It’s pure marketing. Very tiny. But very cool.

This may be a tiny experiment, planting a flag for a first. But, we live in the future, and it truly is full of wonder.

Meanwhile, back on Earth, electricity supply and distribution is a huge limiter on data center constructions (along with other tech, such as electric vehicles). Recently, Satya Nadella said some servers can’t be turned on because there’s not enough electricity.

Speaking of recent bit flip news with airplanes, in space that’s every day reality and radiation is a problem.

I didn’t find reference to shielding, but they did build software-level error correction and can shut down physically damaged cores.

Details:
• The computer is fully submerged in liquid cooling.
• Heat is radiated into space as infrared.
• Power comes from thin solar panels in a sun-exposed orbit.

Like many folks already stated, a lot of LLM success in offensive can be attributed to tool use, but there is much more to it than that.

There is no doubt that recent models have made a huge difference in capabilities.

Evals are extremely hard, and Irregular is doing a good job at it.

I was confused at first, but
believe they might be re-filling it live, which is a smart human engineering move that keeps me hooked.
But, a bit more verbose warnings would lower my stress levels. For now, it forces breaks.

Now I get warnings that I’m about to run out. Or later on that they moved to the extra charge. But…

It seems to me they changed something drastic. They give me the warnings, but then it resets!

Thank you :)

Complexity of problem: 5 versions, 30 days, 205K research tokens
Complexity of solution: 5 lines of JSON (settings.json)

{
"hooks": {
"PreToolUse": [{
"matcher": "Write|Edit",
"hooks": [{"type": "command", "command": "./.claude/hooks/pre-tool-use.sh"}]
}]
}
}