ChatGPT Examples, Good and Bad

[fresh_42]

Now that they see the problem in how models process math, I would assume that people will start working on getting them to reason over the problems instead.

I wonder how they will deal with the setup. The equation $2+2=4$ does not hold everywhere, and I haven't seen an example yet of a setup that is sufficiently described. The automatism to assume integers (reals) is all over the place, but it is a requirement, not a property, and nobody mentions it.

 

[Borg]

I wonder how they will deal with the setup. The equation $2+2=4$ does not hold everywhere, and I haven't seen an example yet of a setup that is sufficiently described. The automatism to assume integers (reals) is all over the place, but it is a requirement, not a property, and nobody mentions it.

From the article, they noticed some correlations in the landscape for memorized and reasoned processes. Assuming that the correlations hold up, I would guess that the next step would be to adjust the backpropagation step to penalize the sharper memorization landscapes when dealing with math.

 

[Borg]

Here's a good LLM example. I have to create a Programming Guide for the architecture that I designed and built this year. Last week, I threw sections of the code at an LLM and asked it to generate the guide. I tossed the results in a doc and hadn't reviewed it much though. This morning, I was working on a new factory method in the code and thought that would be good to add. When I reviewed last week's output, there was a nice section on generating the new factory that I'm working on!

 

[jack action]

Now that they see the problem in how models process math, I would assume that people will start working on getting them to reason over the problems instead.

I have a problem with mixing the terms "reasoning" and "LLM". I'm reassured with this quote from the same article:

It’s worth noting that “reasoning” in AI research covers a spectrum of abilities that don’t necessarily match what we might call reasoning in humans. The logical reasoning that survived memory removal in this latest research includes tasks like evaluating true/false statements and following if-then rules, which are essentially applying learned patterns to new inputs. This also differs from the deeper “mathematical reasoning” required for proofs or novel problem-solving, which current AI models struggle with even when their pattern-matching abilities remain intact.

 

[Borg]

Pakistani newspaper mistakenly prints AI response with the article.

 

[Borg]

On some AI platforms, users are marrying and having virtual children with their chatbots.
https://www.psypost.org/new-study-f...and-having-virtual-children-with-ai-chatbots/

 

[fresh_42]

On some AI platforms, users are marrying and having virtual children with their chatbots.
https://www.psypost.org/new-study-f...and-having-virtual-children-with-ai-chatbots/

It can at least create a typical conversation:

"I didn't know."
"I told you last week, Monday, 3:15:25 pm."

 

[Hornbein]

On some AI platforms, users are marrying and having virtual children with their chatbots.
https://www.psypost.org/new-study-f...and-having-virtual-children-with-ai-chatbots/

That's why I stopped reading science fiction. Real life is much more weird and imaginative.

 

[Borg]

It can at least create a typical conversation:

"I didn't know."
"I told you last week, Monday, 3:15:25 pm."

It won't truly be like a wife until it can remind you of the toilet seat that you left up two years earlier.

 

[gmax137]

Maybe that's the new Turing test: does the chatbot male partner forget anniversaries, leave the seat up, pretend to listen while watching football...

 

[DaveC426913]

Maybe that's the new Turing test: does the chatbot male partner forget anniversaries, leave the seat up, pretend to listen while watching football...

I had a story idea kicking around about the idea that the last barrier between androids becoming human ultimately required them to be able to misremember and even forget stuff.

Data, on Star Trek, for example, cannot understand humanity, or become human, until his memory and thought processes are messy and unreliable.

And of course, once that happens, we're removed the big advantage of androids; they're as imprecise as humans.

A sort of 'you can't have your cake and eat it too' kind of thing. (shades of Twilight Zone's 'The Mighty Casey'.)

 

[nsaspook]

https://www.scworld.com/news/copy-p...fer-as-top-corporate-data-exfiltration-vector

Copy-paste now exceeds file transfer as top corporate data exfiltration vector​

It is now more common for data to leave companies through copying and paste than through file transfers and uploads, LayerX revealed in its Browser Security Report 2025.

This shift is largely due to generative AI (genAI), with 77% of employees pasting data into AI prompts, and 32% of all copy-pastes from corporate accounts to non-corporate accounts occurring within genAI tools.

“Traditional governance built for email, file-sharing, and sanctioned SaaS didn’t anticipate that copy/paste into a browser prompt would become the dominant leak vector,” LayerX CEO Or Eshed wrote in a blog post summarizing the report.

 

[Hornbein]

https://www.scworld.com/news/copy-p...fer-as-top-corporate-data-exfiltration-vector

Copy-paste now exceeds file transfer as top corporate data exfiltration vector​

So who's the bad guy looking at that data? Do the AI companies sell it?

 

[DaveC426913]

So who's the bad guy looking at that data? Do the AI companies sell it?

They don't need to sell it raw. That would go at "wholesale" prices. They parcel it out with every use of the their product, meaning the can effectively charge "retail" prices for it.

Think of it like an ore mine. Rather than selling off the raw ore by the ton, they can keep it as their product, and sell each rock on-by-one, at a fabulous markup - as if they carved the rock themselves.

 

[WWGD]

Well, I wrote a small review of recent material, submitted to ChatGpt just to edit. I accidentally submitted a second time without any changes of my own. ChaptGpt suggested several edits, changes to the edits it had itself made.

 

[Borg]

AI-generated clocks
https://clocks.brianmoore.com/

Every minute, a new clock is displayed that has been generated by nine different AI models.

Each model is allowed 2000 tokens to generate its clock. Here is its prompt:

Create HTML/CSS of an analog clock showing ${time}. Include numbers (or numerals) if you wish, and have a CSS animated second hand. Make it responsive and use a white background. Return ONLY the HTML/CSS code with no markdown formatting.