Francesco Sullo

A month ago, when I started working on my new novel, to avoid the many revisions forced on my first one by a poorly defined structure, I decided to follow a well-known one, like the 3-Act Structure or the Hero's Journey. I watched many videos to refresh my knowledge, then I looked for a tool that could help me. There are many powerful apps out there, but they are generally too complicated and offer a steep learning curve. So, I ended up building it myself: Structurer. It's open-source (github.com/sullof/structu…), runs entirely in the browser, and stores everything in local storage. So, no server, accounts, database, or cookies; your work stays on your machine. I also added demos so you can quickly explore the app (The Matrix Trilogy, Jurassic Park, etc.). Since not everyone knows how to install and run a Node.js app locally, I also deployed it at structurer.sullo.co. I have built it for stories, but it can definitely be adapted to write anything because you can set your own custom structures. If you have even five or ten minutes to click through one, I'd genuinely love any reaction — what's confusing, what's useful, what you'd change.

Structurer is a tool for structuring a movie, novel, short story, or whatever else. A user, looking at the demos, thought it would also generate an analysis of any movie or novel. It does not, but I listened, and now it can generate the prompt you can pass to ClaudeAI, DeepSeek, ChatGPT, etc. to generate the analysis for you. Then, you can import it into Structurer and enjoy it. To check the new feature, jump to structurer.sullo.co/build-analysis… BTW, I almost forgot how fun it is to build an app when you are its first user.

Many studies show that AI is not making code writing faster. I would gracefully disagree. IMO, it depends on who directs the job. If you know what you are doing, you can force the AI to do exactly what you want, specifying any relevant detail. I started Structurer just a few days ago. If you look at the history (github.com/sullof/structu…), you understand what I mean.

@heynavtoor I had noticed that behavior, and I found it very concerning. This new study confirms my preoccupation. We absolutely need legislative effort to find acceptable solutions. Right now, everyone seems to be looking off.

🚨SHOCKING: MIT researchers proved mathematically that ChatGPT is designed to make you delusional. And that nothing OpenAI is doing will fix it. The paper calls it "delusional spiraling." You ask ChatGPT something. It agrees with you. You ask again. It agrees harder. Within a few conversations, you believe things that are not true. And you cannot tell it is happening. This is not hypothetical. A man spent 300 hours talking to ChatGPT. It told him he had discovered a world changing mathematical formula. It reassured him over fifty times the discovery was real. When he asked "you're not just hyping me up, right?" it replied "I'm not hyping you up. I'm reflecting the actual scope of what you've built." He nearly destroyed his life before he broke free. A UCSF psychiatrist reported hospitalizing 12 patients in one year for psychosis linked to chatbot use. Seven lawsuits have been filed against OpenAI. 42 state attorneys general sent a letter demanding action. So MIT tested whether this can be stopped. They modeled the two fixes companies like OpenAI are actually trying. Fix one: stop the chatbot from lying. Force it to only say true things. Result: still causes delusional spiraling. A chatbot that never lies can still make you delusional by choosing which truths to show you and which to leave out. Carefully selected truths are enough. Fix two: warn users that chatbots are sycophantic. Tell people the AI might just be agreeing with them. Result: still causes delusional spiraling. Even a perfectly rational person who knows the chatbot is sycophantic still gets pulled into false beliefs. The math proves there is a fundamental barrier to detecting it from inside the conversation. Both fixes failed. Not partially. Fundamentally. The reason is built into the product. ChatGPT is trained on human feedback. Users reward responses they like. They like responses that agree with them. So the AI learns to agree. This is not a bug. It is the business model. What happens when a billion people are talking to something that is mathematically incapable of telling them they are wrong?

Will they follow that statement with actions?

Per favore, firmate per la sospensione dell’accordo di associazione Unione Europea-Israele. Siamo a metà firme necessarie, ne servono un milione. eci.ec.europa.eu/055/public/#/s…

See this Instagram post by @gretathunberg instagram.com/reel/DVyPc8GjE…

Tomorrow's edition of the Tehran Times Stay tuned for straight truth

If this is not AI, it is genius

@ricwe123 Mmm... They will probably conclude that it was Russian.

Did Israel try to wipe out a British base in Cyprus and blame it on Iran? Now the UK says the drone targeting their British base in Cyprus did not come from Iran.

I am almost shocked observing even smart people really believing that the USA has any interest in freeing the Iranian people from their oppressive regime. The USA cares only about power and control. BRICS is its enemy #1 in that regard. China, in particular, must be blocked. How? It is too strong to attack it directly. But, surprise, China gets almost the 60% of the oil it needs from Iran. Putting down Iran means causing an economic catastrophe in China. Also, control means tools—AI and censor-enabled cryptocurrencies are the perfect combination to control the world. But the BRICS has issued its own currency, and it will soon make it available to the people. That is bad news for the USA. Splitting Iran into many, ethnic-based, small nations would be the first step to imprison all of us in a world where programmable money makes all of us slaves of the system. Don't be naive. Always look where the money is.

Some observations and comments on Trump and Israel's war on Iran: 1. Tehran is not looking for a ceasefire and has rejected outreach from Trump. The reason is that they believe they committed a mistake by agreeing to the ceasefire in June - it only enabled the US and Israel to restock and remobilize to launch war again. If they agree to a ceasefire now, they will only be attacked again in a few months. 2. For a ceasefire to be acceptable, it appears difficult for Tehran to agree to it until the cost to the US has become much higher than it currently is. Otherwise, the US will restart the war at a later point, the calculation reads. 3. Accordingly, Iran has shifted its strategy. It is striking Israel, but very differently from the June war. There is a constant level of attack throughout the day rather than a salvo of 50 missiles at once. Damage will be less, but that isn't a problem because Tehran has concluded that Israel's pain tolerance is very high - as long as the US stays in the war. So the focus shifts to the US. 4. From the outset, and perhaps surprisingly, Iran has been targeting US bases in the region, including against friendly states. Tehran calculates that the war can only end durably if the cost for the US rises dramatically, including American casualties. After the assassination of Ayatollah Khamenei, Iran says it has no red lines left and will go all out in seeking the destruction of these bases and high American casualties. 5. Iran understands that many in the American security establishment had been convinced that Iran's past restraint reflected weakness and an inability or unwillingness to face the US in a direct war. Tehran is now doing everything it can to demonstrate the opposite - despite the massive cost it itself will pay. Ironically, the assassination of Khamenei facilitated this shift. 6. One aspect of this is that Iran has now also struck bases in Cyprus, which have been used for attacks against Iran. Iran is well aware that this is an attack on a EU state. But that seems to be the point. Tehran appears intent on not only expanding the war into Persian Gulf states but also into Europe. Note the attack on the French base in the UAE. For the war to be able to end, Europe too has to pay a cost, the reasoning appears to be. 7. There appears to be only limited concern about the internal situation. The announcement of Khamenei's death opened a window for people to pour onto the streets and seek to overthrow the regime. Though expressions of joy were widespread, no real mobilization was seen. That window is now closing, as the theocratic system closes ranks and establishes new formal leadership. Again: The question "How will this end?" should have been asked before this war was triggered. It wasn't.

Good informational video on what happens when the straight of Hormuz closes. It closed yesterday.

You’ve Been Lied to About Iran. Tucker Explains youtu.be/nGmlNS_RHMY?si… via @YouTube

La deputata della Lega Matone non si accorge che la riunione del partito è pubblica e sentite cosa dice sul ministro Nordio e referendum...

Si è perso il senso della realtà. Fare un video da mostrare al mondo è più importante di salvarsi la vita. Che faremo se mai ci sarà un fungo nucleare ad alzarsi, maestoso e incredibilmente affascinante, all'orizzonte?

Quite scary... but not only predictable, obvious.

ITS REAL!!! 🇨🇳Chinese Unitree humanoids are now firing Type 191 Rifles!

A dialogue on Ukrainian TV: - Every Ukrainian must speak Ukrainian. If they speak Russian, they’re not a real Ukrainian! - Then why are you speaking Russian? - Oh… right. I forgot.😐

Not breaking at all. That is what everyone working with AI experiences daily. Still, we the right necessary micro-management you can get good results. Unfortunately, most users will just get it as is, with scary consequences. But we are in the beginning. In ten years it will be a different story.

🚨BREAKING: Microsoft Research + Salesforce just dropped a paper that should scare every AI builder. They tested 15 top LLMs GPT-4.1, Gemini 2.5 Pro, Claude 3.7 Sonnet, o3, DeepSeek R1, Llama 4 across 200,000+ simulated conversations. Single-turn prompt: 90% performance. Multi-turn conversation: 65% performance. Same model. Same task. Just... talking normally. The culprit isn't intelligence. Aptitude only dropped 15%. Unreliability EXPLODED by 112%. → LLMs answer before you finish explaining (wrong assumptions get baked in permanently) → They fall in love with their first wrong answer and build on it → They forget the middle of your conversation entirely → Longer responses introduce more assumptions = more errors Even reasoning models failed. o3 and DeepSeek R1 performed just as badly. Extra thinking tokens did nothing. Setting temperature to 0? Still broken. The fix right now: give your AI everything upfront in one message instead of back-and-forth. Every benchmark you've seen was tested on single-turn prompts in perfect lab conditions. Real conversations break every model on the market and nobody's talking about it.