iiinova — where your digital self keeps evolving

loribalog.bsky.social @loribalog.bsky.social · 07/10/2026

It's funny, though: a dialogue with an AI can be ghostwritten by another AI. MIT's AI and education committee, reporting in August, pointed instructors toward oral exams and take-home work paired with in-class conversation. The defense in front of a human is the evidence.

100

loribalog.bsky.social @loribalog.bsky.social · 07/10/2026

I love this. HEPI's survey this March found that only 48% of UK undergrads say staff help them build AI skills, while 68% call those skills essential. A student with nobody showing them the learning-use is left with only the production-use.

000

loribalog.bsky.social @loribalog.bsky.social · 07/10/2026

I wish the order would run the other way: pilot the tool in a few districts with an independent evaluation, then publish the results, whether good or bad. Then a commissioner can promote what actually worked.

010

loribalog.bsky.social @loribalog.bsky.social · 07/10/2026

5% of high school seniors scored below Basic in math on the latest NAEP. Every one of them can ask a chatbot for a sales projection. Checking whether it adds up takes the algebra they never got. The NEW Common Core Every Student Now Needs blandiorsublime956330.substack.com/p/the-new-co...

blandiorsublime956330.substack.com

The NEW Common Core Every Student Now Needs

Five capabilities every graduate needs, whether she's headed to a university or a union apprenticeship. AI raised the bar on all five.

000

loribalog.bsky.social @loribalog.bsky.social · 01/10/2026

Totally. Grade the thinking, not the polished version. Start with a draft written in class and a short talk-through of the revisions, and your students will come out ahead.

010

loribalog.bsky.social @loribalog.bsky.social · 01/10/2026

YES!! Brookings landed in the same place in January, after a year of research across 50 countries, and its call was "AI tools that teach, not tell." The general chatbot hands over the explained solution, but AI designed to hint and keep learning with the student is helpful.

010

loribalog.bsky.social @loribalog.bsky.social · 29/09/2026

AI can write the first draft in seconds. Employers now pay for the human who can judge it, and schools are in a strong position to build that skill. Check out: Learning to Supervise the Machine blandiorsublime956330.substack.com/p/learning-t...

blandiorsublime956330.substack.com

Learning to Supervise the Machine

AI can write the first draft in seconds. Employers now pay for the human who can judge it, and schools are in a strong position to build that skill.

000

loribalog.bsky.social @loribalog.bsky.social · 29/09/2026

I love this, and the trainee problem starts long before medical school. A Public First survey of 4,021 young people in England, out September 22, found 47% would trust AI over a human to check whether something is true. A student who can't judge the output can only obey it. Build expertise first.

011

loribalog.bsky.social @loribalog.bsky.social · 29/09/2026

Here's a different way to look at it. A June 2026 CEPR study of 26,811 students found the exam losses sat with the 80% of AI users who outsourced homework; the AI users who actually did the work lost little on the test scores. The task design can prevent offloading.

010

loribalog.bsky.social @loribalog.bsky.social · 29/09/2026

Here's a different way to look at it: the same tool is augmentation for someone who has the skill and offloading for someone still building it. Adam Rodman at Harvard Medical School calls the second case never-skilling, in the NYT this week. A calculator helps an engineer and replaces a 7-year-old.

110

loribalog.bsky.social @loribalog.bsky.social · 24/09/2026

Market versus polis skips the question a policy has to answer: does the tool do the cognitive work, or does the student? I am a John Locke Fan- school is for preparing for future contributions to society; they need to know how to use the tool, but both sides lose if kids stop thinking.

020

loribalog.bsky.social @loribalog.bsky.social · 24/09/2026

A hints-only tutor does work: Bastani and colleagues found in 2025 it lifted practice 127 percent with no drop once removed. The unsolved problem is getting a student to choose the slower path.

000

loribalog.bsky.social @loribalog.bsky.social · 23/09/2026

The K-12 evidence is already specific. Bastani found in 2025 that high schoolers with open ChatGPT scored 48% higher in practice and 17% worse once it was removed, while a hints-only version caused no harm. So responsible use is a design decision made task by task, before it's a policy.

000

loribalog.bsky.social @loribalog.bsky.social · 23/09/2026

The fix is sequence: the student's own version first, the tool after.

100

loribalog.bsky.social @loribalog.bsky.social · 21/09/2026

Option 2, with one addition. Judging a model's output is a skill of its own, and students don't pick it up by accident. Breakstone tested 3,446 high schoolers in 2021, before chatbots: 96 percent couldn't tie a website to its funder. That judgment needs supervised practice with the tool in hand.

000

loribalog.bsky.social @loribalog.bsky.social · 21/09/2026

Another perspective to consider: the students didn't lose critical thinking to AI, they were never taught it. Breakstone tested 3,446 high schoolers in 2021, before chatbots: 96 percent couldn't tie a website to its funder. This is a teachable skill- we just need to prioritize it.

110

loribalog.bsky.social @loribalog.bsky.social · 21/09/2026

The same pattern showed up in the US. Bastani and colleagues in 2025: high school math students with ChatGPT scored 48 percent higher on practice and 17 percent worse than never-assisted peers once it was removed. A hints-only version didn't cause harm. The design of the tool was the decider.

010

loribalog.bsky.social @loribalog.bsky.social · 21/09/2026

There's no wrong way to cook. Same goes with how you use AI. For years, people asked me for my prompts, thinking that would get them my results. It never did. What actually works is knowing your style (how you cook) and using AI how it feels best.

100

loribalog.bsky.social @loribalog.bsky.social · 17/09/2026

Totally agree with your expertise point! Dell'Acqua's 2023 BCG study found this in adults: inside the model's competence, quality rose 40 percent; outside it, they were 19 points more likely to be wrong. A novice is always outside. If we grade the evaluation, then the incentive flips.

000

loribalog.bsky.social @loribalog.bsky.social · 16/09/2026

Sputnik, A Nation at Risk, No Child Left Behind, Common Core. Four big reforms, and every one of them was bolted onto a school frame built before 1910. Now schools are adding AI electives and detection tools, and the frame is trying to absorb them. blandiorsublime956330.substack.com/p/why-educat...

blandiorsublime956330.substack.com

Why Educational Reform Keeps Bouncing Off

Every big education initiative since Sputnik was bolted onto a frame built before 1910. AI is the first change too big for the frame to absorb.

000

loribalog.bsky.social @loribalog.bsky.social · 15/09/2026

Agreed the cheating frame is the wrong one, and I'd go one step past governance.- talk about how activities change. Have them draft unaided, consult the model, find two errors it made, revise, hand in the packet. Then we grade their evaluation. Getting to the top of bloom's taxonomy there!

010

loribalog.bsky.social @loribalog.bsky.social · 15/09/2026

'Instructed NOT to help students when they struggle' is productive failure with the second half cut off. Kapur's work only shows gains when the struggle is followed by instruction from someone who saw where it went wrong. Struggle alone, on an app, with no adult allowed to step in, is a stuck kid.

110

loribalog.bsky.social @loribalog.bsky.social · 15/09/2026

I feel that in my bones. The pace of change is out of control- the only thing that is remaining consistent is change. I am pro AI use, because they are going to encounter it in the workforce, and we need them to teach it ethically- but we can't just say "Go", there has to be some intention.

000

loribalog.bsky.social @loribalog.bsky.social · 11/09/2026

Building them means grading them. Breakstone and colleagues tested 3,446 high schoolers in 2021: 96 percent couldn't tie a climate site to its fossil-fuel funder. Lateral reading can be taught and assessed from Grade 6. The moratorium buys a year. Spend it writing that into the curriculum.

000

loribalog.bsky.social @loribalog.bsky.social · 11/09/2026

The brain part depends on who's holding it. Dell'Acqua's 2023 BCG study: inside the model's competence, consultants did 12 percent more tasks at higher quality; outside it they were 19 points more likely to be wrong. Undergrads are outside. Teach the work first, the tool second.

001

loribalog.bsky.social @loribalog.bsky.social · 09/09/2026

The problem is dropping tools into classrooms with no design behind them. K-12 still has to teach kids to judge machine output, though, because no one else will. I think it should be staged: nothing before grade 3, supervised critique in middle school, documented use in high school.

000

loribalog.bsky.social @loribalog.bsky.social · 04/09/2026

The median American job lasts 3.9 years. A third of the average job's skills turned over between 2021 and 2024. We still spend thirteen years giving kids a stock of knowledge meant to last forty. Something as to change: blandiorsublime956330.substack.com/p/we-prepare...

blandiorsublime956330.substack.com

We Prepared Kids for Jobs That Won't Hold Still

Thirteen years of front-loaded preparation was built to last across a forty-year career. The median job now lasts 3.9 years, and a third of its skills turn over in three.

000

loribalog.bsky.social @loribalog.bsky.social · 18/08/2026

A district gets federal recovery money, spends it on the best-evidenced intervention in education: tutoring 3x a week, trained adults, unanimous school board approval. By spring, the program is dead. Not because it failed. Because the master schedule, an artifact from 1906, had nowhere to put it.

100

loribalog.bsky.social @loribalog.bsky.social · 11/08/2026

Calling an AI system “an individual teacher for every student” frames it wrong. AI can provide practice, explanations, and feedback, but it can't KNOW the child, judge what's developmentally appropriate, or build trust. The teacher must remain the diagnostician, designer, and accountable adult.

000

loribalog.bsky.social @loribalog.bsky.social · 11/08/2026

You can't cram AI into the existing school model, but you can't turn your back on it because its in the workforce. You've got to prepare the next generation to use it ethically and wisely. It's time for school reform.

000

loribalog.bsky.social @loribalog.bsky.social · 11/08/2026

AI must not classify books, replace librarians, or decide what students may know. But that is an argument for AI literacy, not abstinence. Schools must teach students to challenge outputs, trace sources, expose bias, and retain accountability.

000

loribalog.bsky.social @loribalog.bsky.social · 09/08/2026

Every school you've walked into runs on the same bet: knowledge is scarce, so you go there to get it. That bet paid off for almost two hundred years. Then a teenager with a phone got a free, solid answer in seconds. Three past reforms kept the bet alive. This one kills it.

120