Henry Fisher

Interesting investigation argues that a lobbying campaign behind U.S. “age verification” laws for devices and operating systems is largely driven by Meta (Facebook’s parent company) and allied advocacy networks, rather than grassroots child-safety groups.

---
I'm wondering whether Meta is the final beneficiary, or if there's something behind its actions and funding?

https://old.reddit.com/r/linux/comments/1rshc1f/i_traced_2_billion_in_nonprofit_grants_and_45/

Dilettantism is an engagement with a subject without deep knowledge, serious study, or professional competence. And the real danger of the “vibe-coding” culture is that it strongly normalizes and promotes an entirely dilettant approach to code engineering. Very soon, we will see more and more consequences of that. The hangover is closer than we think.

https://www.theregister.com/2026/02/27/lovable_app_vulnerabilities/

What's the point of reading books if I remember max 5% after all?

After my dad died, I received several friends’ condolences written with.... ChatGPT. That was quite easy to identify by the style and specific punctuation. To be honest, it disturbed me because it created a strange, mixed feeling. On one hand, I understand how hard it is to find words of support for someone who has lost a loved one, and ChatGPT can indeed act as a protective layer for both sides in such a situation. On the other hand, it felt so fake and uncomfortable to realize that you were, in fact, talking to a soulless machine even in a moment like that. I immediately remembered myself writing birthday congrats with LLMs and myself pushing every email through Claude, putting linguistic perfection over authenticity. And I feel something truly sad about it...

Creating an app costs almost nothing today. Seriously. I am not talking about “Lovable”-style stuff, which cannot be taken seriously from an engineering perspective. I mean a well-managed engineering process that delivers reliable code that still counts on a human engineer, but is amplified by AI coding agents. For 20 bucks for Claude Code and a kilo of coffee beans, you can ship a brand-new app and upload it to a marketplace. Easy!

And think about the people who have already lost, or will soon lose, their jobs at Google and Meta, or who spent years and vast amounts of money earning a computer science degree. Will they all end up flipping burgers at McDonald’s after making $350K a year? Obviously not. They will desperately look for new opportunities and ways to utilise their software engineering skills. Many will try their hand at the SaaS market, which sounds cool until you realize they will flood the global commercial arena with millions of similar apps and drive market saturation through the roof.

Another pain point is that as software development costs trend toward zero, you can not rely on the subscription model anymore. The only way to make money with a newly built AI app is to add a margin on top of token pricing. In fact, you’re not selling your software; you’re wrapping tokens… that’s it. The problem is that this margin cannot realistically be very large, which leaves no path to building a long-running, scalable business model. More users = More tokens = More cost. A vicious circle...

This looks like we are witnessing a great heat-death scenario for the SaaS model itself, with LLM hosts as the only real winners.

Hmm... Are we catching the first vibes of AI hangover? Despite the many Instagram reels I watched recently expressing deep disappointment in AI automation, I came across two interesting documents that expose AI automation struggles at different scales. The first one is an article, “America's $1T AI Gamble” by Joseph Politano, which highlights that despite the US making an unusually large, unusually concentrated capital bet on AI capability (around 3.5% of US GDP), the revenue growth of the tech sector has not reaccelerated enough to “cover” the investment pace. Another interesting report is “Leading through uncertainty in the age of AI” by PWC, which reveals that most of the companies adopting AI see no payoffs and only 30% report increased revenue from AI in the last 12 months. More concerning is that this is accompanied by massive layoffs and widespread hiring freezes that are dominating the labour market. ---- After going through and reading all these articles, one is left with a faint impression that the innovation system is not yet capable of digesting AI technology. The system is imbalanced. On one side, there are enormous expectations and promises tied to AI adoption and application. On the other, the system is burdened by the energy and financial requirements necessary for corporations and companies to participate in the race for AI-driven automation. This raises another question. Who are the beneficiaries of this transformation? If companies are not increasing their revenue while automating processes to the point where human labor becomes unnecessary and obsolete, what is the collective benefit of deploying these technologies? From a technological determinist perspective, everything appears obvious. But what, then, is the role of the human being in this future?

@chmod644@mastodon.social seems like you needed so much support recently

The current wave of AI hype is pushing automation faster than safety can keep up. As systems gain more autonomy, the risks scale just as quickly. Many of the AI disasters people warn about are not theoretical. They are already forming. Now!

A clear and sad example is OpenClaw, a new do-it-yourself AI assistant that runs directly on users’ hardware, operating systems, and everyday applications. It automates emails, calendars, web browsing, and interactions with online services. At its core, it is a cascade of LLM agents given broad control over a machine. Insanely broad access!

In practice, this looks less like a smart assistant and more like a modern Trojan horse. With full system access, OpenClaw inherits every major weakness of today’s AI models: unreliable behavior, susceptibility to prompt injection, and massive data and privacy exposure.

And it's already here! Recent security advisories have already highlighted vulnerabilities that allow attackers to trigger remote code execution simply by feeding the system attacker-controlled content. Simple as that!

The lesson is... When an unreliable LLM is handed deep automation power without extra control or proper containerization, convenience turns into an attack surface. The faster we chase overautomation, the more security debt we accumulate.

Here is a good analysis on why poor countries become poorer and rich countries are richer.

https://davidoks.blog/p/why-poor-countries-stopped-catching-690

https://www.chat-gdp.org/we-were-wrong-about-convergence/