← AI for Educators Daily with Dan Fitzpatrick
Loading summary
Transcript1 lines
- [00:00]
A
If this episode makes you think, please let us know in the comments and support us by subscribing and leaving a review. Thank you. Today we are exploring a fascinating analysis of the sheer pace of AI acceleration we've seen in just one week, drawing from a recent expert commentary that really captures the feeling that the individual stories are adding up to something much more than the sum of their parts. It's that sense that we can feel the acceleration all around us, not just intellectually recognize it. Now, this analysis touches on acceleration in so many different contexts, how AI models are developing, the policies being shaped, the business models evolving, and more. But I want to start where the analysis does, with profitability acceleration and a corresponding shift in market sensibility One of the really big stories this week is that Anthropic, for the first time ever expects to have a profitable quarter. And while the analysis points out some caveats, it's projections, not realized revenue yet, questions around how they recognize revenue, and some discounted access to compute through partners like SpaceX. For most people, these are just quibbles. The bigger picture, the reset of expectations, is around just how much money these AI labs can actually make. Think back to the bubble narrative from late last year and early this year. It was all about concerns that we'd overbuild compute infrastructure, or that these big labs would never be able to serve tokens profitably. But the fact that Anthropic is on the cusp of doing so even with a few strategic boosts, rather than really impacts how people think about the acceleration of business model development in the AI sector. And it's not just anthropic. OpenAI had a fantastic first quarter too, generating a billion dollars more in revenue than Anthropic, though Anthropic's revenue acceleration has since outpaced them. Even Nvidia, the chip maker, had a massive quarter, blowing past analyst expectations to the point where the market isn't quite sure how to value them anymore. This suggests the market is really gearing up for full send mode when it comes to AI. So why does this matter for schools? Well, this profitability acceleration and the shift in market expectations deeply corresponds with another acceleration, a move away from one pricing paradigm to another. For a while, many of us, including me, talked about the subsidy era of AI that's ending and we're entering what some call the trade off era. What's happening is that the introduction of more sophisticated, token hungry AI agents means flat rate plans that subsidize the most active users are just not economically viable anymore. We saw this play out with Google. Their price cut on the ultra plan from $250 to $200 a month sounded great on the surface, but when you dug into it, it came with a shift to usage based billing for certain token hungry use cases. It's not dissimilar to what Anthropic did recently. The analysis suggests that token based pricing is forcing every enterprise customer, and that includes schools, to really confront the actual cost of running these models at scale. And it turns out that number can be far higher than the flat rate experiment suggested for a school leader or a department head. This shift in pricing is a significant practical concern when you're trying to roll out AI tools across your school, your budget is finite. This change means you can't just set it and forget it. You need to really understand the purpose behind your AI use. Are you getting value for those tokens? Are you using them efficiently? This is where my start with why not how? Philosophy becomes incredibly important. You're forced to evaluate to audit your tool usage and really understand what learning goal each AI interaction serves, rather than just chasing the latest shiny tech. It also makes a strong case for exploring more efficient models, which some companies like Cursor are now introducing at significantly lower costs and performing at comparable levels to the big players. More efficient models mean you can get more bang for your buck, stretching your budget further to serve more students. Now, alongside these shifts in business models and pricing, we're also seeing an acceleration in getting access to more compute power. The analysis highlighted Elon Musk's SpaceX moving fully into a shifted role as an AI computer. Offer an AI compute as a service at significant scale. This isn't just talk. Anthropic is expanding its partnership with SpaceX, scaling up on new data centers. While this might sound like something happening far away in the corporate stratosphere, it has future implications for us in education. More widespread and potentially cheaper access to powerful compute means the AI tools available to schools are only going to get better, faster and more sophisticated. It could enable truly personalized learning paths, incredibly rich simulations, or even AI tutors that are far more capable than what we see today. The infrastructure is being built and that underpins everything. Let's move on to another area of serious acceleration. AI's consumer surfaces. Even with some mixed messaging at Google I O, one thing was crystal clear. AI is now pervasive across the Google ecosystem and people are using it. The Gemini App now has 900 million monthly active users, essentially closing the gap with ChatGPT and the growth and overall tokens Processed each month jumped a staggering 700%. What's truly significant here, and this is where it really hits home for educators, is is how Google is integrating AI into its existing experiences, especially search. Soon Google search will feature not just AI's information consolidation capability, but also its agentic capacity. What does this mean? It means you'll be able to create and manage multiple AI agents right within search. These agents will intelligently look across the entire web, including blogs, news, social posts, real time data, and then send you an intelligent synthesized update with links at the right moment to help you take action. Think about your students for a moment. Historically, search has been a one time query. What's available right now? The analysis uses the example of someone looking for an apartment. You search, you get results, and if nothing fits, you close the browser. But with these new persistent agents, that search becomes ongoing. You can ask Google to keep you updated when apartments that meet your criteria become available. For our students, this is transformational. It completely reshapes how they'll conduct research and gather information. We're not just teaching them to find information, we're teaching them to manage an ongoing stream of synthesized information. This demands a whole new level of AI literacy, which as I often say, isn't about technical skill, but collaborative reasoning ability. Students will need to think with AI, not just use tools. They'll need to understand how to manage these agent conversations with precision, critically evaluate the synthesized updates and understand the limitations of these information agents. The real value is not in what the machine produces, but but in how the student responds to an acts upon that persistent curated information. We're teaching them not to outsmart machines, but to outthink them to apply their human judgment and creativity to what the agents deliver. Google is also taking this principle into other areas like Docs Live, where users can dictate prompts and then edit to create full documents using just their voice plus AI. This is a clear acceleration towards a voice first live interaction pattern. Imagine a year 7 student with dysgraphia, or even just any student being able to articulate their ideas verbally and have AI draft the document, allowing them to focus purely on refining their thoughts and arguments. This is outsourcing the doing, not the thinking. It frees up cognitive load from the mechanics of transcription and formatting to so the student can pour their capacity into creativity, structure and critical judgment. Okay, so we've talked about business models, compute and consumer services accelerating, but this week also brought us model capability acceleration in a really striking way. Even if it's for a model, we don't have access to Yet OpenAI announced a breakthrough in an 80 year old mathematics problem first posed by Paul Erdos in 1946. The problem asks how many pairs of points can be exactly one unit apart if you place ne shown points on a plane. The prevailing view for decades was that a square grid arrangement was optimal. But this week an internal model at OpenAI disproved this conjecture, leveraging multi dimensional math flattened into 2D, producing more pairs than the grid. What's so significant about this, according to Fields medalist Tim Gowers, who wrote a companion paper, is that it's the first really clear example of AI solving not just an unsolved math problem, but a really well known horrid one. And here's the kicker. OpenAI researcher Noam Brown said it was just a general purpose LLM with no specific training for this problem or mathematics. The prompt wasn't even particularly tricky. It was largely just a clear statement of the problem using appropriate mathematical terminology. This is huge. It really makes you pause and think about what thinkin means. OpenAI's Alexander Way summed up the feeling of acceleration, writing that 10 months ago he was ecstatic that AI could win International Math Olympiad gold. But 10 today that excitement feels quaint. He also wrote that math is a leading indicator of what is to come. Soon, perhaps sooner than we all think, AI will begin autonomously producing landmark results in CS, physics, econ, bio, etc. We should be prepared for a new world where the nature and methods of science will have changed. So what does an AI solving an 80 year old math problem mean for your year 10 maths class? It profoundly reinforces the idea that we need to design learning that cannot be faked because it demands depth, care and imagination. Traditional assessments that rely on recall or simple problem solving are increasingly vulnerable. We need to shift towards what I call the three P's of assessment. Looking at the product, yes, but also the process, including how students interact with AI and their performance through live demonstrations. And we need to design tasks for cognitive stretch, asking if AI could complete this without the students unique context, perspective or judgment. This isn't about replacing human intelligence. It's about enhancement, augmenting human capability so we can tackle even more complex, previously unsolvable challenges. Machines can compute, they cannot wonder, they cannot care. These remain our uniquely human domains and our curriculum needs to nurture them. This idea of AI autonomously advancing was also at the heart of another big model development acceleration. Former OpenAI co founder Andre Karpathy announcing his return to the playing field, this time with anthropic to work on what many call recursive self improvement, or rsi, essentially using AI to accelerate pre training research itself. This means the pace of AI development is only going to get faster. As educators, this calls for us to embrace an evolution, not revolution mindset, understanding that we're always adapting, always learning. Now, on a slightly lighter but important note, the discussion around the Erdos problem led to some interesting facts about how much energy an AI uses. Ethan Mollick highlighted that solving the problem took a tiny amount of electricity and water, less than three almonds worth of water. This reflects an accelerating counter narrative about data centers. On one hand, opposition to data centers is growing, but on the other, facts are emerging to counter some of the bigger critiques. For example, data centers use less than a fifth of the water used on golf courses annually. This is crucial for education leaders. When we introduce AI into schools, we often encounter resistance rooted in fear or misinformation, not just about the technology itself, but its wider impact. The conversation about data centers is a great parallel. It's better when it's based on real facts and evidence, not just fear of AI or big tech. This is about being transparent and proactive in addressing concerns, framing change in terms of what stakeholders value, whether that's purpose, adventure, or knowledge. Finally, on the policy front, we saw both acceleration and a strange pause. This week, California Governor Gavin Newsom signed an executive order aimed at preparing workers for potential AI labor disruption. It's exploratory, directing state agencies to work with academics and labor groups to develop new policies and track early warning signs. While some have raised practical questions about the difficulty of measuring AI's impact on employment, and others see it as political show. But it's still significant. For the first time, a governor of a major state is directly tackling the potential implications of AI on the workforce for schools. This underscores our role in preparing students for a future where adaptability and AI literacy are paramount. It's about teaching students skills that complement AI, fostering creativity and critical thinking, ensuring they can thrive in an evolving job market. Meanwhile, at the federal level, an anticipated AI executive order from the current president was scuttled just hours before a planned signing ceremony. A draft had circulated proposing things like safety, benchmarking standards, and requiring AI companies to submit models to the government ahead of public release. But the order was postponed due to concerns about it getting in the way of innovation, particularly in the air race with China. This policy back and forth, this moment of acceleration followed by a pause, highlights the volatile landscape for school leaders. It means we can't afford to wait for perfect clarity from the top. We need to act now, within our own context, building from strengths, identifying early adopters, and aligning our AI strategies to our specific institutional needs. It's about empowering teachers as change agents, giving them the time and space to explore, and hankering AI to existing friction points in their workflow, not just novelty Ultimately, what this expert analysis conveys is that the acceleration isn't just in the news itself, but in the feeling that the implications of the news create a regular AI model is solving 80 year old math problems. Companies are defying profitability expectations, leading researchers are moving to accelerate AI development itself. And everywhere the jockeying for narrative and positioning is getting more dramatic. Demis Hassabis, Google DeepMind CEO, recently framed where we are as a beginning, not an end. He argued that these developments would help unlock AGI's incredible potential for the benefit of the entire world, suggesting we're standing in the foothills of the Singularity. He sees this technology as a force multiplier for human ingenuity, ushering in a new golden age of scientific discovery and progress. It's a powerful, optimistic message, and as educators, it's our job to ensure that vision becomes a reality by preparing our students not just to use these tools, but to shape this new world with wonder, care, and human judgment. That's all for today. Thanks for listening.