ChatGPT

[Gavster]

Been playing around with this chat AI and it's pretty amazing, the kind of responses it can create are mind blowing. Apparently it's great at coding too. Well worth playing around with.

https://openai.com/blog/chatgpt/

I asked it to write a poem about BMW and here's what it said (copywriters better watch their asses)

The sleek and shining car,
A symbol of luxury and power,
A BMW, a work of art.

With lines so smooth and sleek,
It glides along the road,
A shining testament to German engineering.

The engine roars with strength,
Pushing the limits of what is possible,
A symphony of power and precision.

Inside the cabin, luxury reigns,
With leather seats and polished wood,
A haven of comfort and sophistication.

The ultimate driving machine,
A BMW, a work of art,
A symbol of excellence and style.

[Beany]

Been playing around with this chat AI and it's pretty amazing, the kind of responses it can create are mind blowing. Apparently it's great at coding too. Well worth playing around with.

Oh, it's not.

https://www.theregister.com/2022/12/05/ ... s_chatgpt/

[Explosive Newt]

It basically bullshits to a high level. It will churn out screens of poetry or the sort of copywritten stuff you might put in an advert but ask it a technical question and it will put out something that *sounds* plausible but isn’t.

So the people who do copywriting for a living should be afraid but most professionals can sleep easy for now.

[integrale_evo]

Elon musk been using it for years

[Jobbo]

Merry Christmas!

[Broccers]

It basically bullshits to a high level. It will churn out screens of poetry or the sort of copywritten stuff you might put in an advert but ask it a technical question and it will put out something that *sounds* plausible but isn’t.

So the people who do copywriting for a living should be afraid but most professionals can sleep easy for now.

It's something I am actually starting to worry about to be honest

https://www.livemint.com/technology/tec ... 44505.html

[Rich B]

Have you seen it?

[JonMad]

Tried to get on there yesterday to have a play but signups are currently disabled.

[Gavster]

I've been using it more, it's really good for smashing out low-level tasks, such as creating social media posts, basic brainstorming around a topic, summarising text, editing text. I tried to get it to write the exec summary for a big report that I was working on, however the outputs weren't nuanced enough to do it. For a small, internal report I'd use it all day long.

The biggest challenge has been modifying my language from the kind I'd use to query Google, to talking like I would an intern, giving it tasks with multiple specificities, criteria and boundaries for the outputs.

As mentioned, it's definitely just killed off a load of basic copywriters.

[Mito Man]

Mentioned it in the Trump thread but China has jumped on the AI bandwagon. I asked it a few straightforward questions and it gave an answer that you would expect from something Chinese. Sometimes it begins typing an answer then deletes it!

[Simon]

Nvidia is down over 8.5% on this news that this model can run on older hardware.

[Swervin_Mervin]

Mentioned it in the Trump thread but China has jumped on the AI bandwagon. I asked it a few straightforward questions and it gave an answer that you would expect from something Chinese. Sometimes it begins typing an answer then deletes it!

You're now on a watch list

[Ascender]

That's amazing

[ZedLeg]

Openai crying about Deepseek violating its ip is incredibly funny.

[Gavster]

Oh how funny that a platform built upon the mining of other people’s copyrighted material without credit or payment is now worried about their work being copied

[Sundayjumper]

"no fair you plagiarised our plagiarising machine"

[DeskJockey]

This doesn't feel scary at all...

https://arstechnica.com/tech-policy/202 ... aint-says/

Publicly available GenAI is a binfire waiting for someone to come along with a tanker full of petrol. It'll go very badly wrong.

Closed models/restricted training sets are a bit less bad although I'm very unimpressed with output from the one at work. It misses *a lot* of information when you ask it to extract data based on themes or keywords. Meaning, it can't be trusted, and that it'll become a contractual risk at some point.

[Jobbo]

Virtually every day I get e-mails from our practice management software provider inviting me to AI training sessions. I have made very clear to them that I am not going to use AI in my firm - humans don't make up the law, in general, so anything produced by AI is going to be less trustworthy than the product of the most junior member of staff.

I even attended a sales call from them trying to pitch AI. I gave them short shrift. If AI becomes a reliable and useful tool for law firms, it will have first been tested and refined in much bigger practises than mine; I am certainly not going to use my time to be their guinea pig.

Oh, and thus far most things labelled 'AI' in law are simply the old tools for filling in the blanks in a document to produce a first draft, but relabelled.

[Gavster]

I’ve been using it a lot more recently as a blank page filler. E.g. a way to quickly spitball some ideas that I can take away and make my own. I’m working on making my prompts to get better results, but a lot of the outputs are still generic trash to me.

[dinny_g]

I've started using it more and more to write queries and export routines for me from SharePoint - a tech I know very little about (Oh to be back in Toad and Oracle...)

Doesn't always get it right but it generally gets me 90% of the way there