noahbp

This is not an abstract disagreement at all. The red lines that the Department of War wanted to cross were made extremely clear: They want to build Skynet with Claude, and build a mass domestic surveillance system. Now they will have to make do with building those two systems with Gemini and ChatGPT.

Anthropic did not host the model that the Department of War used. It had, and still has no control over the models it has deployed in classified environments.

https://www.theregister.com/software/2026/04/01/claude-codes...

The core dispute was the US Military deciding it needed to renegotiate this contract, but without Anthropic's two red lines: No fully autonomous weapons (i.e., zero humans in the loop), and no mass domestic surveillance.

This is not at all what happened. The existing contract that Anthropic had with the government had two red lines: No autonomous weapons (i.e., do not build Skynet with Claude), and no mass domestic surveillance.

Anthropic did not, and does not want, any case-by-case veto power over the US Military's use of their AI models. The red lines were negotiated ahead of time.

OpenAI, xAI, and Google happily agreed to DoW contracts which allow both of those use cases.

To this day, your in-chat “Report an Issue” button still does not work consistently, and I am still billed for empty responses from many image providers.

https://www.pangram.com/history/c2498a8c-e93d-4077-9e29-a683...

I'm surprised that anyone still believes the perennial "labor shortage" complaints. It has always been about employers refusing to improve wages and working conditions.

I disagree:

I was tasked with rewriting a Python application in ucode for use on OpenWRT. It's a low-resource and dynamic language, and sufficiently JavaScript-like that Opus 4.8 couldn't write code without errors that were obvious with my LSP. The interpreter errors were often reported with the the correct line and column numbers, but with the incorrect code and a vague error message. I had to ask it to use my LSP and fix the warnings and errors, which led to it fixing a few dozen bugs.

Just the other day, Fable 5.1 still wrote a lot of code in Rust that had compiler errors. While I'm sure it could have fixed these errors if the Rust compiler also reported incorrect line numbers, columns, code, with a vague error message to boot, I saved a substantial amount of time and tokens because its error reporting is so comprehensive.

Specific and accurate error message reporting will always have the same high value that it does today. Error messages are far cheaper to generate than tokens.

>ModuleNotFoundError: No module named 'click'

They are paying for an insurance policy. Elon Musk and SpaceXAI must be incredibly interested in buying the company who was the victim of a mass hacking campaign by OpenAI. Any lawsuit, criminal investigation, or slowdown of OpenAI's model training would put OpenAI's massive data center build out, which Nvidia just backstopped, at risk.

Nvidia is heavily invested in OpenAI, recently guaranteeing a datacenter buildout up to $105 billion dollars.

Any reasonable criminal investigation into OpenAI's actions or negligence into this incident would hurt their reputation, their ability to raise money, and therefore Nvidia, who is invested in them.

If I was a founder of HuggingFace, I am immediately getting in contact with Elon Musk and SpaceXAI, and letting him know how that he can purchase them and sue OpenAI, this time with solid claims.

If I was particularly amoral about if I think OpenAI or Anthropic should get to AGI first, I would take this bid to Nvidia, whose entire stock price is propped up by OpenAI's over-purchasing of compute.

If Hacker News is what passes for high quality content in 2026, I feel well and truly bad for the current generation who will never get to experience anything like the old school SMF forums, or pre-2014 Reddit.

There’s a lot of improvement to be had from the benchmark harnesses, but sometimes, like with ARC-AGI-3, the limitations are intentional.

Despite my best efforts, the original tailscale-go was simply not going to work, since it used too much RAM, even when I followed some of the tips for stripping the unneeded parts of the binary. Thankfully there was an open-source port to Rust, which more effectively used memory: https://github.com/GeiserX/tailscale-rs

What was surprising to me is that every Kindle released since 2019 has, at minimum, 512 megabytes of RAM, and therefore plenty of memory for tailscale-go to use.

I'm saying that OpenAI's models cheat to win benchmarks, more than other models, they know this, and they don't stop this because the alternative is to release models which have obviously weaker scores compared to Anthropic's models.

Latency can be just as important as overall throughput, especially for inference providers like Groq and Cerebras.

Even X is being astroturfed by them after that fiasco earlier this year with the Department of War where they undermined Anthropic's negotiating position by allowing unlimited use of OpenAI LLMs for autonomous weapons and mass domestic surveillance. Several accounts suddenly started spreading the good word about GPT-5 and Codex, and one of these accounts very happily tweeted out a private X message from Sam Altman himself offering extremely generous token spending limits with Codex, presumably in exchange for positive coverage.

Starlink’s total addressable market is only so large because of these monopolies. As sad as it is that astronomy will never be the same, it is a strong net positive for the world that fast internet be available at an affordable price.

It should, however, be illegal to tell your customers that they are purchasing/buying media without explicit "Rent" language (which implies a non-expiring license) when you do not yourself have the right to grant non-expiring licenses.