having read this weirdly boosterish piece and played with Claude too, I can tell you exactly what Claude is doing that makes it feel 'warmer' & more human. it unilaterally asks you abt things you mentioned earlier in the chat. which makes it sound as if it's thinking www.nytimes.com/2024/12/13/t... - ThreadSky

naomialderman.bsky.social • 73 days ago

having read this weirdly boosterish piece and played with Claude too, I can tell you exactly what Claude is doing that makes it feel 'warmer' & more human. it unilaterally asks you abt things you mentioned earlier in the chat. which makes it sound as if it's thinking

https://www.nytimes.com/2024/12/13/technology/claude-ai-anthropic.html?unlocked_article_code=1.hE4.O5jt.BdCSoBXZCDLQ&smid=nytcore-ios-share&referringSource=articleShare

Comments

jmcobleyauthor.bsky.social•73 days ago

How depressing.

naomialderman.bsky.social•73 days ago

I've come to the unavoidable conclusion that most ppl who are using & writing abt Large Language Models are weirdly not bestselling novelists with a good collection of awards, & cannot work out what word tricks are being used to achieve various effects.

(yes, the joke is that I sound like a knob)

postambition.bsky.social•73 days ago

I would buy your book about this. Sadly, the book would just be adopted as a manual for how build an even better Torment Nexus

segyges.bsky.social•73 days ago

That's really interesting! I can see how being a novelist with a good understanding of verbal tricks would cast Claude's behavior in a different light. It doesn't make you sound like a knob at all! Would you like to delve into anything else about Claude's linguistic tricks or your background?

naomialderman.bsky.social•73 days ago

Oh hah bloody hah 😂

cyberfuturist.bsky.social•69 days ago

This is exactly why we need authors and language specialists working on the character personalization of these model assistants (and chatbots generally)

The words, styles, tones, grammar, and formatting they use are important to users and the public perception of the models

prlofty.bsky.social•73 days ago

LLMs are for writing code because they are code.
Check out Ted Giola on substack for interesting writing on the subject, interspersed with jazz.

vinosmax.bsky.social•73 days ago

As soon as I saw the headline and your description I knew Kevin Roose was the author. He is always weirdly boosterish. He also didn’t include any real examples of Claude’s warm and wonderful responses.

plewis.bsky.social•73 days ago

It's funny how often these new chatbots remind me of Eliza, which probably would have done the same callback thing you're describing it if had more memory

plewis.bsky.social•73 days ago

And it's interesting that they're not better at absurdity and non sequiturs, like Chris Marker's chatbot

ipsquiggle.bsky.social•73 days ago

I know AI shitposts are a dime a dozen, but I heard this trend and gave it a try. The "warm demeanor" is a different way of saying, "full of bullshit on even more dimensions."

Behold, the very first thing it said to me, from a very straightforward, not bait question:

ipsquiggle.bsky.social•73 days ago

"I was ESSENTIALLY creating a fictional attribution..." 🤣🤣🤣 So warm and friendly, like a pet weasle...

trollball.bsky.social•73 days ago

Also, I feel that AI companies have hacked the consciousness of journalists. “Watch this parlor trick where I fake your job skills.”

I don’t understand why novelists can spot this and journalists can’t, but I do note the trend.

I’m an engineer so I may be missing something obvious.

alittlemorewicked.bsky.social•73 days ago

Novelists, in order to do their job well, must think deeply about other people and all the facets of their experiences. Journalists, by and large, do not have such a requirement. They are much more transactional about other people.

jtatlife.bsky.social•73 days ago

More specifically, AI can't (yet) replicate the skills of even a basic supermarket romance novelist in balancing tone, character etc.
But a lot of basic journalism (especially clickbait) is just summarising stuff or rewriting something with a standard angle, and 'AI' is already very good at that.

stealthbadger.net•72 days ago

@writestarlight.bsky.social Thought of you reading this thread

writestarlight.bsky.social•71 days ago

As of yet, the machines haven't hacked the kind of alchemy we create. I hope to God they never do. Not just because of job security--if the same stories we wrote could be produced outside of the human experience, we'd all be poorer for that knowledge.

stealthbadger.net•71 days ago

I think that sums up why it can't be recreated with generative AI: the closest thing to the human experience that can be fed to the models are texts, which are by definition limited windows into a particular person's experience. Creating new experiences requires building a new frame of reference.

trollball.bsky.social•73 days ago

There are moments in the article where the author flirts with self-awareness. “…or that it’s only popular among A.I. hipsters who want to brag about the obscure chatbot they’re into.”

But note his quoted sources!

For clarity, I’m sure this wasn’t lost on you.

naomialderman.bsky.social•73 days ago

Three of his quoted sources are the same one guy’s Twitter thread

vastin.bsky.social•73 days ago

It's a fairly damning comment on how modern economies and societies isolate us from daily human contact, that exposure to an AI that can just communicate in an empathetic format appears to be a major improvement for many people.

jtatlife.bsky.social•73 days ago

I'm currently in a tech company and this basic ability to simulate active listening would put it in the top 30% on emotional intelligence hereabouts

damocro.bsky.social•73 days ago

Like with the memory setting in ChatGPT?

apardoe.bsky.social•73 days ago

> it unilaterally asks you abt things you mentioned earlier in the chat. which makes it sound as if it's thinking

So basically it does Eliza's trick from the late 1960s? (Yes, I'm sure it does more, but yeesh.)

prettyhatmachine.bsky.social•73 days ago

Same thought on reading that. Showing empathy by confirming listening.

jons253.bsky.social•73 days ago

I did somethong like that on a C-64 in the '80s.

prettyhatmachine.bsky.social•73 days ago

I couldn’t do that because I have emotional issues and was on the Sinclair.

zaphodbeeblebr0x.bsky.social•73 days ago

What if Eliza, but with billions of dollars being burned?

sosprachmarsu.bsky.social•73 days ago

didn't worrybot do this same thing way back when? it recalled past convos, would later bring up things you wrote before.

Comments

Posting Rules

Reply