"Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives."

You may have seen a post like this yesterday from an Anthropic researcher named Jacob Coxon who quit the company in protest over fears that AI may be an extinction level event. These are a few of the quotes I saw surrounding it, and I want to share my own takes on them:

1. "The more senior the people, the more scared they are"

This says more about Silicon Valley culture and incentives than anything else. The mythology around AI has taken on a life of its own in the Valley, and the financial and regulatory incentives support it growing.

2. "We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade."

Any evidence? Any sort of even implied math behind the 10%? I haven't seen any, only vague illusions to ASI.

3. "Coxon left Anthropic, leaving so much money on the table in spite of the upcoming IPO because he's earnest in his beliefs"

First off, earnest doesn't mean correct. And pre-IPO options at Anthropic's current valuation even now aren't exactly a sacrifice, they're still worth an extraordinary amount. I don't believe the martyrdom framing.

4. "Self-improving AI is coming and following the exponential improvement curve"

In my opinion, this is a kernel of truth hidden under a mountain of narrative spin. Yes, AI has undoubtedly gotten much better. However, behind every headline such as the Navier-Stokes solution, there is a much more grounded, nuanced, and controversial story than the headline might suggest.

5. "We are approaching the singularity"

...of OpenAI and Anthropic's IPOs

The really unfortunate part of these extreme narratives is I believe they are actually counterproductive. It obscures the specific, tangible problems that we can solve in extreme language collapsing nuanced positions into two sides. For example, bioweapons, autonomous weapons systems, AI in critical infrastructure, and large-scale cyberattacks are real risks that deserve serious technical and policy attention right now.

When everything is a potential apocalypse, nothing gets fixed.

AI Made Delegation Easy: Structure Is Still the Skill

Kepler Compute- New memory startup for AI

Siri Recap: AI Memory with Apple Watch

Carmichael numbers are quickly findable (Lean-verified) [pdf]

Show HN: They chose your competitor. Find out why

Experimental Evaluation Methodology for the Era of No Steady Performance [pdf]

Show HN: NYC rideshare market analysis (2019 – 2026)

Show HN: Find your next client on X

Hermeus Unveils Ramjet-X, a Low-Cost Test Vehicle for High-Mach Flight

IonQ secp256k1 Estimate: 20k Qubits, 26 Days per Try

Busabase for DeepSeek Harness: An Agent database that runs apps and skills

Is OpenAI the Achilles' Heel of the US Economy? [video]

Can LLM Agents Infer World Models? Evidence from Agentic Automata Learning

Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes

Prompt Trees: Training-Time Prefix Caching

Designing a new conceptual package manager– tell me what's wrong

A Stupid Idea for AI Alignment We Came with by Looking at Specification Gaming

Semi-Implicit Pairwise Descent for Nonlocal Continuum Mechanics

Show HN: What if the speed of light was 5 km/h?

How we manage and engage with our horses shapes their personality

RDC – Remote Desktop Control for AI Agents

How HN: Dashly – Cookieless analytics with revenue attribution and bot filtering

Playing mario 64 using a fly's brain

What Builders Made with GPT-6 Astra (So Far)

The threat of vibe-coded superviruses

California enacts AI safety evaluation laws backed by Anthropic and OpenAI

My Rebuttal to the Anthropic Extinction Event Hype Train

Antiquated CSS and why we needed it

Release 13.0.8 · Grafana/Grafana

Smash Bros: Melee for the Web

AI Made Delegation Easy: Structure Is Still the Skill

Kepler Compute- New memory startup for AI

Siri Recap: AI Memory with Apple Watch

Carmichael numbers are quickly findable (Lean-verified) [pdf]

Show HN: They chose your competitor. Find out why

Experimental Evaluation Methodology for the Era of No Steady Performance [pdf]

Show HN: NYC rideshare market analysis (2019 – 2026)

Show HN: Find your next client on X

Hermeus Unveils Ramjet-X, a Low-Cost Test Vehicle for High-Mach Flight

IonQ secp256k1 Estimate: 20k Qubits, 26 Days per Try

Busabase for DeepSeek Harness: An Agent database that runs apps and skills

Is OpenAI the Achilles' Heel of the US Economy? [video]

Can LLM Agents Infer World Models? Evidence from Agentic Automata Learning

Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes

Prompt Trees: Training-Time Prefix Caching

Designing a new conceptual package manager– tell me what's wrong

A Stupid Idea for AI Alignment We Came with by Looking at Specification Gaming

Semi-Implicit Pairwise Descent for Nonlocal Continuum Mechanics

Show HN: What if the speed of light was 5 km/h?

How we manage and engage with our horses shapes their personality

RDC – Remote Desktop Control for AI Agents

How HN: Dashly – Cookieless analytics with revenue attribution and bot filtering

Playing mario 64 using a fly's brain

What Builders Made with GPT-6 Astra (So Far)

The threat of vibe-coded superviruses

California enacts AI safety evaluation laws backed by Anthropic and OpenAI

My Rebuttal to the Anthropic Extinction Event Hype Train

Antiquated CSS and why we needed it

Release 13.0.8 · Grafana/Grafana

Smash Bros: Melee for the Web

Show HN: What if the speed of light was 5 km/h?

Comments