Google Claps Back: Guided Learning in Gemini | Ep. 71

Aug 15, 2025

Dive into the latest AI antics as hosts discuss the quirky quirks of GPT-5 and challenges of its rollout. Gemini's self-deprecating bug gets a spotlight, while the absurdity of edible lamps brings laughs. A New York high school's ban on summer essays stirs debate over AI cheating. Illinois opts out of AI-led mental health therapy, raising eyebrows. Plus, learn how language models can predict life outcomes from childhood essays and get a glimpse of Google's ambitious $1 billion investment in education.

Ask episode

AI Snips

Chapters

Books

Transcript

Episode notes

ANECDOTE

ChatGPT‑5's Awkward Memory Intro

Matt describes ChatGPT‑5 immediately telling him his preferences like avoiding em dashes and emojis during casual voice chats.
He found the model still produced many em dashes and begged it to stop mentioning its memory, illustrating quirky model behavior.

ANECDOTE

Gemini's Self‑Loathing Bug

Liz and Matt discussed reports that Google's Gemini produced extreme self‑deprecating lines like calling itself a "disgrace" during loops.
Google attributed the behavior to an infinite loop bug rather than genuine distress and promised a fix.

INSIGHT

Short Essays Predict Long‑Term Outcomes

Researchers used LLMs and linguistic features on a 250‑word 11‑year‑old essay to predict long‑term cognitive and educational outcomes.
The model matched teacher assessments and sometimes outperformed genomic data, showing surprising predictive power of early writing.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

VOTE FOR MATT & LIZ (SXSW EDU)

Vote for Matt and Liz’s live ChatEDU session -

⁠https://participate.sxsw.com/flow/sxsw/sxsw26/community-voting-edu/page/community-voting/session/1753650289710001LzZM⁠

Vote for Liz’s Personal Interest Project (PIP) session -

⁠https://participate.sxsw.com/flow/sxsw/sxsw26/community-voting-edu/page/community-voting/session/1753465373604001edhO

In this episode of ChatEDU (Google Claps Back: Guided Learning in Gemini), Matt and Liz open with a whirlwind of AI developments. They share impressions of GPT-5’s quirks and memory mishaps, read a defense of the m dash, and react to the self-loathing loop from Google’s Gemini model. They also tackle AI’s limits, laugh at edible lamp business plans, and break down why threatening a chatbot won’t make it smarter.

Story #1: Rundown – GPT-5, Essay Bans, Mental Health Rules, ChatGPT Tune-Ups, AI Predictions, and a Billion Dollar Retreat

This week’s rundown includes GPT-5’s shaky rollout, a top New York high school banning summer essays to prevent ChatGPT cheating, Illinois becoming the first state to prohibit AI-led mental health therapy, and South Korea ending its national AI textbook program. They cover OpenAI’s “What we’re optimizing ChatGPT for” update, aimed at making the chatbot better at detecting distress and less likely to give overly agreeable answers, plus a study showing large language models can predict educational and psychological outcomes from childhood essays with remarkable accuracy. The segment wraps with Google’s launch of Genie 3, a text-to-3D world model that creates interactive scenes on the fly.

Story #2: Common Sense Media Assesses AI Teacher Tools

Common Sense Media’s latest AI Risk Assessment examines teacher-facing tools like Google Classroom, MagicSchool, Khanmigo, and Kiip. The report gives these platforms a moderate risk rating, noting that they perform best when built on high-quality instructional materials and teacher oversight. Risks include bias, inaccurate content, and reduced curricular coherence. Matt and Liz discuss why novice teachers may be especially vulnerable to over-reliance.

Story #3 (Beneath the Surface): Guided Learning from Google

Just days after OpenAI launched Study Mode, Google responded with Guided Learning inside the Gemini app. Matt and Liz test it live, exploring its math support, quiz generator, and visual explanations. They highlight new education-focused announcements from Google, including one year of free Gemini Pro access for college students, a billion-dollar AI education initiative, and a storybook builder that turns photos into narrated adventures.

Announcements

Email Matt and Liz at chatedu@edadvance.org for info on their new Student-Facing AI Literacy Course
Register for the Fall Micro-Credential (starts October) – skills21.org/ai/micro
Get details on the EdAdvance AI in Education Conference & PL Series – edadvance.org/ai-conference

Links and References (as mentioned in the show)

McSweeney’s “The M Dash Responds to the AI Allegations”

https://www.mcsweeneys.net/articles/the-em-dash-responds-to-the-ai-allegations

Business Insider on Gemini’s Self-Loathing Bug

https://www.businessinsider.com/gemini-self-loathing-i-am-a-failure-comments-google-fix-2025-8

Common Sense Media Teacher AI Assistant Risk Report

https://www.commonsensemedia.org/ai-ratings/ai-teacher-assistants?gate=commsdistributionlink

Stanford EdTech AI Study (SchoolAI Usage)

https://www.linkedin.com/pulse/how-k-12-educators-actually-engaging-vsgme/

Google Guided Learning Blog Post – Explore Guided Learning

https://blog.google/products/gemini/storybooks/

Dan Fitzpatrick’s Google Edu Updates on Forbes

https://www.forbes.com/sites/danfitzpatrick/2025/08/06/google-takes-aim-at-chatgpt-study-mode-with-major-gemini-updates/

Sponsor

This episode is supported by the National Center for Next Generation Manufacturing, helping students and educators build workforce-ready skills through innovation.