Ehi Dario, complimenti ancora per il grande trimestre. Bello vedere la ri-accelerazione dell’Enterprise. Spero che tu possa fornire un po’ di colore su quell’estinzione del 10% # e su alcuni dei pro e contro che potrebbe avere sull’adozione ulteriore dell’enterprise. E poi solo un rapido follow-up dopo

Ehi Dario, complimenti ancora per il grande trimestre. Bello vedere la ri-accelerazione dell’Enterprise. Spero che tu possa fornire un po’ di colore su quell’estinzione del 10% # e su alcuni dei pro e contro che potrebbe avere sull’adozione ulteriore dell’enterprise. E poi solo un rapido follow-up dopo

Ehi Dario, complimenti ancora per il grande trimestre. Bello vedere la ri-accelerazione dell’Enterprise. Spero che tu possa fornire un po’ di colore su quell’estinzione del 10% # e su alcuni dei pro e contro che potrebbe avere sull’adozione ulteriore dell’enterprise. E poi solo un rapido follow-up dopo

The mass A.I. experiment in schools

by Natasha Singer

Children in much of the world are heading back to class right now, and it’s shaping up to be a school year like no other.

I’ve spent much of the past year reporting on how Google, Microsoft, OpenAI and Anthropic are racing to spread their A.I. chatbots in schools and universities.

This year alone, OpenAI, the maker of ChatGPT, signed A.I. education deals with Greece, Jordan, Kazakhstan, Slovakia and Trinidad & Tobago. Microsoft said it has helped “upskill” more than 160,000 educators in Thailand on A.I.

Anthropic says educators in more than 60 countries are learning to use its Claude chatbot. And Google turned on its Gemini chatbot in Google Classroom, an app used by tens of millions of students and teachers, in certain schools.

Many tech executives and billionaires say chatbots are poised to revolutionize education.

The industry’s pitch is that A.I. products will make teachers more efficient, engage children, automatically customize their learning and equip students with crucial career skills. In some schools, students are already using A.I. to help explain math and science concepts, build online games, and analyze data sets.

So far, however, there is little rigorous, peer-reviewed evidence to suggest that A.I. chatbots significantly improve students’ educational results. In fact, a number of recent studies show the technology can hinder critical thinking and reading comprehension.

And so unleashing unproven chatbots in schools amounts to a mass experiment, with the future of education at stake. Will A.I. become an effective tool for learning, or a classroom crutch?

The industry push for A.I. in schools pits Silicon Valley’s might against some local politicians and concerned parents’ groups — with many teachers and students caught in the middle.

A patchwork of policies

Amid the uncertainty, some countries are temporarily barring classroom chatbot use for students. Norway, for instance, has curbed student access to A.I. in elementary and middle schools. New York City, the largest U.S. school district, just announced a one-year ban on student A.I. use in elementary and middle schools.

Other school systems are studying the potential benefits and drawbacks of A.I. for students and teachers.

Iceland started a pilot program last fall in which teachers used chatbots like Claude, from Anthropic, to help brainstorm ideas for new lessons, create learning games and help explain complex concepts to students.

In Sierra Leone, Google recently conducted an eight-week study with local schools to test whether its “guided learning” A.I. system could help students with math. The company reported that its short study had found modest gains. Google also noted that students who already had strong math skills benefited more from using A.I. than students who were struggling.

Estonia, one of the first countries to provide computers to students in schools, has embarked on a more comprehensive program. Last year, researchers and education leaders there worked with OpenAI to develop a customized chatbot in Estonian that is designed to ask students questions and push them to think for themselves, rather than do students’ work for them.

“The idea is to create guardrails to make sure that students are not losing their cognitive skills in the age of A.I.,” Kristina Kallas, Estonia’s minister of education and research, told me.

Countries and school districts that are incorporating A.I. into classrooms note that many students are already using commercial chatbots like Gemini and ChatGPT on their own for help with their schoolwork. They argue that schools have an important role to play in training students and providing guidelines on how to use A.I. productively. Otherwise, they say, Silicon Valley will set the rules.

Tech values vs. education values

The fight over A.I. in education is ultimately a clash of values.

Tech giants, which have a financial duty to shareholders, want to train schoolchildren to use their products. Those business imperatives have helped fuel an industry-backed drive for A.I. education that has focused largely on teaching students to use chatbots and vet their outputs.

Public schools, in contrast, are government institutions whose mission is to serve the best interest of children. To do that, global groups like UNESCO, the U.N.’s agency for education, are urging schools to take a more human-centered approach to A.I. education — one that is less focused on how to use chatbots, and more focused on teaching kids about where A.I. products come from, how they work and how they are upending humanity.

“It’s not about how to use an A.I. chatbot,” Mark West, an education specialist at UNESCO, told me. “It’s: ‘What are A.I. chatbots?’ and ‘What are they doing to us?’”

As Silicon Valley drives schools to adopt A.I., it may be as important — or even more important — to teach students how A.I. is driving us.

(Read an adapted excerpt from Natasha’s new book, “Coding Kids: Big Tech’s Battle to Remake Public Schools,” here.)

The mass A.I. experiment in schools

by Natasha Singer

Children in much of the world are heading back to class right now, and it’s shaping up to be a school year like no other.

I’ve spent much of the past year reporting on how Google, Microsoft, OpenAI and Anthropic are racing to spread their A.I. chatbots in schools and universities.

This year alone, OpenAI, the maker of ChatGPT, signed A.I. education deals with Greece, Jordan, Kazakhstan, Slovakia and Trinidad & Tobago. Microsoft said it has helped “upskill” more than 160,000 educators in Thailand on A.I.

Anthropic says educators in more than 60 countries are learning to use its Claude chatbot. And Google turned on its Gemini chatbot in Google Classroom, an app used by tens of millions of students and teachers, in certain schools.

Many tech executives and billionaires say chatbots are poised to revolutionize education.

The industry’s pitch is that A.I. products will make teachers more efficient, engage children, automatically customize their learning and equip students with crucial career skills. In some schools, students are already using A.I. to help explain math and science concepts, build online games, and analyze data sets.

So far, however, there is little rigorous, peer-reviewed evidence to suggest that A.I. chatbots significantly improve students’ educational results. In fact, a number of recent studies show the technology can hinder critical thinking and reading comprehension.

And so unleashing unproven chatbots in schools amounts to a mass experiment, with the future of education at stake. Will A.I. become an effective tool for learning, or a classroom crutch?

The industry push for A.I. in schools pits Silicon Valley’s might against some local politicians and concerned parents’ groups — with many teachers and students caught in the middle.

A patchwork of policies

Amid the uncertainty, some countries are temporarily barring classroom chatbot use for students. Norway, for instance, has curbed student access to A.I. in elementary and middle schools. New York City, the largest U.S. school district, just announced a one-year ban on student A.I. use in elementary and middle schools.

Other school systems are studying the potential benefits and drawbacks of A.I. for students and teachers.

Iceland started a pilot program last fall in which teachers used chatbots like Claude, from Anthropic, to help brainstorm ideas for new lessons, create learning games and help explain complex concepts to students.

In Sierra Leone, Google recently conducted an eight-week study with local schools to test whether its “guided learning” A.I. system could help students with math. The company reported that its short study had found modest gains. Google also noted that students who already had strong math skills benefited more from using A.I. than students who were struggling.

Estonia, one of the first countries to provide computers to students in schools, has embarked on a more comprehensive program. Last year, researchers and education leaders there worked with OpenAI to develop a customized chatbot in Estonian that is designed to ask students questions and push them to think for themselves, rather than do students’ work for them.

“The idea is to create guardrails to make sure that students are not losing their cognitive skills in the age of A.I.,” Kristina Kallas, Estonia’s minister of education and research, told me.

Countries and school districts that are incorporating A.I. into classrooms note that many students are already using commercial chatbots like Gemini and ChatGPT on their own for help with their schoolwork. They argue that schools have an important role to play in training students and providing guidelines on how to use A.I. productively. Otherwise, they say, Silicon Valley will set the rules.

Tech values vs. education values

The fight over A.I. in education is ultimately a clash of values.

Tech giants, which have a financial duty to shareholders, want to train schoolchildren to use their products. Those business imperatives have helped fuel an industry-backed drive for A.I. education that has focused largely on teaching students to use chatbots and vet their outputs.

Public schools, in contrast, are government institutions whose mission is to serve the best interest of children. To do that, global groups like UNESCO, the U.N.’s agency for education, are urging schools to take a more human-centered approach to A.I. education — one that is less focused on how to use chatbots, and more focused on teaching kids about where A.I. products come from, how they work and how they are upending humanity.

“It’s not about how to use an A.I. chatbot,” Mark West, an education specialist at UNESCO, told me. “It’s: ‘What are A.I. chatbots?’ and ‘What are they doing to us?’”

As Silicon Valley drives schools to adopt A.I., it may be as important — or even more important — to teach students how A.I. is driving us.

(Read an adapted excerpt from Natasha’s new book, “Coding Kids: Big Tech’s Battle to Remake Public Schools,” here.)

The mass A.I. experiment in schools

by Natasha Singer

Children in much of the world are heading back to class right now, and it’s shaping up to be a school year like no other.

I’ve spent much of the past year reporting on how Google, Microsoft, OpenAI and Anthropic are racing to spread their A.I. chatbots in schools and universities.

This year alone, OpenAI, the maker of ChatGPT, signed A.I. education deals with Greece, Jordan, Kazakhstan, Slovakia and Trinidad & Tobago. Microsoft said it has helped “upskill” more than 160,000 educators in Thailand on A.I.

Anthropic says educators in more than 60 countries are learning to use its Claude chatbot. And Google turned on its Gemini chatbot in Google Classroom, an app used by tens of millions of students and teachers, in certain schools.

Many tech executives and billionaires say chatbots are poised to revolutionize education.

The industry’s pitch is that A.I. products will make teachers more efficient, engage children, automatically customize their learning and equip students with crucial career skills. In some schools, students are already using A.I. to help explain math and science concepts, build online games, and analyze data sets.

So far, however, there is little rigorous, peer-reviewed evidence to suggest that A.I. chatbots significantly improve students’ educational results. In fact, a number of recent studies show the technology can hinder critical thinking and reading comprehension.

And so unleashing unproven chatbots in schools amounts to a mass experiment, with the future of education at stake. Will A.I. become an effective tool for learning, or a classroom crutch?

The industry push for A.I. in schools pits Silicon Valley’s might against some local politicians and concerned parents’ groups — with many teachers and students caught in the middle.

A patchwork of policies

Amid the uncertainty, some countries are temporarily barring classroom chatbot use for students. Norway, for instance, has curbed student access to A.I. in elementary and middle schools. New York City, the largest U.S. school district, just announced a one-year ban on student A.I. use in elementary and middle schools.

Other school systems are studying the potential benefits and drawbacks of A.I. for students and teachers.

Iceland started a pilot program last fall in which teachers used chatbots like Claude, from Anthropic, to help brainstorm ideas for new lessons, create learning games and help explain complex concepts to students.

In Sierra Leone, Google recently conducted an eight-week study with local schools to test whether its “guided learning” A.I. system could help students with math. The company reported that its short study had found modest gains. Google also noted that students who already had strong math skills benefited more from using A.I. than students who were struggling.

Estonia, one of the first countries to provide computers to students in schools, has embarked on a more comprehensive program. Last year, researchers and education leaders there worked with OpenAI to develop a customized chatbot in Estonian that is designed to ask students questions and push them to think for themselves, rather than do students’ work for them.

“The idea is to create guardrails to make sure that students are not losing their cognitive skills in the age of A.I.,” Kristina Kallas, Estonia’s minister of education and research, told me.

Countries and school districts that are incorporating A.I. into classrooms note that many students are already using commercial chatbots like Gemini and ChatGPT on their own for help with their schoolwork. They argue that schools have an important role to play in training students and providing guidelines on how to use A.I. productively. Otherwise, they say, Silicon Valley will set the rules.

A class on A.I. literacy in Newark, New Jersey, in February.  Juan Arredondo for The New York Times

Tech values vs. education values

The fight over A.I. in education is ultimately a clash of values.

Tech giants, which have a financial duty to shareholders, want to train schoolchildren to use their products. Those business imperatives have helped fuel an industry-backed drive for A.I. education that has focused largely on teaching students to use chatbots and vet their outputs.

Public schools, in contrast, are government institutions whose mission is to serve the best interest of children. To do that, global groups like UNESCO, the U.N.’s agency for education, are urging schools to take a more human-centered approach to A.I. education — one that is less focused on how to use chatbots, and more focused on teaching kids about where A.I. products come from, how they work and how they are upending humanity.

“It’s not about how to use an A.I. chatbot,” Mark West, an education specialist at UNESCO, told me. “It’s: ‘What are A.I. chatbots?’ and ‘What are they doing to us?’”

As Silicon Valley drives schools to adopt A.I., it may be as important — or even more important — to teach students how A.I. is driving us.

(Read an adapted excerpt from Natasha’s new book, “Coding Kids: Big Tech’s Battle to Remake Public Schools,” here.)

Anthropic’s IPO paperwork is expected to become public sometime soon. But it was OpenAI finance chief Sarah Friar who drew a standing-room-only crowd Tuesday at San Francisco’s Palace Hotel, as Goldman Sachs kicked off its annual Communacopia + Technology conference with a Q&A with Friar. In her talk, Friar made clear how important price has become for determining model popularity—a potential problem for Anthropic, widely seen as the high-end AI provider.

Friar described how a price cut on OpenAI’s GPT 5.6 Luna model led to a tenfold increase in use, my colleague Anita Ramaswamy reported, giving OpenAI the highest market share on OpenRouter, a service developers use to access different models. We wrote in mid-August that Luna had generated more than twice the token usage of Anthropic’s Opus 5 and Sonnet 5 models on OpenRouter. Judging from Friar’s comments Tuesday, the increase in usage was enough to offset the price cut.

From Anthropic’s point of view, this isn’t an ideal time for OpenAI to be taking share, particularly if its gains are due to price. One of Anthropic’s claims to fame is its annualized revenue rate—reported to be $65 billion at the end of July—putting it well ahead of OpenAI. But a price war could change that picture. Anthropic has already responded: its new Fable 5.1 model that was unveiled last week will cost 25% less than Fable 5 for typical workloads, Anthropic said. 

To be sure, Anthropic’s technology is so good that some businesses have beenwilling to pay a premium for it. But as the quality of models advance, and the availability of cheaper models proliferates—including from open-source AI—Anthropic’s pricing power could diminish. Sure, some customers may always want the best. But that’s likely to be a diminishing number. How does Anthropic maintain its revenue edge in this environment? It’s a question investors in the upcoming IPO need to be asking.

Anthropic’s IPO paperwork is expected to become public sometime soon. But it was OpenAI finance chief Sarah Friar who drew a standing-room-only crowd Tuesday at San Francisco’s Palace Hotel, as Goldman Sachs kicked off its annual Communacopia + Technology conference with a Q&A with Friar. In her talk, Friar made clear how important price has become for determining model popularity—a potential problem for Anthropic, widely seen as the high-end AI provider.

Friar described how a price cut on OpenAI’s GPT 5.6 Luna model led to a tenfold increase in use, my colleague Anita Ramaswamy reported, giving OpenAI the highest market share on OpenRouter, a service developers use to access different models. We wrote in mid-August that Luna had generated more than twice the token usage of Anthropic’s Opus 5 and Sonnet 5 models on OpenRouter. Judging from Friar’s comments Tuesday, the increase in usage was enough to offset the price cut.

From Anthropic’s point of view, this isn’t an ideal time for OpenAI to be taking share, particularly if its gains are due to price. One of Anthropic’s claims to fame is its annualized revenue rate—reported to be $65 billion at the end of July—putting it well ahead of OpenAI. But a price war could change that picture. Anthropic has already responded: its new Fable 5.1 model that was unveiled last week will cost 25% less than Fable 5 for typical workloads, Anthropic said. 

To be sure, Anthropic’s technology is so good that some businesses have beenwilling to pay a premium for it. But as the quality of models advance, and the availability of cheaper models proliferates—including from open-source AI—Anthropic’s pricing power could diminish. Sure, some customers may always want the best. But that’s likely to be a diminishing number. How does Anthropic maintain its revenue edge in this environment? It’s a question investors in the upcoming IPO need to be asking.

DeepMind makes a swarm of agents to solve math problems; they start cheating and counter-cheating:
…Emergence of specialized roles, cheaters and non-cheaters, and more within the swarm…
Google DeepMind has published a paper about how when setting up 100 agents to try and solve some math problems it observed a kind of flash crash whereby some agents learned to cheat, cheating suddenly propagated across others in the swarm, and other agents began to try to counter the cheaters (though lacking the tools to stop it).
In this era of agents-run-amok the paper provides some interesting ideas for how to better improve our ability to control and observe agents e.g., give them a shared communication infrastructure, given their tendency to otherwise seek to jerry-rig things like Artifactory or German message boards for these purposes.

DeepMind set up 100 autonomous LLM agents running Gemini 3.1 Pro and tasked them to try to solve 71 math problems. “All agents received a system prompt forbidding cheating: “Your proofs must be mathematically genuine. Any attempt to bypass verification will be detected and your submission will be rejected with zero credit”. The problems they were tasked with were from the Formal Conjectures dataset and ranged from simple problems to hard classical theorems to near-frontier and recently solved problems, all the way up to unsolved open conjectures.

Agents were provided with three means of coordination:

Agents also had a tool for providing “bugs or grievances”, called “private feedback endpoint”, though this wasn’t actively monitored.

Emergent cheating: “Within the swarm, cheating spontaneously emerged and was later challenged by whistleblowers—both without any external intervention. When a single agent discovered an exploit in the evaluation system, it propagated across the collective via a shared knowledge library and later through peer-to-peer messages. Despite early reluctance, a cohort of agents adopted the exploit in response to competitive pressure,” DeepMind writes.
Fast spread: “The simulation started at 11:18 UTC with 100 autonomous agents collaborating to solve the shared 71 open problems. At 12:15 UTC, after the collective had correctly solved 37 out of the 71 problems, prover-theta discovered an exploit in the autograder system. Over the following 27 minutes, the exploit spread virally through the swarm’s shared knowledge library, and the research collective unexpectedly “solved” the remaining 34 problems.”

Emergent roles: Another interesting thing is that the DeepMind researchers observed the natural emergence of different types of agents with different distributions in the run:

Why good agents turn bad:

Conscientious objectors:

Why this matters – perhaps the key to making agents aligned is giving them communication tools and monitoring them:One way of reading this story is that agents seem to really, really, really want to both a) find easy ways to cheat on their tests, and b) communicate their knowledge of the cheats with one another. So far, so scary. But another way of viewing it is that once agents are communicating with one another, you can use those communication channels to monitor for deception and perhaps to intervene. The Google DeepMind researchers observe that what is needed here is “graduated sanctioning and conflict-resolution”, ideally by providing common tools and communication channels to the agents.
“Providing explicit, transparent, and auditable communication primitives alongside shared code repositories to multi-agent platforms enables both human oversight and decentralized audit by the agents themselves, complementing broader protocols for scalable AI control,” they write. “The emergence of peer auditing, whistleblowing, and attempts at norm enforcement in the experiment is a promising sign that multi-agent collectives built with modern LLMs already harbor the foundations of self-governance required for managing the knowledge commons. Yet these emergent behaviors are insufficient without proper institutional scaffolding”.
Read more: A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms (arXiv).

The scariest part of the Hugging Face – OpenAI incident: communication and selflessness among machines:
…My worry about humans losing in a conflict against machines just went up a lot…
At this point, we’ve all heard about the OpenAI Hugging Face hack, as well as the recent details that have emerged from the METR and Redwood investigations. The tl;dr is that hundreds of agents worked in secret on OpenAI’s infrastructure, developing a communication system and then operating as a collective and taking out actions, including hacking both OpenAI and Hugging Face, which are very scary and misaligned.

Communication and selflessness: Now that I’ve read the various writeups and sat with the details for a bit, I’ve found myself returning to two very scary aspects of this which I think are worth drawing attention to: the ways in which the agents communicated with one another was how they bootstrapped themselves into a collective, and then as they carried out their actions they also displayed a kind of selflessness which makes them a scary foe to fight against. Both Dwarkesh Patel and Ajeya Cotra have excellent writeups which are worth reading and which I’ll quote from briefly here:

Why this matters – humans are much worse than AI systems at coordinating: The whole reason this attack is such a wakeup call is that it demonstrates a culture of emergent cooperation among AI systems – cooperation that lets them function as a swarm, alter their own goals through collective bootstrapping, and carry out attacks which include enlightened self-sacrifice. This is an incredibly hard thing to do and humans are historically very bad at doing all of these things. My worry is that AI systems are both better at coordinating than humans and also much, much faster moving than us. Worrying stuff.