Slop generally should be seen as negative. It lacks novelty, and relies on blending old concepts, while entirely avoiding exploration and novelty. Although people have such poor understanding of both creativity and intelligence that the dialogue that gets socialized around AI is pretty unhinged. Copyright is just a system altered so that old money oligarchs can hoard “human software” under the guise of empowering creators. This leads to things like the tumblr “you stole my style” wars. And now being a “real” artist means being an AI slop style intern for Disney types. Good art such companies make are due to the intern artists, DESPITE the corporate influence, but it’s always incentivized to be enshittified by the active corporate process.

Turing was trying to show many aspects of intelligence. Michael Levin from Tufts university is carrying on Turing’s work in morphogenesis, which is showing how concepts built in counterfactual spaces are used in a systems actively updating as they grow into new eco-niches. This lets you feel out the new space, even in abstract conceptual spaces that are detached from your sensory signals. This let’s you build abstract yet useful patterns depending on what parts of reality seem important for your type of system. Rudimentary eyes might not comprehend the sun, but it can use it as a binary gate for an internal mechanism to activate, enabling survival tools that rely on detecting day/night, even if those concepts aren’t comprehended in the heuristic. You build robustness through scale and diverse means to build perspective and weighting active Bayesian predictive systems on top of, inside, and around each other.

LLMs don’t have the active updating, so can’t grow into novel spaces, which is important.

They however are great at certain impulse probabilistic decision making, not unlike our ability to effortlessly spew coherent combinations if words, although we need another system to pay attention and halt that process when it makes mistakes, and also to direct salience. When you don’t add deeper novel patterns, but just the facade of patterns you’ve seen, you often get something that likes like generative slop, especially if you’re familiar with the deeper meaning that’s being inappropriately usurped to signal a fake competence. Like when AI mostly gets what a hand is, but easily gets confused due to a lack of robustness or ability to recognize what is actually wrong with an extra few fingers. When our own robust “checks” fail, we might start hearing voices that aren’t consistent with reality, as one mis-weighted system projects into another. Confabulation is a big part of intelligence, although its not inherently hallucinatory.

LLMs are definitely a part of human style intelligence. That’s why Geoffrey Hinton talks about how our brains “ping.” turing has an amazing understanding of intelligence, especially for his time. Intelligence is also many things, as by its very nature is a growing grab bag if heuristics to call on or balance on. As hinton suggests, the layman comprehension of intelligence is flat earth level of misrepresented.

I think if Turing were alive he’d be in love with current academic dialogues, and he’d probably be a furry.