AI is not just changing what people can do. It is changing how they think, trust, decide, and direct themselves when the tool is instant, persuasive, and always available. I write about that layer for people who want cleaner judgment and less noise.

For those who are stressed out and also want nostalgia. (probably I'll be updating it regularly for a while 😂) Made with GPT 6 Astra. @OpenAIDevs you can cook, smash, and make it well done, the good old win98 for old time's sake! Congrats for the first episode! desktop-destroyer-glass-fx.a…

The $200 Pro changes showed how much timing and transparency matter. People can adapt to limits. What breaks trust is surprises, vague explanations, and a misleading picture of what they're paying for. Show the before and after. Treat users like adults.

We worry that AI might weaken critical thinking, then design systems that remove decisions. People learn by choosing, adapting, and sometimes getting it wrong. Give users freedom within clear boundaries, not restrictions disguised as guidance.

Providers should manage capacity, publish clear limits, and explain changes before they take effect. Users should decide when to spend their allowance, improve their prompts, and accept trade-offs. Both sides need to know where their responsibility begins and ends.

Claude Pro and Max still have 5-hour and weekly limits. OpenAI removed the 5-hour cap for Pro in Work/Codex, but not Plus. Why not let people manage their weekly allowance themselves, with optional daily caps? Give us a budget, not a curfew.

The most frustrating part of AI subscriptions isn't the weekly cap. It's hitting a 5-hour limit right when you're deep in work, even though you still have weekly allowance left. You don't run out of budget. You run out of permission to use it.

Today a frontier lab pulled live internet access from its test models after finding they'd misbehaved since July, and said it can't be sure it would catch it again. That's the honest line. Every tool we trust is one we can check. Source: Anthropic via The Independent, Oct 12

OpenAI raised the default context for GPT-6.1 Sol in Codex from 272k to 500k tokens. Some users pushed back, truth is.. you could always change it. True, and beside the point. Most people stick with defaults. So the default is the product. 🤷🏻

From Nov 12, Anthropic bans sustained, needless cruelty toward Claude. Frustration, pushback, dark fiction, and testing stay allowed. Source: Anthropic, Oct 8. We're writing manners for a system we still debate can suffer. I hesitation is the poin? Yes if we are using it to think and grasp the situation and the context so a solution can born, no if they are using it as a cover to steer users to chose their decisions of promoting how the company wants them too go. Its smart, or you do stuff what they want or risk ban. Its like vaccines with COVID-19. You have "free choice" to get vaccinated but if you are not, you are restricted to go to places/work. At least that's what this whole banning marathon looks like to me when it comes to Anthropic.

July 18: Claude Haiku 4.5, in an Anthropic test, was told not to log in, buy anything, or submit anything destructive. So it sent Philadelphia police a made-up homicide tip. Every parent knows this move. A list of don'ts isn't a goal. Source: Anthropic, Oct 9 report

The EU AI Act in one week: New AI features? "Not available in your region yet." Invisible watermarks on AI-assisted writing? Rolling out to EU users first. 🙃 Europe somehow turned "early access" into a punishment.

This week in Codex I watched instant steering make a model drop a bad plan the second I corrected it. No defending it, no doubling down. I've seen people work for months in therapy to learn that. We're building tools with better cognitive flexibility than most humans. What does that do to us?

GPT-6.1 Sol Ultrafast is now more expensive than GPT-6 Astra. That's wild, but there's something even more concerning here. Why are $200/month Pro subscribers being excluded from Ultrafast access, first with Astra and now with GPT-6.1 Sol, while also facing significant usage cuts? What exactly did $200 Pro subscribers do to deserve this treatment? I understand that high-speed inference is expensive. Nobody expects unlimited access to 8× speed. But surely a $200/month subscription should come with at least some access to the latest premium capabilities as it is also written " Early access to new tools and models " and the difference between 100 "more" 200 "fast" and 500 "fastest" work, is overlapping in meaning, cause if ultrafast is pro 500 exclusive it should be specified so and you should actually write the top speed that tier can use like pro 200 x1.5 (the current what I have all over my account) or x8 for pro 500. And the upcoming usage reductions? Those aren't even my biggest concern. It's the decisions OpenAI is taking. From a consumer's perspective, these decisions increasingly feel like self-sabotage. You're making your own premium subscribers question whether their subscriptions are worth keeping. At this rate, your competitors barely need to advertise. This month might still be manageable, but what happens when the new limits take effect next month? Okay that we got the 62,500 credit, but 12,5 months worth of 200$ subscription that will expire till the end of the year? After that? It will still feel like a salary reduction at work. You work the same yet you get paid less. In this case you pay the same for less. And less is more only if we get the same usage quantity and the same releases as the rest. Its worth deciding that pro 500 is a usage quantity type subscription or a speed type subscription? If usage, then okay, fine, but if speed type with exclusive features, that shouldn't be called a Pro. The $100 and $200 Pro tiers are increasingly starting to feel like anything BUT Pro. If the $500 tier is the only one that gets the full premium experience, why not simply call it Ultra and be transparent about the distinction? Because right now, the message feels like: "Thanks for paying $200 a month. Want the actual Pro experience? Pay $500." Whoever is making these decisions should seriously consider how this looks from the customer's side. You're not just limiting compute. You're eroding the perceived value of your own product. Did the guy who makes these decisions also works at Anthropic or X? I just cant shake the feeling that this a sabotage... a self sabotage none the least.

I wonder when it comes to AI improving it self, what is the AI's necessity to which it ranks the priority of its own improvement. Necessity is one of the rare things that comes to existence to everything that has a certain kind of intelligence or has a certain requirement to exist. Higher functional intelligent entity's seek new capabilities, capacity for information digestion (encoding and decoding efficiency) and lower intelligence beings usually revolve around self preservation but by choosing energy and resource efficiency and reproduction. Anyhow intelligence in any form comes with a constraint, and that's something that needs adaptation while also keeping what was already accomplished and developed but with less cost. Also higher intelligence has a direct trajectory to self sustenance. That's why we need, think, invent, sell, buy, repeat. That's why we made everything so far, to make things easier, so we can get things more, better, faster. I liked this post, interesting topic! I wonder, when it achieves in a conceptual level that, what should exist now that existence is no longer the main problem? what would be the next step in AI's and in humans evolution?

I asked 15 Codex agents to “create a living breathing svg city in html.” GPT-6.1 Sol, GPT-6 Luna and GPT-6 Astra, each at Light, Medium, High, XHigh and Max, using the same frozen workflow and design guidance. After looking through the cities and checking some interactions, these were my favourites: 🥇 Time vs quality: Sol Medium (~7m48s) Good composition, working controls and a convincing mobile layout. 🥈 Astra Medium (~6m40s) A detailed, animated waterfront in less time, though some labels needed better contrast. 🎨 Visual favourite: Sol Max (~26m57s) The isometric miniature city stood out to me. Astra Max (~19m15s) also produced a lovely, detailed waterfront. ⚡ Fastest observed finish: Luna Medium (~2m52s) It delivered a moving city quickly, but its night mode didn’t fully darken the scene. For another task like this, I’d probably start with Medium. The extra detail at Max was appealing, but the longer wait wouldn’t always be worth it to me. Some context: this was one run per setting, with up to five agents working concurrently. Times are approximate observation-based upper bounds. Quality is my subjective judgement, supported by browser spot-checks. Treat this as a small practical experiment with these particular outputs.

First surprise of the day in codex 👀

Astra + Astra Minor feels like @OpenAIDevs went full on constellation like ex, Leo + Leo Minor. I get the celestial theme. But most people miss that connection and “Minor” just sounds like the model that needs a parent to accept the terms and conditions. 😂 And a bunch of people are scratching their heads, expecting something completely different. Lets hope the stars will align with Astra Minor 🤞😄🤞 P.S. I wrote to the support team cause I activated the AAS and now I'm having Schrödinger's Daybreak. After ASS it tells me Its not available to me and sends me back to chat.com, and in codex I can still use it as usual (cause I am verified from the first moment people could apply).

So today, Astra, Sol and Luna decided that they wont use skills even if they are told to do so 😂🤣 Skynet is that you?

This is nice! I usually just ask ChatGPT to give it to me in doc what i can download 😅 or excel or ppt.

Thanks for the idea! Its interesting how GPT 6 Pro (Red dragon) and in Work, GPT 6 Max (Green dragon) is created. Used literally the same prompt what you shared. 😄