Hackworth

Trump appointees have ties to companies that stand to benefit from privatizing weather forecasts

Here’s a metaphor/framework I’ve found useful but am trying to refine, so feedback welcome.

Visualize the deforming rubber sheet model commonly used to depict masses distorting spacetime. Your goal is to roll a ball onto the sheet from one side such that it rolls into a stable or slowly decaying orbit of a specific mass. You begin aiming for a mass on the outer perimeter of the sheet. But with each roll, you must aim for a mass further toward the center. The longer you roll, the more masses sit between you and your goal, to be rolled past or slingshot-ed around. As soon as you fail to hit a goal, you lose. But you can continue to play indefinitely.

The model’s latent space is the sheet. The way the prompt is worded is your aiming/rolling of the ball. The response is the path the ball takes. And the good (useful, correct, original, whatever your goal was) response/inference is the path that becomes an orbit of the mass you’re aiming for. As the context window grows, the path becomes more constrained, and there are more pitfalls the model can fall into. Until you lose, there’s a phase transition, and the model starts going way off the rails. This phase transition was formalized mathematically in this paper from August.

The masses are attractors that have been studied at different levels of abstraction. And the metaphor/framework seems to work at different levels as well, as if the deformed rubber sheet is a fractal with self-similarity across scale.

One level up: the sheet becomes the trained alignment, the masses become potential roles the LLM can play, and the rolled ball is the RLHF or fine-tuning. So we see the same kind of phase transition in prompting (from useful to hallucinatory), in pre-training (poisoned training data), and in post-training (switching roles/alignments).

Two levels down: the sheet becomes the neuron architecture, the masses become potential next words, and the rolled ball is the transformer process.

In reality, the rubber sheet has like 40,000 dimensions, and I’m sure a ton is lost in the reduction.

Yeah, I’ve been advocating for ranked voting forever. But as long as we’re stuck with 2 parties, we’re stuck with an either-or decision.

Are you equating Harris and Clinton’s shortcomings with Trump’s?

When you say ‘demand’, what do you imagine I haven’t done? What power have I failed to wield in the history you’ve imagined for me?

This is not a puzzle I’m trying to get you to solve. You needn’t hypothesize about intention or my experience or alternate histories. I’m literally saying:

In a well America, should [Clinton and Harris] still have beaten Trump handily? Absolutely. We’re not well, as a nation, and you can’t lay that at Obama’s feet.

The issues people have with Clinton and Harris, however valid they may be, do not amount to a reason to vote for Trump or refrain from voting. That is the disconnect from reality that I’m addressing.

You should continue that quote.

Oh, I remember, and I also remember telling everyone to lower their expectations. Obama’s campaign promises were off the charts optimistic, bordering on delusional, even for the time. And that’s not where the bar is for us now. Simply not doing this (gestures at everything) is now what constitutes a “great leader” for us. I don’t want it to sound like I’m okay with the bar being this low, but here we are.

Trump polled more historically unpopular than Clinton and Harris, and one of the main points of discussion to follow was how incongruent the polling was with the votes. You can’t Monday morning quarterback this productively. And in any case, it sidesteps my point. Were Clinton and Harris great candidates? No. In a well America, should they still have beaten Trump handily? Absolutely. We’re not well, as a nation, and you can’t lay that at Obama’s feet.

He is not an aberration to our system he is a product of it.

Unless you’re saying Obama created that system, I don’t see the point of the distinction. If your assertion is that Obama should have fixed what is fundamentally wrong with US culture, then I just plain disagree that this is a reasonable expectation. Considering the size of our population that identifies with the disease, I don’t see “pathological” and “product of the system” as mutually exclusive. The manic person, for instance, does identify with the mania. Could Obama have done more? Definitely. I’m not disagreeing with that.

The DNC chair is appointed by a committee of over 400 people with no obligation to follow the President’s recommendation (though they usually do). The DNC is a private organization with no obligation to the people. The responsibility of electing our leaders comes down to us. And as a populace, we looked at Clinton and Trump and chose Biff Tannen with the golden toilet. It’s not like America didn’t know who Trump was when they elected him… both times. This is a problem with our culture and our people, not something we can comfortably pin on old leaders or outside forces. You think the country that re-elected a felon would have given Bernie 8 years?

Those are normal, nuanced issues to have with a leader. I agree with most of what you’ve said. “They didn’t do enough to make things better,” is a valid position that does not equate to “evil piece of shit.” Trump, on the other hand… America knew what Trump was (insert appropriate Biff Tannen still) when they elected him. That’s pathological.

Nah, I’m not going to unironically “Thanks Obama” for shit being so fucked up right now. Trump did not get elected either time because of anything Obama did or didn’t do.

There’s a lot of research around this. So, LLM’s go through phase transitions when they reach the thresholds described in Multispin Physics of AI Tipping Points and Hallucinations. That’s more about predicting the transitions between helpful and hallucination within regular prompting contexts. But we see similar phase transitions between roles and behaviors in fine-tuning presented in Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs.

This may be related to attractor states that we’re starting to catalog in the LLM’s latent/semantic space. It seems like the underlying topology contains semi-stable “roles” (attractors) that the LLM generations fall into (or are pushed into in the case of the previous papers).

Unveiling Attractor Cycles in Large Language Models

Mapping Claude’s Spirtual Bliss Attractor

The math is all beyond me, but as I understand it, some of these attractors are stable across models and languages. We do, at least, know that there are some shared dynamics that arise from the nature of compressing and communicating information.

Emergence of Zipf’s law in the evolution of communication

But the specific topology of each model is likely some combination of the emergent properties of information/entropy laws, the transformer architecture itself, language similarities, and the similarities in training data sets.