Without the system prompt, asking its name results in it responding with the name of the model they're ripping from. That would certainly draw your eyes to the right places.
Why is this? Do labs reinforce the model name during training? I was under the impression that this sort of "self-knowledge" always came from the system prompt, but I guess not...
Yes. In this case, during fine tuning. Other blurbs are also baked in during fine tuning that are perfectly reproducible from the Nex model. The details inside the linked issue are quite accessible.
Yes, and they are higher due to 2 reasons. And both seem to work.
Number 1 reason is that power was quite highly taxed, since it is directly linked with pollution and CO2. All powers, but that also means electrical power. The effect is, that european cars are smaller and use less. And also european houses are better insulated. You can measure this, the typical german 4 person household uses less than 50% of the electric power of a 4 person US household. Before COVID I even saw a statistics that this less usage compansated the higher electricity prices, so both norm-households payed the same for electricity. Unsure if that is still true post-COVID.
The other reason is that also a good amount of money is directly invested into the grid, to make it more resilient. And you can also measure that. If you lookup the SAIDI (system average interuption duraction index) of e.g. USA and compare it to Germany, you immediately see why over there uninteruptible power supplies are hardly used except in data centers. SAIDI Germany 12.2 minutes per customer per year, USA 125.7 minutes per customer per year. That's a whopping 10x worse. Not just as number, but also for the industry.
And I heard that the SAIDI in Texas is even worse than the US average.
The story is actually quite interesting. The Serbs observed that a nighthawk would routinely fly the same route but their radar couldn’t lock on it unless the missile hatch were open, which they managed to elicit.
In short, it took 2 rare events to occur for it to happen.
Radiation poisoning. Fossil fuels fall into the same category. The alternative is not nuclear vs. fossil. We should focus entirely on renewable energies. Of course Germany is now in a transition period, and there are a lot of conservative politicians who have been shying away from the high investments required for a fast transition. I think the main problem here is that the fossil-nuclear advocates shift the main problems to future generations (climate change, long-term storage of nuclear waste), the general public (large subsidies, minimal security standards, no or unsufficient insurance of power plants against desasters) or other countries (placing nuclear plants or waste deposits at the border, relaying on other countries for long term storing of nuclear waste). Together with extremely optimistic estimates, this makes their energy costs appear low on paper, when in reality the overall costs are immense. In contrast, there do not appear to be many cost elements of renewable energy installations that can be concealed or embellished. The reserve funds for their demolition are perhaps the only exception. But these only account for a small part of the costs at any rate.
This is not how it works. The interviewer knows 1-2 problems and there is no time for profiling since they are rushing through their day, probably focused on their day to day work. You are the least of their concern, believe me.
