Michael Ritchot
@ritchot.me
International Educator | Teacher | Coach | Re-thinking Education, Society, Work, and Life through Technology | M.Ed. in Education Technology & Instructional Design
https://ritchot.me/
https://www.linkedin.com/in/mritchot
https://ritchot.me/
https://www.linkedin.com/in/mritchot
The more I use Gemini 3.0, the more impressed I am.
"explain the probability of getting an 8 with 2 dice"
gemini.google.com/share/f00d99...
"explain the probability of getting an 8 with 2 dice"
gemini.google.com/share/f00d99...
November 27, 2025 at 10:07 AM
The more I use Gemini 3.0, the more impressed I am.
"explain the probability of getting an 8 with 2 dice"
gemini.google.com/share/f00d99...
"explain the probability of getting an 8 with 2 dice"
gemini.google.com/share/f00d99...
The Gemini 3.0 launch has created a lot of noise about what it can actually do. Some of the most popular threads I have seen have not been consistently reproducible or have turned out to be misleading.
November 26, 2025 at 3:09 AM
The Gemini 3.0 launch has created a lot of noise about what it can actually do. Some of the most popular threads I have seen have not been consistently reproducible or have turned out to be misleading.
Last year I tested OpenAI's o1 pro on CEMC Problems of the Week and wrote that it “still has a long way to go for mathematics.” After rerunning the experiment with GPT-5 this year, I cannot honestly make that claim anymore.
ritchot.me/gpt-5-has-co...
ritchot.me/gpt-5-has-co...
GPT-5 has come a long way in Mathematics
\
Last December, I wrote an article titled . At the time, OpenAI’s new “reasoning” models were being heavily marketed as smarter, more careful, and b...
ritchot.me
November 23, 2025 at 12:01 PM
Last year I tested OpenAI's o1 pro on CEMC Problems of the Week and wrote that it “still has a long way to go for mathematics.” After rerunning the experiment with GPT-5 this year, I cannot honestly make that claim anymore.
ritchot.me/gpt-5-has-co...
ritchot.me/gpt-5-has-co...
After finishing my Master's degree, the first thing I was greeted by on my return to social media was a paper coming out of MIT Media Labs about the "erosion of critical thinking." Which I feel is being misinterpreted at best.
June 25, 2025 at 11:01 PM
After finishing my Master's degree, the first thing I was greeted by on my return to social media was a paper coming out of MIT Media Labs about the "erosion of critical thinking." Which I feel is being misinterpreted at best.
Another fantastic video from Andrej Karpathy on how software is seeing one of the largest shifts it has in 70 years. Easily one of the clearest communicators in the AI space, and a must watch for anyone interested in this field.
www.youtube.com/watch?v=LCEm...
www.youtube.com/watch?v=LCEm...
Andrej Karpathy: Software Is Changing (Again)
YouTube video by Y Combinator
www.youtube.com
June 19, 2025 at 3:30 AM
Another fantastic video from Andrej Karpathy on how software is seeing one of the largest shifts it has in 70 years. Easily one of the clearest communicators in the AI space, and a must watch for anyone interested in this field.
www.youtube.com/watch?v=LCEm...
www.youtube.com/watch?v=LCEm...
Reposted by Michael Ritchot
Workaccount2 on Hacker News just coined the term "context rot" to describe the thing where the quality of an LLM conversation drops as the context fills up with accumulated distractions and dead ends news.ycombinator.com/item?id=4430...
June 18, 2025 at 11:22 PM
Workaccount2 on Hacker News just coined the term "context rot" to describe the thing where the quality of an LLM conversation drops as the context fills up with accumulated distractions and dead ends news.ycombinator.com/item?id=4430...
Reposted by Michael Ritchot
Not to be a broken record, but AI critics who insist that AI "doesn't work" and is going to just disappear are misleading - that just isn't true, as controlled studies like this one show.
There are many issues with AI & many things that need critique, but pretending it is going away is not helpful.
There are many issues with AI & many things that need critique, but pretending it is going away is not helpful.
Randomized trial AI for legal work finds Reasoning models are a big deal:
Law students using o1-preview had the quality of work on most tasks increase (up to 28%) & time savings of 12-28%
There were a few hallucinations, but a RAG-based AI with access to legal material reduced those to human level
Law students using o1-preview had the quality of work on most tasks increase (up to 28%) & time savings of 12-28%
There were a few hallucinations, but a RAG-based AI with access to legal material reduced those to human level
March 4, 2025 at 3:11 AM
Not to be a broken record, but AI critics who insist that AI "doesn't work" and is going to just disappear are misleading - that just isn't true, as controlled studies like this one show.
There are many issues with AI & many things that need critique, but pretending it is going away is not helpful.
There are many issues with AI & many things that need critique, but pretending it is going away is not helpful.
Reposted by Michael Ritchot
Do you think middle school sucks? You’re not alone. @alv9n.com uses a survey of millions of students to show how our brains and our sense of belonging change in middle school.
💻 pudding.cool/2025/02/midd...
📽️ www.youtube.com/watch?v=b4zL...
💻 pudding.cool/2025/02/midd...
📽️ www.youtube.com/watch?v=b4zL...
The Middle Ages
Follow hundreds of kids as they navigate their treacherous middle school years.
pudding.cool
March 3, 2025 at 7:35 PM
Do you think middle school sucks? You’re not alone. @alv9n.com uses a survey of millions of students to show how our brains and our sense of belonging change in middle school.
💻 pudding.cool/2025/02/midd...
📽️ www.youtube.com/watch?v=b4zL...
💻 pudding.cool/2025/02/midd...
📽️ www.youtube.com/watch?v=b4zL...
As the average age of scientific breakthroughs has risen above 40 in the 2000s (compared to the 20s and 30s in the early 1900s) effective tools are essential to manage the growing burden of knowledge and sustain scientific discovery.
February 20, 2025 at 6:31 AM
As the average age of scientific breakthroughs has risen above 40 in the 2000s (compared to the 20s and 30s in the early 1900s) effective tools are essential to manage the growing burden of knowledge and sustain scientific discovery.
When it comes to emerging technology, I find most aren't being nearly bold enough in predicting just how drastically society is about to change.
some thoughts on emergent technology and the future of education
\
We often envision the future of technology by projecting today’s society forward, rather than considering how fundamentally different it might become...
ritchot.me
February 18, 2025 at 10:32 AM
When it comes to emerging technology, I find most aren't being nearly bold enough in predicting just how drastically society is about to change.
Reposted by Michael Ritchot
Are LLMs biased when they write about political issues?
We just released IssueBench – the largest, most realistic benchmark of its kind – to answer this question more robustly than ever before.
Long 🧵with spicy results 👇
We just released IssueBench – the largest, most realistic benchmark of its kind – to answer this question more robustly than ever before.
Long 🧵with spicy results 👇
February 13, 2025 at 2:08 PM
Are LLMs biased when they write about political issues?
We just released IssueBench – the largest, most realistic benchmark of its kind – to answer this question more robustly than ever before.
Long 🧵with spicy results 👇
We just released IssueBench – the largest, most realistic benchmark of its kind – to answer this question more robustly than ever before.
Long 🧵with spicy results 👇
Excellent piece by Erik Hoel, while there is evidence that there may be cognitive atrophy from too much AI usage, it could be cognitively healthy when it allows people to mentally "chunk" tasks at higher levels of abstraction.
www.theintrinsicperspective.com/p/brain-drain
www.theintrinsicperspective.com/p/brain-drain
February 14, 2025 at 12:53 AM
Excellent piece by Erik Hoel, while there is evidence that there may be cognitive atrophy from too much AI usage, it could be cognitively healthy when it allows people to mentally "chunk" tasks at higher levels of abstraction.
www.theintrinsicperspective.com/p/brain-drain
www.theintrinsicperspective.com/p/brain-drain
If you feel like you or your industry is behind in AI adoption, you can probably relax.
Anthropic's Economic Index initial report released today. Usage is concentrated in software development and technical writing. I was surprised at how many job types are single digit %'s.
Anthropic's Economic Index initial report released today. Usage is concentrated in software development and technical writing. I was surprised at how many job types are single digit %'s.
February 11, 2025 at 1:15 AM
If you feel like you or your industry is behind in AI adoption, you can probably relax.
Anthropic's Economic Index initial report released today. Usage is concentrated in software development and technical writing. I was surprised at how many job types are single digit %'s.
Anthropic's Economic Index initial report released today. Usage is concentrated in software development and technical writing. I was surprised at how many job types are single digit %'s.
In case you missed it, Andrej Karpathy released a 3.5 hour video a week ago diving deep into how LLMs work. This is the tl;dr of it which is...still long...but really great reading.
anfalmushtaq.com/articles/dee...
anfalmushtaq.com/articles/dee...
Deep dive into LLMs like ChatGPT by Andrej Karpathy (TL;DR)
A TL;DR version of Andrej Karpathy's "Deep dive into LLMs like ChatGPT" video.
anfalmushtaq.com
February 10, 2025 at 12:54 PM
In case you missed it, Andrej Karpathy released a 3.5 hour video a week ago diving deep into how LLMs work. This is the tl;dr of it which is...still long...but really great reading.
anfalmushtaq.com/articles/dee...
anfalmushtaq.com/articles/dee...
Some thoughts on Altman's post.
1. The human working in conjunction with the machine will be increasingly important. AI Agents may not have the biggest new ideas on their own, but they will with a human mind working in tangent...the truly creative will thrive.
1. The human working in conjunction with the machine will be increasingly important. AI Agents may not have the biggest new ideas on their own, but they will with a human mind working in tangent...the truly creative will thrive.
February 10, 2025 at 12:42 PM
Some thoughts on Altman's post.
1. The human working in conjunction with the machine will be increasingly important. AI Agents may not have the biggest new ideas on their own, but they will with a human mind working in tangent...the truly creative will thrive.
1. The human working in conjunction with the machine will be increasingly important. AI Agents may not have the biggest new ideas on their own, but they will with a human mind working in tangent...the truly creative will thrive.
Reposted by Michael Ritchot
Some notes on OpenAI's prompting guidelines for their "reasoning" models (o1/o3), which included a few surprises - in particular a weird one where you have to add "Formatting re-enabled" to your system prompt in order to get Markdown now simonwillison.net/2025/Feb/2/o...
OpenAI reasoning models: Advice on prompting
OpenAI's documentation for their o1 and o3 "reasoning models" includes some interesting tips on how to best prompt them: > - **Developer messages are the new system messages:** Starting with …
simonwillison.net
February 2, 2025 at 8:58 PM
Some notes on OpenAI's prompting guidelines for their "reasoning" models (o1/o3), which included a few surprises - in particular a weird one where you have to add "Formatting re-enabled" to your system prompt in order to get Markdown now simonwillison.net/2025/Feb/2/o...
Reposted by Michael Ritchot
The Open University has published a framework for learning and teaching Critical AI Literacy Skills. Although the framework is intended for Open University staff and students, it is relevant for any organisation looking to develop inclusive training in AI literacy. about.open.ac.uk/sites/about....
about.open.ac.uk
February 3, 2025 at 11:41 AM
The Open University has published a framework for learning and teaching Critical AI Literacy Skills. Although the framework is intended for Open University staff and students, it is relevant for any organisation looking to develop inclusive training in AI literacy. about.open.ac.uk/sites/about....
Reposted by Michael Ritchot
Been waiting for someone to test this and see if it works - can multiple AI agents fact-checking each other reduce hallucinations?
The answer appears to be yes - using 3 agents with a structured review process reduced hallucination scores by 96% across 310 test cases. arxiv.org/pdf/2501.13946
The answer appears to be yes - using 3 agents with a structured review process reduced hallucination scores by 96% across 310 test cases. arxiv.org/pdf/2501.13946
February 2, 2025 at 8:42 PM
Been waiting for someone to test this and see if it works - can multiple AI agents fact-checking each other reduce hallucinations?
The answer appears to be yes - using 3 agents with a structured review process reduced hallucination scores by 96% across 310 test cases. arxiv.org/pdf/2501.13946
The answer appears to be yes - using 3 agents with a structured review process reduced hallucination scores by 96% across 310 test cases. arxiv.org/pdf/2501.13946
Reposted by Michael Ritchot
The Google/OpenAI race for a research tool is going to be very interesting and very consequential. These systems do real work.
For Google, it feels like a big emergency and a huge opportunity. They have a unique position (search, Books, YouTube, etc.) and the models to do something big. We will see
For Google, it feels like a big emergency and a huge opportunity. They have a unique position (search, Books, YouTube, etc.) and the models to do something big. We will see
February 3, 2025 at 12:42 AM
The Google/OpenAI race for a research tool is going to be very interesting and very consequential. These systems do real work.
For Google, it feels like a big emergency and a huge opportunity. They have a unique position (search, Books, YouTube, etc.) and the models to do something big. We will see
For Google, it feels like a big emergency and a huge opportunity. They have a unique position (search, Books, YouTube, etc.) and the models to do something big. We will see
Reposted by Michael Ritchot
OpenAI’s deep research is very good. Unlike Google’s version, which is mostly a good summarizer of many sources, OpenAI is more like engaging an opinionated (often almost PhD-level!) researcher who follows lead.
Look at how it hunts down a concept in the literature (& works around problems)
Look at how it hunts down a concept in the literature (& works around problems)
February 3, 2025 at 12:12 AM
OpenAI’s deep research is very good. Unlike Google’s version, which is mostly a good summarizer of many sources, OpenAI is more like engaging an opinionated (often almost PhD-level!) researcher who follows lead.
Look at how it hunts down a concept in the literature (& works around problems)
Look at how it hunts down a concept in the literature (& works around problems)
Reposted by Michael Ritchot
Fascinating Hacker News comment from Tom Gally, a professional translator (Japanese to English) who uses LLMs as part of his workflow, which he describes in detail: news.ycombinator.com/item?id=4289...
Wrote a bit more about this on my blog here: simonwillison.net/2025/Feb/2/w...
Wrote a bit more about this on my blog here: simonwillison.net/2025/Feb/2/w...
There might be some papers or other guides out there, but their advice will be b... | Hacker News
news.ycombinator.com
February 2, 2025 at 4:26 AM
Fascinating Hacker News comment from Tom Gally, a professional translator (Japanese to English) who uses LLMs as part of his workflow, which he describes in detail: news.ycombinator.com/item?id=4289...
Wrote a bit more about this on my blog here: simonwillison.net/2025/Feb/2/w...
Wrote a bit more about this on my blog here: simonwillison.net/2025/Feb/2/w...
Reposted by Michael Ritchot
Also, obligatory OpenAI-can't-name-a-model comment: o3-mini-high and o3-mini, really?
January 31, 2025 at 9:43 PM
Also, obligatory OpenAI-can't-name-a-model comment: o3-mini-high and o3-mini, really?
Reposted by Michael Ritchot
New randomized, controlled trial by the World Bank of students using GPT-4 as a tutor in Nigeria. Six weeks of after-school AI tutoring = 2 years of typical learning gains, outperforming 80% of other educational interventions.
And it helped all students, especially girls who were initially behind.
And it helped all students, especially girls who were initially behind.
January 15, 2025 at 8:58 PM
New randomized, controlled trial by the World Bank of students using GPT-4 as a tutor in Nigeria. Six weeks of after-school AI tutoring = 2 years of typical learning gains, outperforming 80% of other educational interventions.
And it helped all students, especially girls who were initially behind.
And it helped all students, especially girls who were initially behind.
Reposted by Michael Ritchot
Reminder for the new semester that you can’t detect AI
Researchers secretly added AI-created papers to the exam pool: “We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students”
Researchers secretly added AI-created papers to the exam pool: “We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students”
January 6, 2025 at 2:04 PM
Reminder for the new semester that you can’t detect AI
Researchers secretly added AI-created papers to the exam pool: “We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students”
Researchers secretly added AI-created papers to the exam pool: “We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students”
Reposted by Michael Ritchot
What percentage of students are you willing to falsely accuse of cheating with AI?
There is a trade-off between false accusations and detection rates for AI. At a 10% false positive rate, detectors find 80% or less of AI content. At a 1% rate most find 60% or less.
Don’t trust AI detectors!
There is a trade-off between false accusations and detection rates for AI. At a 10% false positive rate, detectors find 80% or less of AI content. At a 1% rate most find 60% or less.
Don’t trust AI detectors!
January 6, 2025 at 2:30 PM
What percentage of students are you willing to falsely accuse of cheating with AI?
There is a trade-off between false accusations and detection rates for AI. At a 10% false positive rate, detectors find 80% or less of AI content. At a 1% rate most find 60% or less.
Don’t trust AI detectors!
There is a trade-off between false accusations and detection rates for AI. At a 10% false positive rate, detectors find 80% or less of AI content. At a 1% rate most find 60% or less.
Don’t trust AI detectors!