https://news.ycombinator.com/front?day=2009-03-11

https://en.wikipedia.org/wiki/Vogtle_Electric_Generating_Pla...

Even if the US federal government refuses to act, smaller government entities can deploy their own compute clusters running strong open-weights models like Kimi K3, and provide their own citizens with free access to strong AI models. There may come a time when even Chinese providers no longer release strong model weights, but then the previous generation of strong open-weights models can be used to accelerate the development of data pipelines and training setups to build the next generation. California or Australia (or many others) could fund university-led R&D efforts to build completely open models.

Despite the angry noises that leading providers make about their model outputs being used to train competing models, that's not illegal. States can use Common Crawl, Anna's Archive, Books3, commercial model outputs, and whatever other data sources that are useful for building models. States have sovereign immunity to ignore copyright complaints, if they decide to pursue model development.

The big AI companies don't have much of a moat. Let the big companies spend a fortune figuring out what works best, then openly build imitations having 95% of the capability at 5% of the cost.

https://www.world-nuclear-news.org/articles/amazon-updates-s...

However, building new nuclear reactors is a slow process. These reactors aren't expected to be built before the 2030s.

China is currently the fastest at bringing new reactors online. Even there it takes 6-7 years from construction start to commercial power generation, e.g. https://www.world-nuclear-news.org/articles/chinese-reactors...

https://en.wikipedia.org/wiki/Big_Gulp

Even bigger sizes arrived in the 1980s:

Called Big Gulp because it was initially the largest such drink available at any retailer, 7-Eleven eventually introduced larger sizes. In 1986, they introduced the 44-US-fluid-ounce (1,300 ml) Super Big Gulp, followed by the 64-US-fluid-ounce (1,900 ml) Double Gulp in 1989 (later reduced to 50-US-fluid-ounce (1,500 ml))...

https://www.utilitydive.com/news/worlds-largest-grid-battery...

https://www.opb.org/article/2026/06/18/federal-order-keeps-w...

https://en.wikipedia.org/wiki/THTR-300

But it was admittedly a huge cooling system for a reactor that only produced 750 megawatts of thermal energy.

https://en.wikipedia.org/wiki/Google_Books#Scanning_of_books

EDIT: I commented before reading the FT article. The comment by jackb4040 makes more sense to me now: some AlphaFold team members are moving to Anthropic, the company that has been in the news for shredding scanned books.

For all I know (I don't know a lot about biology), it's possible to develop a virus that spreads as easily as measles, but has the lethality and long induction period of HIV. If that's true, and if LLMs would make it easy to develop such a virus, I see how no-guardrails LLMs could be truly disastrous.

The other justifications for restricting model training don't seem nearly as compelling. Worse, restrictions on model training have an infectious quality if they mean something more than "leading AI labs will voluntarily slow down together." (Like if the signatories mean that better models and software for building them should be regulated.) It's too dangerous to publish stronger models. So it's also too dangerous to publish training software. So it's also too dangerous to publish tools that allow you to build training software. (Claude's Fable won't help you write generic model training pipelines for this very reason.)

If you keep following the implications of this safety argument, it's as broad an assault on the distribution of software and computing as has ever been proposed.

https://news.ycombinator.com/item?id=46313991

https://simonwillison.net/2026/Jun/10/if-claude-fable-stops-...

In light of the ability of recent models to accelerate their own development, we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design).

...

Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT).

(And although the "silent" downgrade part was quickly dropped, Fable still won't help you here.)

Anthropic won't teach you how to build bioweapons, or enable you to make your own software infrastructure so that you can train your own biology model. That's where lawmakers may arrive too if they buy Anthropic-style safety arguments. It's too dangerous to publish models that understand biology. It's too dangerous to publish training software. It's too dangerous to publish tools that allow you to build training software.

If you keep following the implications of their safety argument, it's as broad an assault on the distribution of software and computing as has ever been proposed. Worse than the Clipper Chip proposal of the 1990s era Crypto Wars. I have seen how "children must be protected online" has in practice turned into an attack on adult privacy affecting a wide swath of services and devices. I'm taking a maximalist position on openness now because I think that I can anticipate the next steps on the safety side, and I reject those steps.