supported answers, but they can't quite seem to get it right (yet).

AI News of the Week

• ChatGPT logs your IP address. This is not surprising, pretty much all websites do. The funny thing is, it wouldn't admit to it. Check out this conversation where I used ChatGPT search to tell me about local breweries near me (I didn't tell it where I live and I erased ChatGPTs memory before starting the chat). It kept justifying that it "guessed" I was in Kelowna because of the growing brewery scene here🤣 . https://chatgpt.com/share/67252074-465c-800c-b029-988d2a23a595

• OpenAI introduces SimpleQA. "A facutality benchmark...that measure the ability for language models to answer short, fact-seeking questions". Even OpenAI's best model only scored 43%.

• Have you heard of jailbreaking an LLM? It's when you give an LLM a series of prompts to get it to say harmful or potentially harmful things (e.g., getting an LLM to say racist things or teach you how to create a bioweapon). Well now you can also jailbreak AI robots...and it's as scary as you might think and surprisingly easy

• A new research game called Intelligence Rising is revealing potentially concerning patterns about the race to develop artificial general intelligence (AGI). In the game players act as tech giants or governments each trying develop AGI and/or ensure AGI is developed safely. Analysis of 43 comprehensive gameplay sessions showed that without strong cooperation between governments and industry, the AGI race typically spirals into global instability - triggering job losses, economic gaps, eroded trust, and military conflicts. While it's just a simulation, these consistent outcomes offer insights into how real-world AGI governance needs to unfold.

• ChatGPT now has a Windows desktop app (Mac desktop app also available). ChatGPT pro users only.

• Claude Sonnet 3.5 and ChatGPT are still the best LLMs according to these leaderboards:

  • https://livebench.ai/

  • https://www.vellum.ai/llm-leaderboard

  • https://huggingface.co/spaces/lmarena-ai/chatbot-arena-leaderboard)
    but Google Gemini is not too far behind. It's hard to go wrong with whichever of those three you choose.

• Robert Downey Jr. is starring in the Broadway play, McNeal. It explores the tension between human creativity and artificial intelligence. To hear a full conversation about the play with Downey, the writer (Ayad Akhtar), and the director (Bartlett Sher), check out this podcast.

• AI is bringing in the $'s. In the most recent quarterly reports Google Cloud revenue up 35%, Meta revenue up 19%, Microsoft revenue up 16%, Nvidia up 250%.
  But spending (especially on infrastructure) is up significantly for these companies too.

AI Powered Pedagogy 👩‍🏫

• In the latest working paper from the MLA-CCCC Joint task force on Writing and AI (Building a Culture for Generative AI Literacy in College Language, Literature, and Writing), the authors argue that "it is in our collective professional interest to offer students, colleagues, and administrators balanced and informed perspectives on the risks and harms as well as the potential benefits. They then set out suggested GenAI Literacy learning outcomes for students, educators, and institutions. This is one of the more comprehensive documents describing the why and how of GenAI literacy targetted for educators

• Research into efficacy of GenAI in education is still new (both of these papers are still in pre-print). Results are mixed and even conflicting:

  1. AI tutoring outperforms active learning: https://www.researchsquare.com/article/rs-4243877/v1

  2. Gen AI can harm learning: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4895486

  • While the headlines for these papers seem conflicting, there may be some lessons to learn when taking the two papers together. The most important seems to be: Incorporating pedagogical best practices (active learning, cognitive load management, and growth mindset) when adopting AI to support learning may mitigate the risks associated with GenAI use (over-reliance, hallucinations, not assessing results for credibility and correctness

    • This paper, Efficiency-Accountability Trade-off in integrating AI indicates that student over-reliance on GenAI is a major risk, especially when using it to get direct answers

The mAIn Event

A Toolkit for Safe, Ethical, and Equitable AI Integration

A Toolkit for Safe Ethical and Equitable AI Integration from US Department of Education, Office of Educational Technology

Big picture: The U.S. Department of Education has released a new toolkit helping education leaders safely and ethically integrate AI into teaching and learning. It provides practical guidance while emphasizing student privacy, security and non-discrimination. It is targeted towards K-12 education, but there are a lot of valuable insights for educators at all levels.

Why it matters: Schools need clear frameworks to maximize benefits while protecting students. This toolkit gives leaders actionable steps to develop responsible AI strategies.

Key components:
Risk Management

• Requires careful evaluation of AI tools' impact on privacy, security and civil rights

• Emphasizes transparency about how AI systems collect and use student data

• Warns against high-risk uses like automated disciplinary decisions or biometric surveillance

Strategic Implementation

• Recommends aligning AI with existing instructional goals rather than replacing teachers

• Suggests starting with educator AI literacy before classroom integration

Building AI Literacy. Educators need skills in:

• Understanding AI capabilities and limitations

• Evaluating tools for privacy, bias and evidence

• Integrating AI thoughtfully into instruction

Between the lines: While highlighting AI's potential to personalize learning and reduce administrative work, the toolkit consistently prioritizes human judgment and student protection over automation.

What's next: Schools are advised to:

1. Assess current AI use and risks

2. Develop clear responsible use policies

3. Build educator AI literacy

4. Create implementation plans aligned with learning goals

The bottom line: This toolkit provides education leaders a framework to proactively shape AI use in schools while keeping student wellbeing at the center.

Go deeper: Access the full toolkit