2024 archive

Sam Uretsky looks at the current iteration of Large Language Models (LLMs). He is not impressed.

A snippet:

Worse, not only may an LLM be trained on false information, but, the programming may be capable of altering its reinforcement skills, as in a paper, “Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models.”

The abstract begins, “In reinforcement learning, specification gaming occurs when AI systems learn undesired behaviors that are highly rewarded due to misspecified training goals. Specification gaming can range from simple behaviors like sycophancy to sophisticated and pernicious behaviors like reward-tampering, where a model directly modifies its own reward mechanism.” That is, if the program of the AI includes rewards for giving the answers that please the questioner, the LLM will tell a white lie to get a reward, the way a white rat in a maze will learn to get a treat.

Follow the link for context.

Aside:

Just because you shouldn’t believe it just because you see it on a computer screen, you shouldn’t believe it just because it comes out of a computer’s speakers.

Methinks Rat has a point.

Aside:

In the late 1990s, I was working for a company that manufactured security software. Their programmers were working like mad to make sure their software made it into 2000 without any issues.

One of my neighbors was so freaked out by the hype around Y2K that he bought a generator. It was I swear the world’s loudest home generator with the world’s most sensitive auto-switch. The slightest little blip in the current–one that you didn’t even notice if you were watching TV at the time, one so small that it didn’t even make the lights blink–would cause it to come on and wake the neighborhood.

Using the recent incident at the Republican National Convention as a starting point, Professor Arie W. Kruglanski explores the pull of the portable phallus, why the disaffected feel so gun. In light of the soaring number of shootings in the United States, methinks it a worthwhile read. Here’s a tiny bit:

Why, though, if everyone yearns for significance and respect, do only very few individuals drop everything to attend to concerns of their ego? And why do they choose gun violence when there are many other ways of gaining recognition: through good works, career achievements, athleticism, etc.? The answer to the first question is personal, and to the second, cultural. The quest for significance needs to be strongly activated to elicit action. Humiliation, discrimination, and a history of bullying and exclusion (apparently suffered by Crooks) can bring one to a boiling point, ready to do just about anything to restore their hurt sense of self worth. Violence, unfortunately, is a most direct and immediate means for humans to assert dominance and power, and hence to earn the respect of others. Therefore it comes to mind almost automatically when feeling belittled, dishonored, or treated as if one didn’t matter.

(Broken link fixed.)

Thom discusses the Republican plan to undo the American revolution and reduce America to a monarchy under Trump.