The New York TImes: What’s one issue in politics that you’ve changed your mind about? Zohran Mamdani: The role of the private market in housing construction.

Introducing Sherpa: the most advanced fiction writing AI We accelerated from $250M in ARR to $500M because Sherpa helped increase content production by 1200% in 1 year Sherpa was trained on 5.5B hours of playtime with minute by minute dynamic retention data. 550K+ creators have produced 2.6M hours of content annualised using it Pocket FM is like Netflix for audio-only dramas, with our own pool of one-person studios. 10% of eligible writers on Pocket FM make >$200K One blockbuster produced >$100M in revenue 3 writers have become millionaires in <2 yrs We built Sherpa to enable anyone to make >$1M by writing world-class fiction stories: 1. The Idea: Drop a 1-2 sentence concept. Sherpa interrogates it like a veteran editor on tension, stakes, and psychology 2. World & Characters: It builds out the complete lore, tone, and character psychologies 3. Sub-Plot planning: Breaks the premise into arcs, arcs into episodes, and episodes into scenes 4. Scene-by-Scene Generation: Outlines and drafts entire episodes, with you able to steer, rewrite, or override anytime 5. Editorial Review: Stress-tests every draft for pacing, engagement drop-offs, prose, and coherence before it locks 6. One-Tap Production: Pick a voice, convert to audio drama, and publish directly to Pocket FM’s millions of listeners 7. Global Scale & Monetization: Revenue-share on performance, with automatic localization so you earn across international markets Test Sherpa for free here: pocketfm.com/sherpa _____________________________________________ Generic LLMs fail at serialized fiction because they lack a long-horizon narrative reward function. Sherpa solves this through three core technical leaps: 1. Narrative World Model (State Tracking & Retrieval): Context windows degrade over long runs. Sherpa constructs an evolving semantic knowledge graph tracking character states, secrets, and plot dependencies. High-speed retrieval surfaces exact context on demand, maintaining zero continuity decay across hundreds of episodes 2. Hierarchical Story Planner: When writing a 500-episode story like Naruto, you need to plan 100s of sub plots. Rather than generating linearly, Sherpa decomposes narrative across discrete levels: season -> arc -> sequence -> episode -> scene. Rather than generating everything upfront, like a generic LLM, Sherpa uses progressive planning and dynamic replanning. As the story evolves, it identifies what changed, traces the downstream impact, and replans only the affected parts. 3. Prose Engine (Trained on series' retention data): LLMs write robotically, but serial fiction needs emotion, tension, pacing, and dialogue that sounds like real people. Sherpa's Prose Engine was designed specifically for storytelling. It was built on 1B+ tokens of Pocket's own stories, trained by learning from what listeners engage with, where they drop off, and what keeps them hooked. Feedback is taken from specialized evaluator models that measure every scene against a 40-item checklist. (Evaluator models were benchmarked against human reviewers and matched them 80–90% of the time.) _______________________________________________ Owning distribution and creation puts us in a very unique spot. More shows -> More data -> Sherpa becomes better -> more creator success -> more creators -> more shows Pocket FM has already seen one $100M IP. I believe Sherpa will soon lead to dozens of single-person studios creating billion-dollar shows. Most people are scared of AI but I think it'll unlock more human creativity, help creators earn more, and bring the next great IPs to life. This will create millions of jobs and new income streams.

A few days ago, I shared the World Bank’s summary of the results of the PISA for Schools assessment conducted in 171 Salvadoran public schools participating in our education pilot. Since then, I have seen some legitimate questions about what these results mean, as well as some claims about what they supposedly don’t mean. So let’s be very precise. First, what we NEVER claimed: 1. We never claimed these results represent the national average. They don’t. This was not the national PISA assessment. It was PISA for Schools, conducted specifically in the 171 public schools participating in our pilot program. That is precisely why my original post said: “These are the results of our pilot program, which includes AI tutors and has been running in 171 public schools for just over a year.” A pilot is, by definition, not the entire country. 2. We never claimed that these 171 schools were a statistically representative sample of every school in El Salvador. They weren’t selected randomly from all Salvadoran schools. They were selected to participate in the pilot because they were schools where the program could initially be implemented more easily: they had certain basic infrastructure and connectivity requirements, and school communities capable and willing to implement a new educational model. That creates a selection effect, and it should be acknowledged. But there is an important distinction here. These schools were NOT selected because their students had exceptional academic results. They were not our “best schools.” Not even close. We didn’t search the country for the highest-performing schools or students and then test them. They are normal Salvadoran public schools, urban and rural, in different parts of the country, serving ordinary Salvadoran families. And the characteristics that made implementation easier: connectivity, infrastructure, trained teachers and willingness to adopt the program, are also things that can be replicated in other schools. They are not innate characteristics of the students. We can install connectivity. We can improve infrastructure. We can train teachers. We can provide the same technology. We can deploy the same educational tools. In fact, that is exactly what we are doing. 3. These results also do NOT prove that AI tutors, by themselves, caused the improvement, although we believe they are a MAJOR part of these results. Our pilot is broader than that. It includes changes in teaching, support for teachers and principals, educational materials, monitoring and evaluation, technology, and AI tools, among other interventions. Our AI tutors and learning platforms are being developed with @grok, @Kira_Learning, and support from I Hate Flying Bugs. And the broader transformation of our education system is being supported by the @WorldBankGroup, @AgendaCAF, and @el_BID. All of these interventions can also affect student motivation and engagement. But that distinction matters: if students become more motivated BECAUSE of the program, that is not necessarily a flaw in the program or an explanation that somehow invalidates the results. It may be one of the ways the program works. The same applies to teachers. For example, the implementation of the AI tutors, including the requirement for teachers to log in and out of the system, coincided with teacher attendance in these schools increasing from 76% to 97%. If the program causes teachers to attend school more consistently, and that in turn helps students learn more, that is not an advantage that disappears when we scale the program. It is potentially part of the intervention itself. Of course, to scientifically isolate the causal effect of AI tutors, teacher attendance, student motivation, technology, training, or any other individual component would require a different experimental design. Now, what DO these results tell us? 1. They tell us that 1,198 students in 171 ordinary Salvadoran public schools… 🧵1/2👇🏼