Claude is better IMO, but as another poster mentioned your prompts are important too.
You can also use these within IDEs now - there's the DIY option via Neovim and also there's Cursor (which is basically Visual Studio + Claude).
Chatgpt - Seriously good potential (or just some Internet fun)
- Thread starter Begbie
- Start date
Claude is better IMO, but as another poster mentioned your prompts are important too.
You can also use these within IDEs now - there's the DIY option via Neovim and also there's Cursor (which is basically Visual Studio + Claude).
Anyone tested o1, the strawberry/q-star reasoning engine?
If it can reason, it could sift through hundreds of scientific papers and PHDs. Sabine Hossenfelder just did a video going through the steps taken to peer review. Some stuff like spelling errors and using the same symbol to mean different things seem like easy catches. Presumably if you asked it to write a thesis, the extra thinking time (up to 30 seconds) would lead to much better quality work than the GPT4 theses spouted out, although, the 30 second time limit, that it cuts corners if it gets near, looks like a bottleneck.
/cdn.vox-cdn.com/uploads/chorus_asset/file/25462005/STK155_OPEN_AI_CVirginia_B.jpg)
If it can reason, it could sift through hundreds of scientific papers and PHDs. Sabine Hossenfelder just did a video going through the steps taken to peer review. Some stuff like spelling errors and using the same symbol to mean different things seem like easy catches. Presumably if you asked it to write a thesis, the extra thinking time (up to 30 seconds) would lead to much better quality work than the GPT4 theses spouted out, although, the 30 second time limit, that it cuts corners if it gets near, looks like a bottleneck.
Last edited:
FFS question from previous link V (fortunately didn't scan the answer - the cookie msg had covered it - so, don't open it without taking precautions )
your reasoning challenge for today
If that model can reason that in 30s it's taken my job - took me an hour on and off yesterday
can't work out if Apple is mocking it's Apple intelligence 'AI' capability, people on the spectrum, or, the whole advert is AI generated based on a friends situation comedy
suppose it would be interesting to give it some threads and ask it to summarize posts etc.
your reasoning challenge for today
If that model can reason that in 30s it's taken my job - took me an hour on and off yesterday
can't work out if Apple is mocking it's Apple intelligence 'AI' capability, people on the spectrum, or, the whole advert is AI generated based on a friends situation comedy
suppose it would be interesting to give it some threads and ask it to summarize posts etc.
That and the dead bunny advert is painful as well.
I do laugh at the feature now of adding in the photo taker into the same photo, by taking a 2nd photo of them and the AI doing it's thing, actually quite nifty. Given the way things are going with isolationism of younger generations though, I would imagine the next gen iPhone or whatever it was it does it, will straigh up just AI in a fake person to your photos.
I do laugh at the feature now of adding in the photo taker into the same photo, by taking a 2nd photo of them and the AI doing it's thing, actually quite nifty. Given the way things are going with isolationism of younger generations though, I would imagine the next gen iPhone or whatever it was it does it, will straigh up just AI in a fake person to your photos.
I use chatGPT to write boring JSON synchronising code for me. The other day it made me a rather nice regular expression after giving it some sample text to process.
We use Github copilot at work as well, it's pretty darn good. Though often it gets the wrong end of the stick, but the amount of time it will just write entire functions for you using all the right variables, in your own style, it's quite impressive.
We use Github copilot at work as well, it's pretty darn good. Though often it gets the wrong end of the stick, but the amount of time it will just write entire functions for you using all the right variables, in your own style, it's quite impressive.
Anyone tried using the actual verbal chat on ChatGPT yet?
I've been playing around with it today and it's pretty damn good, its way ahead of anything like Siri or Alexa.
Having actual conversations about anything you might find interesting is mind blowing.
Edit: It's now informed me that I've used up my monthly limit and now need to subscribe. It gives you just enough to realise how useful it is! haha
I've been playing around with it today and it's pretty damn good, its way ahead of anything like Siri or Alexa.
Having actual conversations about anything you might find interesting is mind blowing.
Edit: It's now informed me that I've used up my monthly limit and now need to subscribe. It gives you just enough to realise how useful it is! haha
Last edited:
I might have to try that out. How long does it take to respond to a speaken question?
Last edited:
Nice!
I've used it on and off the last couple of weeks. Pretty impressive stuff to be honest.
The only thing with chatgpt at the moment, it keeps calling me Orbit for some reason. I've told it to stop calling me this but it still keeps calling it me.
Haha, I must be lucky that it remembers my name fine.
Actually considering a subscription to this at least for a month or two to see how it works out.
I find the sub worth it for what I use it for. The time saved is worth it alone.
Following govt sub-committee yesterday seem to be some hype about lack of UK military use/investment of AI versus Ukraine
bbc interview this morning - hyperbola about ukraine ai use for autonomous drone targetting and intelligence analysis without any details - what is the practicality ?
( yes there was all the tories hype with bletchley conference suggesting uk would become an ai powerhouse )
bbc interview this morning - hyperbola about ukraine ai use for autonomous drone targetting and intelligence analysis without any details - what is the practicality ?
( yes there was all the tories hype with bletchley conference suggesting uk would become an ai powerhouse )
I'm currently doing an Artificial Intelligence programme at Oxford Uni. I have to say it's cut the ChatGPT wow factor down with a bump lol.
There's a paper that demonstrates that as LLMs get larger they become more inaccurate. So it's not a simple case of more is better.
The ramification is that we're not going to see a magical shift in employment to AI (Artificial General Intelligence, AGI) in the short or medium term. However we will see it as a more useful tool. To caveat the advances of AI tend to be non linear and "Emergence" (the term for AI advancement) can be stuck for a while, or magically make a massive leap only to take two steps back.
The models of AI are all closed and owned by US companies. This means the UK needs to start, from scratch, to make it's own cloud and AI.
There's two interesting impacts of AI:
1. The AI interactive element canabalises Google's own business case on search, and the companies providing web/mobile interactions. They become SaaS databases of information. This has a knock on effect to web/mobile development which rapidly declines as there's no need to have your own app on the phone when everything is through the AI.
2. As the AI learns it incorporates data into its model and doesn't need that data source in operation. So in the short term the AI will continue using SaaS services during operation, later will only need to use them during training and finally.. the AI will learn from the pervious AI and not need SaaS data at all.
This means a decline of SaaS business cases too with an eventual loss of profitable business cases. The same with databases.
What's also funny is that a training cycle of a large LLM you see like ChatGPT costs upwards of $5M for computation alone. Then you have all the storage etc too, validation etc.. the cycle of training to address any security concerns is also a problem - it leaves holes wide open until the training can patch it (so I'd expect AI to sit in guard rails with lots of patches being applied in regular code until the next training cycle).
Also generative AI only knows the data you have given it. It's making links between data. However what will be fun is Innovation AI which is where the AI uses the geometry of the statistical regression, clustering in multiple dimensions to say "this area over here has no data points but the the maths here points to it being interesting".
There's a paper that demonstrates that as LLMs get larger they become more inaccurate. So it's not a simple case of more is better.
The ramification is that we're not going to see a magical shift in employment to AI (Artificial General Intelligence, AGI) in the short or medium term. However we will see it as a more useful tool. To caveat the advances of AI tend to be non linear and "Emergence" (the term for AI advancement) can be stuck for a while, or magically make a massive leap only to take two steps back.
The models of AI are all closed and owned by US companies. This means the UK needs to start, from scratch, to make it's own cloud and AI.
There's two interesting impacts of AI:
1. The AI interactive element canabalises Google's own business case on search, and the companies providing web/mobile interactions. They become SaaS databases of information. This has a knock on effect to web/mobile development which rapidly declines as there's no need to have your own app on the phone when everything is through the AI.
2. As the AI learns it incorporates data into its model and doesn't need that data source in operation. So in the short term the AI will continue using SaaS services during operation, later will only need to use them during training and finally.. the AI will learn from the pervious AI and not need SaaS data at all.
This means a decline of SaaS business cases too with an eventual loss of profitable business cases. The same with databases.
What's also funny is that a training cycle of a large LLM you see like ChatGPT costs upwards of $5M for computation alone. Then you have all the storage etc too, validation etc.. the cycle of training to address any security concerns is also a problem - it leaves holes wide open until the training can patch it (so I'd expect AI to sit in guard rails with lots of patches being applied in regular code until the next training cycle).
Also generative AI only knows the data you have given it. It's making links between data. However what will be fun is Innovation AI which is where the AI uses the geometry of the statistical regression, clustering in multiple dimensions to say "this area over here has no data points but the the maths here points to it being interesting".
Last edited:
ChatGPT uses 15x more energy than a google search for each prompt.
ChatGPT uses 1,060 MWh per training run.
Google uses 14TWh.
USA total capacity is 4,028 TWh for everything residential, commercial and industrial.
So if everyone used ChatGPT instead of google search, 15x14 = 210 TWh.
16.94 cents per kilowatt-hour
So 210,000,000 * 16.94 = 3,557,400,000 USD per hour (3.6Bn)
ChatGPT uses 1,060 MWh per training run.
Google uses 14TWh.
USA total capacity is 4,028 TWh for everything residential, commercial and industrial.
So if everyone used ChatGPT instead of google search, 15x14 = 210 TWh.
16.94 cents per kilowatt-hour
So 210,000,000 * 16.94 = 3,557,400,000 USD per hour (3.6Bn)
Soldato
- Joined
- 25 Nov 2005
- Posts
- 12,657
I didn't try it because it's associated with Tencent and thus the CCP and the US has just sanctioned them as a military supplier but https://github.com/Tencent/HunyuanVideo looks quite good if you're a little more risk averse and don't mind Winnie the Pooh and his gang spying on you