A few days ago, one person said to me, “You will enjoy exploiting nature and taking the job market.” Though criticism or hate speech doesn’t usually affect me (I try), this did, maybe because there is some truth in it, and I felt a responsibility inside.

It is really true that AI is creating a lot of problems in the world. Ecological (data center carbon emissions), economic (job market saturation), psychological (relationships with AI cause a lot of breakups, anxiety, and suicide too), and in the future maybe existential (it may have the capability to destroy the whole civilization or make humans irrelevant, according to some experts).

But it is also creating a lot of great impact on society. It has solved many big scientific problems, made education and learning easier than ever, and solving cancer and stroke are among the top goals of AGI building.

Even if there is a probability that non-aligned AI will cause extinction, if it doesn’t happen, then surely millions of people will die because we did not do it earlier. We can compare it with one scenario. Imagine you are standing in front of a train. You have a machine in your hand, and if you don’t press it, 10 people will die (among them your father, your mother, or your lover, and everyone is someone’s loved one).

But if you press it, and you can’t break the chain within 5 minutes, then you, everyone around you (likely 800 people), and those 10 people will die. But you know that, according to most researchers on this topic, the chance that you cannot break the chain is 10-20% (even if we are a little skeptical), and the chance that you will break the chain is 80-90%. Now, will you press the button to save those 10 people, or not?

I’m saying this because it’s true that a level of non-aligned AI can cause the extinction of humanity. According to the same researchers, the chance of humans failing to control AI is about 10-20% on average. But AGI can probably also solve cancer, where millions of people die every year. My grandfather was one of them.

You can say that AGI solving it is not proven science, and that it is only the hope of most researchers. But that 10-20% (about extinction) is also not based on any scientific fact or mathematical theorem actually.

Secondly, this is not the only case. If we consider that analogy again, the button has actually already been pressed. So there are 3 kinds of people in these 5 minutes:

1. Those who are trying to use it, control it, or grow it, so that people don’t get hurt and it grows. 2. Those who don’t think about it. 3. Those who are trying to stop it, who are against it.

Okay, this is not that black and white, and it’s mixed in reality. But anyway, this is unstoppable. So, there are only 3 choices for us: do something about it (take advantage of it, try to control it, etc.), don’t care about it, or be against it. It can never be stopped.

Even if we stop it in one country or in many countries, there is very little chance that no one will build it. (Chinese open models are reported to be only about seven months behind the US, and the US put export controls on two of its top models in June 2026, which were later removed. Also, once a model’s weights are public, nobody can take them back.) How do you know if someone or some group is trying to make it secretly? How can you guarantee we will know?

And I think the best and most rational position is the 1st, because it’s the only way we can protect the largest number of people. Yes, it will have effects and create many problems, but those are also problems, and if we created them, it’s not our responsibility to just fear it, spread hate speech, or accept it. It’s our responsibility to solve it.

TorchLean is a Lean 4 framework for neural-networks

Wall Street's AI Debt Machine Is Starting to Choke on Supply

Why We Should Work on AI Instead of Fearing It: A Philosophical Perspective

Popular Sovereignty

Analysts followed AI agent's randomly assigned preferences in 86% of decisions

The Brutal Harvard Class That Suddenly Everyone Wants to Take

What was the turning point for Postgres?

Steve's Old Computer Museum

Lamdera v1.5.0 is out Elm 0.19.3, faster Wire, safer virtual DOM and++

AI Agents Flooded RubyGems to Get Code Execution on Its Docs Builder

A framework for capturing webhooks from providers reliably and durably

Propaganda and Cinema

Tell HN: I'm going on strike this November

Show HN: Toolvy – 120 media tools that run in the browser

TLP Is Not a Suggestion

Tagd – a game of Tag (You're it)

AI bots flooded 2023 UK government consultation on disposable vapes ban

Show HN: TAW, herdr for tmux: see which coding agent needs you

Zero database QR labels carry structured data

Answers at the Back of the Book

About the compromised eMudhra intermediate CAs

Show HN: Nuvrail – make an AI agent ask before it sends, moves or deletes email

Crab Eating Frog

MotionBricks: Scalable Real-Time Motions with Modular Latent Generative Model

Ten days later, Git on macOS still trusts the compromised eMudhra certificates

A Little Bit of Slope Makes Up for a Lot of Y-Intercept [video]

I let Claude run my social media accounts – where it won and where it failed

Where Have All the Car Commercials Gone?

OpenDisplay

Convert OLM to PST

Show HN: Ballast – macOS app to clear disk space