Conversation
No working paper yet, but the results and experiment are written up here. They used Microsoft Copilot and teachers provided guidance and initial prompts:
To make clear the caveats for people who don't read the post: learning gains are measured in Equivalent Years of Schooling, this is a pilot study on narrow topics and they do not have long-term learning measures. And there is no full paper yet (but the team is credible)
And since it isn't clear to everyone who doesn't read the blog post - the fact that this is teacher-led is likely very important. We know that independent use of AI as a tutor can harm learning in some circumstances, because it gives the illusion of learning.
No one reads the thread and now I see viral quotes of the first post in the thread saying “teachers are going away”
That is the opposite of the message of every RCT we have in AI & education so far! You really need instructor guidance for today’s LLMs to be useful for learning.
I wrote more about when AI helps and hurts education here.
There is a new paper from Wharton (strange that you haven't seen it as you are a professor there) that says the opposite:
Yes, that is why it is important teachers are involved. I discuss that paper at length in the post at the end of the thread.
Start your child's independent homeschooling journey with our flexible, self-paced platform. We offer a full PreK-12 curriculum with electives & hundreds of engaging video-based lessons taught by experienced educators! Enroll now!
No Contracts. Cancel Anytime.
that is an impossibly huge effect size for a six-week intervention in education. Not "AI can't be that good" but rather "nothing is that good."
This is exactly why I'm so passionate about AI—opening up access and opportunity for people. Even more astounding to read that many of these students had never even used a computer before this pilot. Gets me so excited for what's possible, for everyone.
big [causal intervention] if significant [Bayesianly]
While being a tutor for courses, we always ask them first "Have you asked GPT/Claude/Gemini about it?" much like how we used to ask "Did you google it?".
Hope they learned the lesson, and not "lazy tutors...".
I'm rolling out LLM tutors (that students co-design) for my university class
—but does anyone have thoughts on how to implement it as a randomised controlled trial?
In terms of fairness and ethics, it seems harsh to withhold the treatment from half the students
throttling AI-led education and medicine is a direct trade off of child and patient outcomes, on behalf of teaching and healthcare jobs.
It seems likely that foundation models solve tutoring much faster than "AI tutor" startups solve distribution.
This will be attacked (unfortunately) by the usual "if not perfect then should not be done" brigade.
I will never comprehend where that hubris comes from. We all had terrible teaching all through childhood. Most teaching is bad—that's how we all remember that one good teacher of
Show more
When I first started with ChatGPT, I personally encountered the potential of a "private" tutor. It's honestly guided learning:
- Ask what you don't understand
- Ask what interests you most
- Skip parts that are obvious to you
- Learn efficiently and purposefully
What's also not known is how helpful is this method beyond learning (in) English language.
Also, data on age groups of the students is important data too.
0-3, 4-6, 7-9, 10-12, 13-15, 16-18 ... Each of these groups have different learning & development needs.
AI as a tool for empowering students to surpass traditional limitations, with teachers evolving into mentors who guide this self-directed, personalized learning journey.
Want to learn how practical AI skills and automations for your business and work?
Check out our 50+ step-by-step video tutorials 100% FREE
20+ hours of Ai and Automation goodness absolutely free

Have valid scientific methods been applied to demonstrate these incredible gains?
Did the authors control for confounding factors, e.g. student self-selection or differing rates of Internet access?
Was the study sponsored by anyone with an interest in publicising certain results?
As homeschool quickly becomes the clearly superior option, how long before it is the norm?
AI in education has transformative potential: closing gaps and personalizing learning. But its success depends on overcoming challenges like equitable access, privacy, and adaptation to local contexts. A huge opportunity if we can address them. #AI #Education #Innovation
This is a big deal. For me, I just could never quite grasp calculus the way it was taught. Learned it from ChatGPT at the dog park the other day and loved it.
This is absolutely fantastic.
I would have loved to have access to AI in the former USSR when I was a kid.
Education has been bad value for money for my entire lifetime
Most of my teachers weren't much
Do you know what the control group was? Didn’t see it in the blog posts.
If it was “tutoring without AI” or “no tutoring” those feel like very different results
This is the future! My son is 3, I'm am so very optimistic for his future, what he'll be able to learn and do with the help of AI (my wife isn't on board yet, working on it, she'll come around)
Having a PHD level tutor, for ANYTHING kids want to learn, that will also learn how
Show more
Still waiting for a specialized AI tutor product. Who is building the cursor of education?
Did control condition students also do after school program without AI? The after school program itself could be a big impact factor , other tech solutions have shown significant gains with after school program years ago
If you're not learning AI in 2024, you're falling behind.
Join 500,000+ readers and learn how to use AI in just 5 minutes a day (for free).
This study (early release) shows why the rollout of technologies such as solar powered Starlink terminals in remote areas of Africa and South America is so important.
There are brilliant minds in these places that don't have the ability to break out beyond what their teacher can
Show more
Relying solely on AI for education overlooks critical thinking and interpersonal skills development. Teaching should be about more than just test scores; it's about promoting holistic growth in students.
I absolutely would have loved having gpt-4 in college. Would have done a much more demanding and intensive curriculum
This is really interesting. AI has the potential to flip education on its head and as seen in this study have a major impact on students learning.
this is exactly the kind of ai impact we should be celebrating... leveling the playing field in places that need it most
AI tutoring is incredible. Kids are going to be incredible thanks to tools like
If you're not learning AI in 2024, you're falling behind.
Join 500,000+ readers and learn how to use AI in just 5 minutes a day (for free).
Imagine how many teachers right now are screaming at students for using AI
They have no clue what's coming next
Showing my wife the "ask a question" feature in notebookLM was a huge eye-opener for her. She immediately understood its potential for tutoring, which none of my AI impact ramblings could convey.
These "ai" models have been sitting here with educational potential and I've been using them for the occasional question and mostly meh generative art that I ultimately don't use for anything (though some actually came out good). I gotta up my game.
AI will change our learning infrastructure for the better. Now, we just need stakeholders to unite on researching, designing and building a new model. A pilot program vastly different than the status quo. Not reform. A complete, modern overhaul.
If true, this is actually insane.
2 years of learning on 6 weeks is absolutly beyond anything we have created so far
How many hours of AI tutoring a day is the study based on?
Current 9-5 Workers..
It's time to make a move.
Our Accelerator 2.0 program shows those with zero prior experience how to find profitbable niches and launch WINNING products under their own brand on Amazon.
Click the link below to see what our
12 Month Mentorship includes

Slide 1 of 5 - Carousel
My sons would benefit from working with AI as a math tutor. We’ve already dabbled with it.
Was thinking that these models would be terrific if embedded in text books. They can explain parts you don't understand, break everything down to it component steps, etc... Could make example problems tailored to your understanding.
Not surprised. Anyone can pretty much learn anything (at least the cognitive part) now, only limited by their effort, genetic potential, an internet connection and a decent computer. Traditional schools and colleges/universities are functionally obsolete. The outliers would be
Show more
Invigorating Education with LLMs
interesting result, but many such studies have novelty effect.
hope this delta continues for the long term
Do we know sample size? Can’t find it in that blog
Promising though. And makes perfect sense. Exciting!
Man I want to read the paper. That’s awesome and bodes well for the future!
Honestly, this is incredible. I would love to know what they taught or how their program provided the tutoring to these students.
"outperforming 80% of other educational interventions"
So many questions.
Have we quantified every possible educational intervention? How could you possibly determine this?
and what are the best educational interventions?? Forget this top 20% garbage lets use those
Certainly one of the best use case possible for LLMs during the next years, at least for serious students. The capability those tools have for patience & clear answers simply cannot be matched by any professors.
It certainly won't be used by our westerners school, but those who
Show more
This is fascinating.
Speaks volumes about teacher-led-AI-education, which I believe will make teachers the rockstars of the next decade, and AI the tutor we always needed.
The education franchise, is how I imagine it!
Thanks for sharing
GPT university is extremely bullish for global GDP growth and technological progress
Great share and great to see the data speaking so clearly.
Advocate a complete change to our education model, schools are an archaic setup designed initially as daycare for the children of factory workers - not to maximize student learning or human potential.
Online schools failed before because people need a real personality to connect to. AI addresses this problem. If it really works, we can start deploying education in all corners of the world.
Fascinating.
Attend School Online With Acellus Gold, A Revolutionary Platform That Accelerates Learning.
And this is just with gpt4, imagine when they can get a voice model and more sota models
This remarkable study underscores not only the transformative potential of AI in bridging educational gaps, particularly for marginalized groups, but also highlights the critical need for equitable access to such technologies to ensure sustainable progress in global education.
Yerp, was pretty fucking obvious this would be the outcome. Ai will revolutionise teaching. We do need some way to teach kids the soft skills ie social interaction with other humans, and places you make friends. I guess teaching could become more like glorified babysitting with
Show more
Who knows someone living in Las Vegas or Nevada? 
That state dishes out $8,000 per grant for each student and processes applicants faster than you can say "Jimmy John's" – under 2 weeks! 
I can help them enroll 8-10 students per month.
You do the math – that's some
Show more
I got help when I read Cosmos. Since I did not study science hard enough in school there were a lot of concepts I did not understand; Chatgpt helped mw understand them.
Thank you for sharing! We will get the working paper out soon cc
Incredible findings. Always feel like the school work my kids get is 2 years behind what they’re actually capable of.