Soldato
Can you explain the law of universal gravitation as if you're speaking to an American?
Chatgpt - Seriously good potential (or just some Internet fun)
- Thread starter Begbie
- Start date
Caporegime
- Joined
- 30 Jul 2013
- Posts
- 28,959
Marques Brownlee made a video about that
Soldato
I can see celebs being made redundant.I can see Celebs being in **** movies
Caporegime
I wonder how these are rendered. Are they traditional 3D geometry models or something else? The computational power for some of them must be substantial if traditionally rendered I'd have thought?
Caporegime
Fixed.
Associate
The GPU power needed for this is going to be astronomical. Wonder how much it costs when is goes public.
the google one has gone woke:
Not just that example but quite a few others:
Amusingly it will draw Asian people to include with Australia etc.. but an Asian person asked it to draw a Chinese woman and it refused:
I wonder if they're overcompensating a bit too much here as previous products from google have had racism issues!
Not just that example but quite a few others:
Amusingly it will draw Asian people to include with Australia etc.. but an Asian person asked it to draw a Chinese woman and it refused:
I wonder if they're overcompensating a bit too much here as previous products from google have had racism issues!
Soldato
"People born in Scotland on 1820" *generates black man* 
You have to laugh at forced diversity these days, otherwise you'd cry.
Modern AI is a truly wonderful tool, that is going to be ruined by censorship and diversity requirements.
You have to laugh at forced diversity these days, otherwise you'd cry.
Modern AI is a truly wonderful tool, that is going to be ruined by censorship and diversity requirements.
Caporegime
Exactly. Its going to rewrite "everything" in 25 years the next generation of kids won't have any reference points in reality"People born in Scotland on 1820" *generates black man*
You have to laugh at forced diversity these days, otherwise you'd cry.
Modern AI is a truly wonderful tool, that is going to be ruined by censorship and diversity requirements.
at least the Western ones.Who ever gets the dominant AI will be shaping the future and past.
I'm so glad I'll be dead in 30 years. These are insane times.
I think some of the above is the insertion of "diversity" prompts, like you're not interacting with the model directly rather you type a prompt but then Google adds some diversity words to the prompt then the model gets the modified prompt.
The model has a representation of say people in the 1800s in Scotland but it's been fed modified prompt to draw people of different ethnicities etc..
You can see an example here where it's refused to draw the image (presumably a restriction on drawing specific people after the Taylor Swift incident) and has replied back with a description, of course that description includes the (unnecessary for this image) diversity stuff:
They will probably adjust this and make it a bit more sensible in the future as it's more likely they want that stuff to apply to general current-day images but I guess they wanted to avoid negative PR upon launch. Like we're so conditioned these days with both adverts and with Netflix, BBC series etc. to expect it even in a historical context that things end up looking out of place without it so they'd probably get absolutely panned by tech journalists without modifying these prompts.
The model has a representation of say people in the 1800s in Scotland but it's been fed modified prompt to draw people of different ethnicities etc..
You can see an example here where it's refused to draw the image (presumably a restriction on drawing specific people after the Taylor Swift incident) and has replied back with a description, of course that description includes the (unnecessary for this image) diversity stuff:
They will probably adjust this and make it a bit more sensible in the future as it's more likely they want that stuff to apply to general current-day images but I guess they wanted to avoid negative PR upon launch. Like we're so conditioned these days with both adverts and with Netflix, BBC series etc. to expect it even in a historical context that things end up looking out of place without it so they'd probably get absolutely panned by tech journalists without modifying these prompts.
To date I have found little practical use for Chat GPT / RTX Chat or Bing and I'd be interested to learn what others are using it for.
That said last week I was tasked at work with writing 5 speeches for senior leaders - put the topics into MS Bing, it did a great job in less than 10 seconds and I received fantastic feedback (didn't tell them AI did it lol). I'd be more interested if it could write spreadsheets, program Power BI and prepare Powerpoints etc - possibly MS Co-Pilot can do this but that is a paid licence.
However, I am seeing many articles popping up today on the net that Chat GPT etc has already been made obsolete by a new AI LLM called Claude 3. I believe there is a free trial available on the Claude AI website
That said last week I was tasked at work with writing 5 speeches for senior leaders - put the topics into MS Bing, it did a great job in less than 10 seconds and I received fantastic feedback (didn't tell them AI did it lol). I'd be more interested if it could write spreadsheets, program Power BI and prepare Powerpoints etc - possibly MS Co-Pilot can do this but that is a paid licence.
However, I am seeing many articles popping up today on the net that Chat GPT etc has already been made obsolete by a new AI LLM called Claude 3. I believe there is a free trial available on the Claude AI website
Caporegime
Claude 3 has definitely not made ChatGPT obsolete. GPT 4 and 4.5 turbo are still pretty solid. Claude 3 OPUS is supposed to be better if your specific use case is programming, but it's not light years ahead. The release of GPT 5 shouldn't be too far off.
Last edited:
Not obsolete per se but Claude is now the best model at the moment (GPT4 was top for a while) and not just for say the specific use case of programming, really it's multimodal and beats everything else among a range of categories.
Caporegime
Only according to Anthropic's own benchmarks (which didn't include GPT 4.5 turbo), and as it was only released yesterday, I think we need to wait for some independent test results before we can say that it categorically beats everything.
I think you've misunderstood that a little bit, those aren't Anthropic's own benchmarks, for example, this is the graduate-level reasoning one:
So these results are comparing established benchmarks:
Someone has already tried the Norway Mensa test too:
Caporegime
This still doesn't confirm that it's categorically better than GPT, and as I said, it wasn't tested against GPT 4.5 turbo.
Claude3 should be better on average in most categories as it's a much newer model, but the real test will be now that it's out in the open.
Last edited:
It pretty much does indicate that though and there isn't a GPT4.5 turbo (yet), there's a turbo iteration of GPT4 and like I said those are public benchmarks - GPT4 turbo for example scores 92.5% on GSMK8, 54% on MATH and 73.17 on HumanEval see here:
Gemini 1.5 Pro vs GPT-4 Turbo Benchmarks
The evolution of AI language models is revolutionizing how we interact with technology. Among the latest advancements are Google’s Gemini 1.5 Pro and OpenAI’s GPT-4 Turbo. This article delves into a detailed comparison, shedding light on their capabilities, architecture, and potential impact...
bito.ai
And if you look at the anthropic's results GPT4 scores 92% 52.9% and 67% respectively for each of those, so turbo isn't a huge improvement.
But then if you look at Claude 3 it scores 95% 60.1% and 84.9% respectively for those tests.
No that doesn't follow, there have been plenty of other newer models released that didn't beat GPT4, beating GPT4 across a range of tasks is quite an achievement.