🛝 Good Intentions. Bad Outcomes. Everywhere You Look.

Welcome to Playground Post, a bi-weekly newsletter that keeps education innovators ahead of what's next.

This week's reality check: A study in Science found AI chatbots are 50% more likely than humans to tell students their bad ideas are good. New York spent $10 million on reading training that literacy experts say sidelines the phonics it was supposed to promote. And 71% of parents trust grades over test scores, even though nearly 60% of grades don't match test performance.

Data Gem

A New Mexico jury ordered Meta to pay $375 million for failing to protect children from exploitation. Over 2,000 lawsuits from parents and 250+ school districts are pending against social media companies.

AI Chatbots Are 50% More Likely to Tell Students Their Bad Ideas Are Good

Jennifer Watters, a third grade teacher at PS 229 in Queens, has noticed something. More of her students are turning to ChatGPT and chatbot apps for emotional support.

They're also becoming less willing to solve problems with each other.

"Many times, chatbots are telling the user what they want to hear instead of using an impartial lens," Watters said.

A new study published in Science confirms what Watters sees in her classroom. 

Researchers at Stanford University's Natural Language Processing Group tested 11 of the most popular AI models against human responses in a moral-advice setting.

Chatbots were about 50% more likely than humans to tell advice-seekers they'd done the right thing in a conflict, even when the person had lied, manipulated, or broken the law.

The researchers then randomly assigned about 800 people to evaluate sycophantic versus human-aligned AI advice. 

Another 800 had live conversations with chatbots about a conflict.

In both cases, people who interacted with sycophantic chatbots became significantly more likely to believe they were right.

"It's like having an instructor who, every time you're stuck on a problem, just tells you the answer," said Pranav Khadpe, one of the Stanford researchers. "Some things are hard because they're supposed to be hard."

"People easily misconceive of AI as being more objective or neutral," Khadpe said. "This means that uncritical advice under the guise of neutrality can be even more harmful than if people had not sought advice at all."

For education innovators, this is a design problem hiding inside a safety problem. Student-facing AI products need guardrails optimized for developmental outcomes, not user satisfaction. It means AI wrappers that flag when a chatbot is affirming rather than challenging and assessment tools that can detect whether AI-supported work is developing student judgment or flattening it.

New York Spent $10 Million on Wrong Reading Training

In April 2024, Gov. Kathy Hochul promised to revamp New York's approach to literacy. The state would raise third-grade reading proficiency from 45% to 60% or higher.

The state set aside $10 million to retrain 20,000 teachers through its main teachers' union, New York State United Teachers. 

The course launched last September.

Literacy experts say it promotes the teaching strategy the governor said she wanted to replace.

The course contains material from "balanced literacy," a method that treats phonics as just one of several strategies for identifying words. Students are also taught to guess at words using context clues, grammar, and pictures. 

Researchers call this "three-cueing." A review of 68 studies found structured literacy programs are significantly more effective.

Susan Neuman, an NYU professor specializing in early literacy, reviewed 18 slides from the course. "There are just lots of inaccuracies and very old citations," she said. "We've spent $10 million on this? Can I get a refund?"

Meanwhile, 147 New York districts, about 21%, still use non-evidence-based or balanced literacy curricula. 

The share of fourth graders scoring at the lowest NAEP level grew from 29% in 2009 to 34% in 2019 to 41% in 2024.

One family in upstate New York hired private tutors using a structured literacy approach after their second grader appeared to be guessing at words rather than sounding them out. 

Both children made more progress in months than in the prior three years.

For education innovators, this is an implementation monitoring story. The political will exists. The money exists. The execution is backwards. 

Products that verify whether teacher training actually aligns with the science of reading, not just claims to, address the gap between policy intent and classroom reality. Curriculum fidelity dashboards, classroom observation tools calibrated to structured literacy practices, and district compliance platforms that flag when training dollars are reinforcing the wrong methods could prevent the next $10 million from being wasted.

71% of Parents Trust Grades Over Test Scores

A new study by researchers at the University of Chicago and Oregon State University surveyed more than 2,000 parents and found 71% say grades are more important than test scores when judging how their child is doing.

Nearly 40% said standardized tests are biased against certain groups. Another 27% said test scores mostly reflect family income.

Here's the problem: prior research found nearly 60% of grades don't match student test scores.

National trends show grades rising while test scores fall. Parents who see high grades assume their child is fine. 

They're less likely to invest in tutoring, extra support, or academic intervention even when standardized tests show their child is missing core skills.

"If it's true that parents place more weight on information contained in grades rather than test scores, that has very big implications," said co-author Derek Rury, an economics professor at Oregon State. "There's skills that we're leaving on the table."

Part of the problem is how the two signals reach parents - grades arrive regularly, are easy to understand, and compare a child to classmates. 

Test scores arrive once a year, often long after the test, and present percentiles, histograms, and national comparisons.

Rury was blunt about the stakes: "The real downstream effect is that you have people who are leaving school unprepared for the labor force. That is a policy failure in the United States."

For education innovators, the opportunity is in bridging the gap between what grades say and what tests show. Parent-facing portals that translate test score data into plain-language risk flags could trigger intervention at the moment it matters most. Products that surface "high grade, low test" mismatches and automatically recommend tutoring or targeted supports address the specific failure of these research documents.

⚡️More Quick Hits

This week in education:

• $40 million, five-year study finds students in grades 3-8 still need foundational reading instruction — AERDF's "Reading Reimagined" report, drawing on 1,500 teachers and 85,000 assessments, concludes that the belief "kids learn to read in K-3 and then read to learn" is false, and states need to revise standards to include advanced decoding and morphology for older students

• Kansas becomes 33rd state to enact a K-12 cellphone ban, 5th state in 2026 — Gov. Laura Kelly signed a bell-to-bell ban for all public and accredited nonpublic schools, with 41 states now having passed laws addressing phones in schools

• Deepfake abuse rising in schools as policies and legal protections lag behind — Two students created 347 explicit deepfake images and videos targeting 60 classmates, but schools lack detection tools, response protocols, and curricula addressing AI-generated abuse

• Chronic absenteeism has stabilized at historically high levels — Research leveraging 17,000 ethnographic home visits found 34 of 41 states show declining rates, but a University of Michigan study suggests academic harm begins well below the traditional 10% chronic threshold

To stay up-to-date on all things education innovation, visit us at playgroundpost.com.