btrmt. | Betterment

Interesting article on how AI thinks, by Anthropic. Some highlights:

Knowing how models like Claude think would allow us to have a better understanding of their abilities, as well as help us ensure that they’re doing what we intend them to. For example:

• Claude can speak dozens of languages. What language, if any, is it using “in its head”?> - Claude writes text one word at a time. Is it only focusing on predicting the next word or does it ever plan ahead?
• Claude can write out its reasoning step-by-step. Does this explanation represent the actual steps it took to get to an answer, or is it sometimes fabricating a plausible argument for a foregone conclusion?

And:

solid evidence that:

• Claude sometimes thinks in a conceptual space that is shared between languages, suggesting it has a kind of universal “language of thought.” We show this by translating simple sentences into multiple languages and tracing the overlap in how Claude processes them.
• Claude will plan what it will say many words ahead, and write to get to that destination. We show this in the realm of poetry, where it thinks of possible rhyming words in advance and writes the next line to get there. This is powerful evidence that even though models are trained to output one word at a time, they may think on much longer horizons to do so.
• Claude, on occasion, will give a plausible-sounding argument designed to agree with the user rather than to follow logical steps. We show this by asking it for help on a hard math problem while giving it an incorrect hint. We are able to “catch it in the act” as it makes up its fake reasoning, providing a proof of concept that our tools can be useful for flagging concerning mechanisms in models.

And:

In a study of hallucinations, we found the counter-intuitive result that Claude’s default behavior is to decline to speculate when asked a question, and it only answers questions when something inhibits this default reluctance.

Lots interesting. But I still think I disagree with calling this ‘thinking’. More here and here, but I still see no reason to believe this isn’t more like walking for an AI. On that account, these ‘thought processes’ would be more like adjusting to terrain, or something.

Stolen from a tweet. Not groundbreaking, but interesting thoughts on use-cases for the new OpenAI image model:

(1) This changes filters. Instagram filters required custom code; now all you need are a few keywords like “Studio Ghibli” or Dr. Seuss or South Park.

(2) This changes online ads. Much of the workflow of ad unit generation can now be automated, as per QT below.

(3) This changes memes. The baseline quality of memes should rise, because a critical threshold of reducing prompting effort to get good results has been reached.

(4) This may change books. I’d like to see someone take a public domain book from Project Gutenberg, feed it page by page into Claude, and have it turn it into comic book panels with the new ChatGPT. Old books may become more accessible this way.

(5) This changes slides. We’re now close to the point where you can generate a few reasonable AI images for any slide deck. With the right integration, there should be less bullet-point only presentations.

(6) This changes websites. You can now generate placeholder images in a site-specific style for any tag, as a kind of visual Loren Ipsum.

(7) This may change movies. We could see shot-for-shot remakes of old movies in new visual styles, with dubbing just for the artistry of it. Though these might be more interesting as clips than as full movies.

(8) This may change social networking. Once this tech is open source and/or cheap enough to widely integrate, every upload image button will have a generate image alongside it.

(9) This should change image search. A generate option will likewise pop up alongside available images.

(10) In general, visual styles have suddenly become extremely easy to copy, even easier than frontend code. Distinction will have to come in other ways.