AI + Education Weekly Update
Week of October 20th, 2025
To catch up on another busy week in AI + Education news, check out our weekly updates for our review and analysis of some of the key highlights worth staying informed about.
Week of October 20th, 2025
School Districts AI Adoption Index
Starbridge's June 2025 index of thousands of U.S. school districts reveals that AI is moving from experimentation to implementation in K-12 education, with a clear divide between leaders deploying comprehensive strategies and districts still in planning phases.
Key Highlights:
68% of districts have purchased AI-related tools, 60% actively discuss AI in board meetings, and adopters typically use 3+ tools simultaneously.
Top performers like Peninsula 401 (WA) and Irvine Unified (CA) combine formal governance, dedicated AI committees, public resources, and multi-tool implementations.
MagicSchool, Brisk Teaching, Microsoft Copilot, Google Gemini, and ChatGPT dominate procurement across districts for lesson planning, personalized learning, and educator support.
The index does not distinguish between GenAI and AI, though tools like ChatGPT and Gemini suggest significant GenAI adoption.
International AI Safety Report First Key Update
The International AI Safety Report documents breakthrough AI performance on expert tasks across mathematics, software engineering, and specialized disciplines, but identifies emerging biological/cyber risks and monitoring challenges as systems become more autonomous.
Highlights:
AI systems now solve International Math Olympiad problems at gold-medal level, successfully complete over 60% of real-world software engineering tasks, and correctly answer 26% of expert-level questions spanning 100+ disciplines—up from under 5% in early 2024.
Major AI developers have implemented safeguards against chemical, biological, radiological, and nuclear (CBRN) threats in their latest models. The UK government projects AI will "almost certainly" amplify cybersecurity threats by 2027, with current systems already detecting 77% of security vulnerabilities.
While 51% of developers now use AI tools daily, broader labor market effects remain negligible. Additionally, performance gaps persist between benchmarks and realistic tasks.
Meta Introduces Parental Controls
Meta is rolling out enhanced parental controls for teen AI interactions, joining similar initiatives by OpenAI and other major platforms addressing concerns about AI's role in adolescent development and digital wellbeing.
Key Updates:
Recently implemented parental controls include: PG-13-rated AI chats, age-appropriate character access, blocking of romance and harmful content, parental visibility, and time limits down to 15 minutes daily.
Parents can turn off chats with AI characters.
Parents can block specific AI characters if they don’t want to turn off all access.
Parents will get insights into the topics their teens are talking about with AI characters and the Meta chatbot.
Gemma Model Helps Discover Cancer Therapy
Google and Yale's C2S-Scale 27B AI model predicted and experimentally validated a novel cancer treatment with no prior supporting literature, proving scaled AI can generate new scientific discoveries.
Key Highlights:
The research involved virtual drug screening validated through lab experiments on previously unseen human cells.
The model discovered that combining a protein called silmitasertib with low-dose interferon (a group of proteins used to treat viral infections) increases antigen presentation by 50%, making cold tumors (tumors that don’t elicit strong immune responses) more visible to the immune system.
This approach demonstrates that larger biological AI models can virtually screen thousands of compounds and generate testable hypotheses, potentially accelerating drug discovery.
What Is ChatGPT Go?
OpenAI introduced ChatGPT Go, a new low-cost subscription tier for users who want more than the free tier but don't need the advanced features of Plus/Pro.
Key Highlights:
Subscribers gain more usage of OpenAI's flagship model with automatic reasoning capabilities, plus enhanced image generation and file upload limits.
The tier includes advanced data analysis with Python, custom GPT creation for tailored applications, and projects/tasks for organized workflow management.
Offers core productivity features without advanced capabilities like legacy models (GPT-4o), Sora video generation, or deep research tools available in higher tiers
Currently available in select countries with plans for gradual global expansion.
