Just like any other evolving software infrastructure.

Maybe it wasn't great at writing exploit code, but so what? Claude "found more high-severity bugs in Firefox than the rest of the world typically reports in two months, Mozilla said."

>> You need inert gas over it to keep it from burning

I'm thinking you did not read the linked articles. The sodium remains liquid and operates at near atmospheric pressure unlike water, which can explosively depressurize and boil off which could result in a meltdown. So in that sense sodium is safer. Yes it will react with air and water and that's a downside as you say, but apparently not insurmountable.

They claim it is a closed fuel cycle system and "Enough fuel for between 40 and 60 years of operation could be included in the reactor during manufacturing" so no need for enrichment or reprocessing.

'A TWR also accomplishes most of its reprocessing within the reactor core. Spent fuel can be recycled after simple "melt refining", without the chemical separation of plutonium that is required by other kinds of breeder reactors. These features greatly reduce fuel and waste volumes while enhancing proliferation resistance.'
https://en.wikipedia.org/wiki/...

I'm generally very skeptical about nuclear power but this project has some interesting features.

There's the 'travelling wave' technology;
'a small core of the enriched fuel in the center of a much larger mass of non-fissile material, in this case depleted uranium. Neutrons from the fission in the core "breeds" new fissile material in the surrounding mass, producing Plutonium-239.

Over time, enough fuel is bred in the area surrounding the core that it can undergo fission, enabling a steady-state reactor composition to be approximated by moving outer fuel rods towards the core as original core fuel rods are moved to the periphery'

The US reportedly has 700,000 metric tons of depleted uranium and "320 metric tons could power 100 million homes for a year".

Then there's the use of liquid sodium as a heat exchanger, potentially much safer than water. The heat is shifted to molten salt, which can be stored in tanks and used to generate steam on demand. A big plus.

Poor tradecraft.
"we noticed an instance where the actor deployed the debug version of the exploit kit, leaving in the clear all of the exploits, including their internal code names."

Apple has patched all the exploits, and automatic system updates were enabled by default in 2023.
"The exploit kit is able to target various iPhone models running iOS version 13.0 (released in September 2019) up to version 17.2.1 (released in December 2023)"

"A 7.2% decline for every tenth of a degree per decade might sound small," he added. "But compounded over time, across entire ocean basins, it represents a staggering and deeply concerning loss of marine life."

That's a massive die-off. And then there's the huge reductions of bird and insect populations. It sounds like ecosystem collapse and if that happens we are screwed.

The link to the study is right there in the report but hey, don't use AI. I sure don't care.

I do coding with various AI models almost every day. Claude 4.6 when I need help with something difficult, otherwise some of the less expensive models will do. They have helped me find quite a many subtle bugs in existing code. They won't do it all on their own of course, but they will write debug code for you if you know how to ask. They will explain exception messages and flag issues in error logs. I've had them spot and fix vulnerabilities, misconfigurations, race conditions. Really very helpful.

>> every actual study has shown AI code generation is 20% or more slower for senior engineers

Here's a Stanford study that doesn't;
https://medium.com/@manusf08/d...

"the study found that AI provides an average productivity gain of about 15–20% across all industries and sectors"

For myself it is far more. At least 4x, especially with the most recent LLM models.

"it can track how many lines of code an engineer wrote with AI assistance"

In my experience the IDE utility (Google's Antigravity, Windsurf, etc) has a dashboard that keeps track of how much code it "writes" compared to how much you did. This includes however much code you accepted in order to fix a bug, to refactor a module, or to document whatever you asked for. It also could be whatever experimental code you caused AI to write as a quick toss-off, which you can now easily do. So it can mount up pretty quick, and it should.

AI assistance is huge boost for people who are willing to learn it. People who aren't willing are a drag on the organization.

" the change to the RSP leaves Anthropic far less constrained by its own safety policies, which previously categorically barred it from training models above a certain level if appropriate safety measures weren’t already in place."

None of their rivals had adopted that ban.

"Instead, the Trump Administration has endorsed a let-it-rip attitude to AI development, even going so far as to attempt to nullify state regulations. No federal AI law is on the horizon."

I get your point but the code that I worked to implement, even though it was with AI assistance, was apparently unique on github. I know this because the AI models, which have ingested all publicly available open source, could not point me to ready-made solutions. I find that some guided experimentation is almost always necessary to come up with something that works properly.

As long as AI merely reads your open source code and gleans a few implementation tips from it, just like a human could do, there won't be a legal problem.

>> having all your open-source contributions digested into a pap

You can host open source projects on Codeberg, right? That makes them equally available for scraping.

>> Many will be less interested in contributing

I doubt it. I've posted a few open source projects to github recently, each of which I created in a day or so using AI. It was so fast and easy that I didn't mind sharing them.