The sheer stupidity of using the r’s in strawberry question as evidence of LLM incompetence

The meme of asking an LLM how many R's are in the word strawberry went viral because of public ignorance about how LLMs work. An LLM doesn't even see the individual letters in a word. It takes its input in the form of word-sized chunks. As such, questions relating to individual letters, like asking it to count the number of letters or asking which words start with a specific letter, are uniquely difficult and completely unrelated to performance on other tasks in general.

If you think about it, the only way it can even infer strawberry has any r's at all is purely based on context clues from somewhere deep in its training set where someone may have spelled out the word "b-e-r-r-y" with dashes in some random blog or forum post. Therefore, the fact they even sometimes get these types of questions right should be considered nothing short of amazing.

On Generative AI Art: A Prompter is not an Artist (yet)

I have heard a number of people claiming that they are creative authors of these works of art because they are refining their prompts until they get what they want. However, this fails the litmus test: Isn't refining a prompt still more similar to what a manager or client would traditionally do? They'd tell the artist what to make, the artist makes it, then they tell it what edits they want, etc. until the piece is done to their liking. Therefore a prompter's role is still more like a client who hired an artist, the artist in this case being the AI.

Additionally, I propose a way to objectively measure creative contribution, without any technical/knowledge gatekeeping nor requiring it be done a certain way; a mathematical definition of creative authorship:

The fewer variations of an output that can match your spec, the more you contributed to the creative process.

Consider the 3 scenarios: A classical piece with all notes fully written out; a jazz piece which allows significant improvisation from the soloists; an AI-generated piece where someone prompted “Epic orchestral soundtrack, sci-fi battle, heroic with French horn melody, high-quality”.

1. The composer already restricted the subspace of possible sounds to a small slice, so they’re recognized as contributing to the majority of the creative process. There is still room for interpretation for dynamics, expression etc., and the musicians are recognized for putting their personal touch when performing these pieces, reducing the subspace from that small slice to that unique recording.

2. There is much more leeway for interpretation via improvisation. In any particular recording, the composer and the soloists are all rightfully recognized as having contributed significantly to the work, because the soloists are composing important parts of the piece on the fly.

3. It is such a vaguely specified instruction that there’s an uncountable number of wildly different pieces that could’ve fit the requirements. As such, your role as the prompter was more like a client who hired the AI to do all the creative work for you, rather than an actual artist.

This concept can be applied universally to both music and visual arts. It also leaves the door open in the future for a prompter to legitimately be an artist; they'd just have to be incredibly specific with their instructions and the model would need to be very good at following those instructions accurately. 

Defeat Sleep Paralysis by Snoring

 I am surprised that this does not seem to be documented anywhere on the internet (or maybe search engines just suck these days), but I discovered a few years back that eye muscles aren't the only thing you can control during deep sleep. There is a muscle in the back of the throat that can cause snoring, which you can also control while lucid dreaming or while in sleep paralysis. I activate that muscle to try to cause the loudest snore possible and wake myself up within 2 snores. I have been using this method to great success most of the time, except today where even after 2 extremely loud snores I still did not wake up, which was quite disconcerting.

I was then awoken by my wife loudly complaining about my snores, which confirms that the control over my snores was real rather than imagined.

The human brain: Is it actually intelligent?

The following is a satire of trendy articles explaining that deep neural nets such as LLM's are "just" predicting the next token, as if it were some sort of grand philosophical insight.

Interviewer: The human brain has captivated the masses. A lot of people are thinking it exhibits true intelligence. What are your thoughts?

Expert Bob: Well, it can certainly seem that way, to be sure. But if you're familiar with how the human brain actually works, you'll know it's just a façade.

Interviewer: Can you explain to us how it works?

Expert Bob: Of course. What happens is there are input sensory neurons, taking in signals from the eyes, ears, nose, and skin. All these electrical signals get processed in the brain, which is at the end of the day just a giant network of neurons. Finally, the brain outputs some signals which get sent as muscle activations.

Interviewer: So you're saying all it does is predict the next best set of muscle activations?

Expert Bob: If you peer into the brain with a microscope or high-tech scanner, you can trace every electrical signal down to its physical causes. It's a very complex biological machine, but ultimately predictable and devoid of true intelligence.

Interviewer: What are your thoughts on the fact that they seem to be capable of reasoning about morality, long-term planning, object recognition, and other tasks thought to require intelligence?

Expert Bob: Imagine you are talking to a human, you ask the human a question about an image you showed them. They respond with a seemingly intelligent answer. You have to understand all their words are just the result of a clever series of muscle activations in their tongue, mouth, and larynx, timed correctly. Basically, the brain is a clever invention that's programmed only to answer a single question: "What is the next best muscle activation?"

Interviewer: So when the ears hear something, is the brain not hearing actual sound? When the eyes see something is it not seeing a real image?

Expert Bob: The brain takes sound waves as input, but it isn't hearing sounds. What's fed into the brain isn't sound; it's just encoded electrons and ions. When the eyes "see" a picture, it's not actually seeing an image the way we see it. It's passing it to the brain as a rapid-fire series of "on" or "off" voltages, similar to how a computer works.

Disclaimer: This piece of satire is not meant to claim that deep neural nets exhibit true consciousness or are as smart as human brains. Rather, I'm simply debunking the notion that you can prove something has absolutely no "true intelligence/understanding" just because it's programmed to predict the next word.

Stop saying chatGPT doesn’t actually know anything

 A common refrain is to say GPT-3 and ChatGPT “don’t actually know anything” because they’re only programmed to predict the next word. That’s like an alien observing that the human brain is only programmed to maximize long-term happiness and hence doesn’t actually know anything.

The correct way to determine whether something “knows stuff” isn’t to theorize about what is or isn’t possible based on how it works, but rather, to scientifically test its knowledge. And GPT-3 technology has already been scientifically proven to perform at the state of the art level for common sense benchmarks such as Q and A, reading comprehension, and SAT questions. It’s nowhere near the human level, but often outperforms previous AI’s specifically engineered for those tasks, even though its only directive was to “predict the next word”. In short, it turns out that when something is good enough at “predicting the next word” it starts to gain emergent logic and reasoning that conventional wisdom would’ve held was “impossible” for something just programmed to predict the next word.

Instead of asserting something can’t possibly know stuff if it’s only programmed to predict the next word, we should be amazed that something only programmed to predict the next word can already know so much stuff.

Expecting quantum mechanics in the brain to explain consciousness is a fallacy

 A very popular theory floating around seems to be that the mysteriousness of our consciousness lies in quantum processes within the brain, and eventually science will discover that this is the true source of consciousness, and it will explain why consciousness happens. 

The overarching motivation which seems to drive most proponents of this idea, is the observation "consciousness is unsolved/mysterious, and quantum physics is unsolved/mysterious; therefore the two must be related".

In order to understand why this specific line of reasoning is a fallacy, let’s review what the Hard Problem of Consciousness really is (please read the linked post). Now that we’ve established why the hard problem of consciousnes is considered unsolvable, can you come up with a hypothetical way for quantum mechanics in the brain to explain consciousness? Spoiler alert: No you cannot, because no matter what objective process you observe in the world, it still has to bridge the gap to the subjective side. No matter what science discovers is the “true” source of consciousness, be it a quantum field or spirit metamaterial or literal magic dust, we’re back at square one, saying, “that’s nice, but why did that thing cause your subjective inner mind to appear out of nothing?”

Rant: Naysayers focus too much on what AI can't yet do

On almost every comment section about generative AI technology (be it with text, images, audio or video), one doesn't have to scroll very far to see comments like "Look how it still can't draw the hands correctly; I'm not worried about art any time soon". Okay, so it still has a ways to go, but what about the stuff that it did accomplish over the last 5 years which many people thought would be impossible? If you want to predict a trajectory, don't you need to take into account how much it's accomplished so far, instead of fixating only on what yet remains to be accomplished? Did people just magically forget that conventional wisdom held for decades that computers should never be able to create new interesting images, at all?

To me when people say this it sounds like someone looking at a snapshot of a car headed towards a cliff, and going like "oh, it's still 1,000 miles away, so we got a lot of time before we need to start worrying", without taking into account the fact that only 1 hour ago it was 2,000 miles away.