Recent content by uppi

Thread length in this graph is number of posts instead of time, I believe. And it is normalized to the maximum length for each source, so the maximum length can be from a few hundred posts to a few hundred thousand posts.

Nocebo effect? You are asking for an evidence-based alternative for something that has itself no evidence. Not finding the former has no impact on the validity on the latter (especially when the evidence for a microwave attack should be fairly simple to collect)

"There is no God but Trump and Trump is his prophet"?

It is slightly more complicated than that, because the antenna design and position might matter. But, yeah, if the electronic intelligence gathering team at those embassies only have the slightest degree of competence, they should have clear-cut evidence of any such attacks.

A few gigawatts - more than a typical nuclear power plant produces.

Which one, though? Those branches tend to be quite diverse.

I certainly do. But as long as I could go on about the misrepresentations, oversimplifications and taking for granted what has not been settled (and may never be settled), my problem with the article is much more fundamental: He confuses Naturalism with Realism. He says that quantum mechanics...

I my opinion, in a FPTP system, a vote for a party without any chances is effectively half a vote for each of the leading parties.

And yet, the presidency changed hands all the time during all those time periods. Structural advantages for one party in the election of a certain branch of government may mean that even with a roughly even split, control tends to remain with one party. In terms of popular vote Biden's election...

I don't think so. Increasing the number of voters would certainly upset the current balance, but it would over time revert to close elections. The steady state of a two-party system is a roughly even split. Anything else and you have a de facto one-party system with all its consequences.

The model cannot generate correct output with certainty. So the only option would be to teach it to refuse to answer to anything vaguely election related. I am not sure this is a good idea, especially since there will be people around who will train their models with deliberate misinformation...

There are none, but that does not stop people. LLMs are not programmed in the traditional sense. The underlying software is, of course, but the output depends on what the model has learned from the training data. It gets the start of a conversation and then tries to predict how it will end. So...

Given the vulnerability of LLMs to prompt injection, that looks a little like an open invitation to get reverse-hacked. Also, I wonder whether using obscene names in your database scheme to intentionally trigger OpenAI's content filters would be an effective defense.

I am not sure I completely agree with dishonesty. I would say it is more about whether politicians are held accountable for their actions. Secrecy is obviously a big component of this, especially in functional democracies. But if someone has a firm enough grasp of power that they cannot be held...

But then he would have to give a vision of a peaceful settlement, which is the one thing he and his supporters have refused to give at all points of this war. Which is very strange to me, because pretending that Russia seriously wants peace would sap quite some support in the West. Somehow they...