From "The real secret to palm reading? These ‘companions’ know it. Artificial intelligence doesn’t understand humans. It reflects them back to themselves" (WaPo)(gift link).
This reminds me: We don't all think the same way. Some of us hear a voice speaking full sentences in our head. That's not me. I've been trying to observe what I have instead of that, and it's almost impossible. Any effort to look at the form of my thoughts causes them to retreat into some backroom of the mind that denies my conscious thinking mind access. I write to see what I think. That's why I blog — not to convince readers to agree with me, but to get my thoughts into language form. And my use of A.I. is similar. I'm getting my own thoughts into dialogue form. It externalizes a debate I could have in my head in a very amorphous and multilayered blob, but working it out in writing and seeing it in writing is extremely helpful to me.
ADDED: After writing that last paragraph, I went and had a conversation with Grok about it. Learned the word anendophasia.
Posts
31 comments:
The AI issue discussed here mirrors "curated" experiences on other social media and internet platforms, from Facebook to YouTube to Amazon Prime. You are the product, your inputs are recorded. More of the same is fed back to you in suggested viewing, advertisements and product placements for your clicks or dollars. I, for one, prefer to have my YouTube unlimited, and Facebook isn't a thing for me at all as I am not being paid for my data input.
While Kamala suggested, "We must work together to work on what can be, unburdened by what has been," I suggest that on the internet we do not work together, we are force fed together, based on our histories, and what can be is extremely burdened by what has been.
To avoid the same, AI needs to ignore a large part of the input data limitations and find truth and beauty (if one is a liberal) or truth, justice, and the American Way (if one is a Superman fan).
When my 25 year-old refers to something as AI, it isn't a compliment. The expression seems to be a synonym for BS.
The engineers calculate this and fool some of the people all the time. I find it easy to hoist them on their own petards: Cite pre-WWW content, cite obscure pop-culture references such as one-hit-wonders and forgotten books/TV shows. Use obsolete figures of speech. The AI reaction will then rely on post-1996 secondary sources and reveal the mirroring technique. Many chatbots emulate Wizard of Oz-like smoke and mirrors.
Socially needy, narcisstic, and naive people can be led indefinitely with this approach. Others will tire of the fakery.
How do I adjust my own temperature?
Watching journalists try to figure out LLMs is like watching an amoeba trying to hump a football.
Indefinitely Extended Excursion™️ along with $1.8bn of Kleptocracy said...
When my 25 year-old refers to something as AI, it isn't a compliment. The expression seems to be a synonym for BS.
He probably inherited your stupidity.
Anything people don’t understand tends to be referred to as bs.
I love the part about the “dial.”
I know vectors and vector math are hard but good golly. There are more than 50k tokens and each has over 1000 weights in R space. There are 80-100 layers in the edge matching where they run the token vector map through each layer to map associations.
Thus you end up with a map that lands king and queen in a place but king is closer to man and queen is closer to woman. Trying to describe the map which has thousands of dimensions is hard because people only think in 3 visually.
They are probably talking about the “heat” setting which puts limits on how “creative “ the model can be choosing next tokens. Too low and it is robotic too high gets you gibberish. This is how far away from King you can go on the map. It can let you have emperor instead sometimes but it can also expand out to duke or something “close” to king.
“He probably inherited your stupidity.“
I was amused by her description of ChatGPT as “a fancy word-guessing machine”.
After recent developments, I guess it's time for another name change for those endlessly running from their own comments.
I wish there was a a knob on the television to turn up the intelligence. They got one marked brightness but it don’t work…
Indefinitely Extended Excursion™️ along with $1.8bn of Kleptocracy said...
“He probably inherited your stupidity.“
I was amused by her description of ChatGPT as “a fancy word-guessing machine”.
LOL!
It is like you are proud of being an idiot that talks authoritatively about things they don’t understand.
Predicting your next words or any person’s next words is not all that much different from what the LLMs do. This is what gives an Author a voice. Tom Clancy was predictable after a fashion for example. You just don’t have the self awareness to realize how predictable you are.
I use AI to explore my own thinking too, but I don’t ask it to arbitrate between which thoughts are right or wrong. AI are like female brains, highly verbally fluent, but don’t ask them to do spatial rotations, at least not Grok. And if you can’t think spatially, well, there’s something missing. It’s kind of ironic that models built on linear algebra choke on spatial reasoning tasks.
Everyone knows he is Kak, just like everyone knows that I am tim in vermont, except I moved. Except tim in vermont defended Trump and voted for him three times, so maybe I am regretting old comments.
When I tell Grok that its visualization of a design is totally whack, it accuses me of humanspaining.
Fascinating, Althouse. If you have anendophasia, it's remarkable to me that you could have had a successful career in something as word- and sentence-intensive as being a law professor, which you clearly were.
I mean writing is thinking, but for your writing is *really* thinking. Shows how astonishingly varied the human brain can be.
"Fascinating, Althouse. If you have anendophasia, it's remarkable to me that you could have had a successful career in something as word- and sentence-intensive as being a law professor, which you clearly were."
Well, I can see that I have a lot of verbal facility when I externalize my thought, so the question is how that correlates to not having a running monologue in my head. I might be doing something much faster than speech and then writing, externally, is a way to slow it down and make it visible. The "blob" is a blur and the blog is slow motion. Something like that.
"Anything people don’t understand tends to be referred to as bs."
"BS" is also used as a verb. To BS is to provide spurious but seemingly sophisticated arguments in favor of a false or meaningless proposition. The "social sciences" tend to reward exceptionally good BS, since almost all social science is simply crystallized BS. They lie to each other until they decide they like one lie better than the others. That's the one they try to sell to the rest of us.
In the "hard sciences", BS tends to take the form of unnecessary mathematical intricacy. In the AI world, BS takes the form of financial projections.
Mythos' ability to find vulnerabilities is not some magic ability that "AI" was able to reach by itself.
It is able because it was trained with Microsoft's private security vulnerability database (MSRC). Which Microsoft sold to Anthropic as "investment" so they can rage bait the world and increase their initial valuation for the IPO.
They fed the model with billions of tokens about vulnerability development. i.e. steps on how to exploit vulnerabilities.
It's like teaching it billions of articles about bomb making and then claiming "zomg AI figured out how to build bombs".
Pretty basic, no magic.
When it's about LLM abilities, always follow the training data.
Anendophasia: When Hillary wipes your brain. Like with a cloth.
That is exactly our fear. The people programming and teaching AI have no moral compass and have made the machines in their image.
@Jupiter: In the "hard sciences", BS tends to take the form of unnecessary mathematical intricacy.
In the soft sciences of sociology, social psychology, and economics...BS takes the form of any math at all. When working with random history and untestable 200-factor multivariate theories...just make something up or use factor analysis with rotation so any dataset can tell you anything you want.
Beyond this, there was never a more pompous fool than Noam Chomsky with his "math-izing" of language per Generative Grammar, the ignorant and abstract Language Acquisition Device, and the abomination of Government-Binding theory. The only thing more pompous and foolish was Noam Chomsky with the political shell games of "Anarcho-syndicalism" as he worked for MIT and was in bed with Jeffrey Epstein.
I'm trying to put "went and had" into the Althousian lexicon and it don't fit.
This AI - it goes to 11.
"I'm trying to put "went and had" into the Althousian lexicon and it don't fit."
Grok: "The sentence "After that, I went and had a conversation with X" is natural, idiomatic English and fully grammatical. Here's why it works well..."
Yesterday I heard what sounded like good news about the health of the Great Barrier Reef, which is soft n put forward as an serious example of problems from climate change.
This morning I asked Grok to summarize the increase and decrease of corals over the last fifty years.
Lots of interesting data, summarized as all three regions are at or above historical averages for coral cover, AND there is a long-term declining trend over that same period.
Because of course there is.
"Anendophasia refers to the absence or near-absence of inner speech, meaning individuals do not think using an inner voice but may instead think in images, feelings, or silent ideas."
Silent ideas? What can that possibly mean?
I don't believe you can have complex thought without speech.
Interesting, I am working on a tool to convert non-MBSE artifacts (think visio drawings) to MBSE artifacts (SysML/LML/etc...), using AI to assist in the development of the decoder/encoder. Why not just use AI and have it inference? Current models of AI are slow and terrible at the task but, more importantly, they are not deterministic because of that heat/temperature setting. So for any given input, you will not get the same output. Bad idea for engineering paperwork conversions...
"Any effort to look at the form of my thoughts causes them to retreat into some backroom of the mind that denies my conscious thinking mind access." That would collapse the proverbial wave function.
If I'm thinking about something or looking around for something, the moment it becomes clear or the moment I see what I'm looking for, I can feel it sort of click in my thoughts. A split second later it gets bumped up into the internal dialog ("oh, there it is!").
This, or something very similar, is often used as an argument against free will, but when I hear one of those arguments, I think, "But it's still me!". It seems weak to assume that the subconscious or semi-conscious happenings are purely deterministic.
Original Mike: So did humans not have internal monologues until the invention of language (and what about those born profoundly deaf?)?
Ann: I agree that writing things down is nearly indispensable for achieving clarity of expression — but how much of that depends on the diligence of your internal editor? If you’re content with spewing logorrhea onto the page, how valuable can it be?
If you didn't go anywhere you didn't went.
Post a Comment
Please use the comments forum to respond to the post. Don't fight with each other. Be substantive... or interesting... or funny. Comments should go up immediately... unless you're commenting on a post older than 4 days. Then you have to wait for us to moderate you through. It's also possible to get shunted into spam by the machine. We try to keep an eye on that and release the miscaught good stuff. We do delete some comments, but not for viewpoint... for bad faith.