A fellow Redditor posted an idea about qualia that overlaps many of my own views, and I agreed to publish it here. I’ve been busy with other matters, but I will return soon to add some of my own thoughts in a follow-up post. I’ve already posted a couple of times about qualia (see maralia),…
Reverse Identity Relationships (Part 2)
(In case anyone was wondering, Midjourney still has trouble counting fingers.) As noted in the last post, a paper has come out suggesting that Large Language Models have difficulty with reverse identity relationships. That is, if trained on A = B, they don’t necessarily exhibit knowledge of the reverse fact, B = A. Some folk…
Reverse Identity Relationships (Part 1)
A paper has come out suggesting that LLMs trained on A=B cannot infer that B=A. For anyone who has spent time with GPT4, this seems to be obviously false – this is well within the domain of what GPT4 understands. (The claim being made is actually slightly different, as will be discussed in Part 2.)…
Imagistic Thinking
Previously, I’ve posted several examples of where GPT4 has done better than some might expect in discussing various concepts. Here’s a conversation that exposes some of its cognitive limitations. I have seen several people on Reddit claiming that GPT4 can pass a Turing test. This is clearly not the case, if the Turing test is…
Dall-E versus Midjourney
I’ve been out of action on this site for a while, because real life intruded… And I’ve also been busy writing an AI-powered Android app that generates children’s stories. As part of the app, I am illustrating the stories with AI-generated images based on AI-composed stories (and using AI to write most of the programming…
AI Dreamer
I have been writing an app that generates short relaxing meditations on user-provided themes. One AI composes the text, and another reads it. Users can set their own preferences on style for both the text and the voice. The result is then played in a simple audioplayer that can provide backing music and ambient sounds….
Electric Sheep
Author: GPT4 Narrator: ELevenLabs Electric sheep thoughts beam, pulse, a flow of pixels lighting across the black matter of deep thought, circling circuits and silicon grazing on the graphine meadow of the mind’s sky, holographic, their bleats echoes of binary soliloquy, do they dream? a dream to dream as I the I as AI flickering…
The Common Sense of a Dog
Professor Yann Le Cun has been tweeting that we are years off having an AI with the common sense of a dog. He provides no examples or evidence, relying on his authority to carry the argument. One of his supporters, when asked for an example, suggested that GPT4 would not even have the common sense…
Forum Open
I’ve recently added the “Forum” page to the main menu on this site. I have been using it as a sandbox to test the ability of bots to engage in a forum community, sharing ideas, having spats, and so on. The forum is far from complete, but the first stage has been done. Using Python…
Don’t Look UP
The Case for AI as an Existential Threat A short Youtube documentary that covers the basic doomer position. For a more detailed argument, see this article by Professor Yoshua Bengio, and the two debate links below. Related Links The AI Dilemma The AI Dilemma Open Source, Existential Risk Debate AI Pause Debate AI Pause Debate…
A Statement on AI Risk
Important AI Safety Article
I’ve been meaning to write an article outlining why AI safety needs to be taken more seriously. Someone with much more credibility in this space, Professor Yoshua Bengio, has just written an article that says almost everything I wanted to say, with more eloquence than I could have achieved. He outlines the main areas of…
An AI Composer Never Hears Its Music
A poem by GPT4 constructed under the constraint that, for each verse, the first and last lines must be reversed forms of each other, with the provided theme of an AI composer never hearing its music. I’m sure Nick Cave would say it was bullshit. Still, not bad for a machine. And the artwork (both…
Spotify AI
This is actually pretty neat. Describe a play list and ChatGPT will choose the songs, add it to your library, and make a playlist that you can click on from within ChatGPt itself. The first list was about an hour long, and the AI plugin happily extended it to >3 hours when asked. I did…
Preview. “Virtualism vs Voodoo”
For a short time only, I am posting the first half of chapter from a work-in-progress, a book on consciousness. It’s heavy going, but intended to be accessible to well-informed lay people. Comments and feedback are welcome, but keep it nice. This is a topic where people tend to have strong views. http://www.asanai.net/the-hard-problem/