What GPT ideas could we create as external tools for Elite?
- Thread starter MUTILADOR
- Start date
Dali is out of a job at least.
Copilot, free, 3 second wait... update yourself, you're left in the past.
Attachments
The mass production of bad art is certainly one area where "AI" has innovated. Look too closely at any of these and you'll see questionable details.
Like the first guy firing a laser without having his finger on the trigger or indeed looking where he's firing. With supporting fire from his backpack?
The ship in the second one reminds me of one of those MC Escher perspective tricks. I think the random barrels attached to it make it look questionable, and it seems confused about where and what the lighting is.
Third one really doubled down on the laser beams, I could assume those in the other pictures were coming from somewhere else but clearly it just chucks several in for good measure, considering how they all line up.
Fourth one has a nifty double laser beam, alongside the most...complicated ship design. He is probably not going to make it though, judging by how he appears to be stood on nothing. And there's a small star very close to him in front of that planet in the top right.
But the hands are fine, I guess. Besides the second one, where I think it added in some gun parts and decided they had to look like finger extensions due to being near the hand. Lighting is too bad to be sure.
It probably won't worry him as long as he continues to insist upon remaining dead.
why the hell would you need gpt to generate and open a url that's just https://inara.cz/elite/search/?search=whatever+you+say
From the samples I've seen people post the predictive text engines are pretty good at coming up with extremely bad BGS advice much faster and often more grammatically than a human could manage, including the invention of entirely new false statements not found anywhere else online.
but you can also get it to roleplay as your grandpa telling you about his job in the pipebomb factory which is pretty unique
And yet, it's better than 90% of concept art created by humans, and in only 3 seconds. And if this can be done right now, imagine what it will do in 3 more months. In one year we went from seeing a deformed Will Smith eating spaghetti to SORA. Truly incredible things are coming next 2025.
hahaha It's not to open a page.
It is so that the AI searches in Inara for what I ask with voice command and responds to me with voice the answer, without having to open another window. For example, in combat.
Sure, you can do that. And if you have a use case for it, good for you. But In the last 10 years, I haven't come up with anything that I'd need to look up on any external site (Inara didn't always exist) while engaged in combat.
Indeed. Like the end of the AI fad. Now it may seem incredible to you, but just you wait.
Unless you're using "AI" so generically as to mean "a computer program" rather than "a computer program with a giant marketing budget", I'm not sure what value the AI portion is adding here.
You've already got to set up as prerequisites
- Voice Attack configuration to convert your spoken request into a URL
- Web browser to retrieve the contents of that URL
- Speech synthesis (within the browser or a separate program) to read out the results.
to get any of this to work at all.
At that point, you're going to get far better quality results from the URL you send requests to being one that's actually designed to give useful results to specific types of ED query (and possibly optimised for voice output), than you are from inserting a predictive text engine into the process as well.
That's all technology we've had available for years.
I wouldn't use GPTs for creating content, even though (and because) it is excellent at producing the kind of oleaginous marketing text that we were all sick of before it could be automated.
In terms of what I would use it for, it starts with a caveat: LLMs with chat interfaces need their hands held at all times if they are to give worthwhile output. They cannot, in the example that @Ian Doncaster gave, separate what is known to be true about the BGS (after principled testing) from a half correct (if you squint) statement some random player wrote on a wiki somewhere. OpenAI have neither the ability nor motivation to fact check. Essentially, everything output by ChatGPT is a hallucination - it has no knowledge of itself and no (explicit) knowledge of the outside world. It just so happens that some of the things it outputs are interpreted as being true because there is a preponderance of language to that effect in its training data, so the output corresponds, more or less, with the reality.
With that said, I can certainly see uses for it. It is very good at paraphrasing and summarising information that it has been told to use. If GPT4o could integrated into, say, Inara, it could summarise the faction influence levels and states through speech and output the data in common formats for use elsewhere. However, Inara is a complex website with multiple integrations itself, so this would be, I suspect, a pretty big project. I haven't used Copilot very much, so I'm not sure if that could provide some (or all) of that kind of functionality. It is integrated into Bing search.
Something much more manageable would be to integrate GPT4o into Google Docs and tell it to only use the player-managed data in specific documents. Provided it can be prompted appropriately, given enough context (so it 'knows' how it should interpret data) and has accurate data to work with, it could give actions advice to players in a spoken or written format. Very often, in my experience, groups end up with one or two people who know the BGS well, and those people end up needing very broad shoulders in order to organise and manage BGS actions. A bot could be very helpful in that situation.
There are also Discord integrations for ChatGPT, which could prove useful as well.
In terms of what I would use it for, it starts with a caveat: LLMs with chat interfaces need their hands held at all times if they are to give worthwhile output. They cannot, in the example that @Ian Doncaster gave, separate what is known to be true about the BGS (after principled testing) from a half correct (if you squint) statement some random player wrote on a wiki somewhere. OpenAI have neither the ability nor motivation to fact check. Essentially, everything output by ChatGPT is a hallucination - it has no knowledge of itself and no (explicit) knowledge of the outside world. It just so happens that some of the things it outputs are interpreted as being true because there is a preponderance of language to that effect in its training data, so the output corresponds, more or less, with the reality.
With that said, I can certainly see uses for it. It is very good at paraphrasing and summarising information that it has been told to use. If GPT4o could integrated into, say, Inara, it could summarise the faction influence levels and states through speech and output the data in common formats for use elsewhere. However, Inara is a complex website with multiple integrations itself, so this would be, I suspect, a pretty big project. I haven't used Copilot very much, so I'm not sure if that could provide some (or all) of that kind of functionality. It is integrated into Bing search.
Something much more manageable would be to integrate GPT4o into Google Docs and tell it to only use the player-managed data in specific documents. Provided it can be prompted appropriately, given enough context (so it 'knows' how it should interpret data) and has accurate data to work with, it could give actions advice to players in a spoken or written format. Very often, in my experience, groups end up with one or two people who know the BGS well, and those people end up needing very broad shoulders in order to organise and manage BGS actions. A bot could be very helpful in that situation.
There are also Discord integrations for ChatGPT, which could prove useful as well.
Is chatgpt and it’s ilk behind some of the translations we see in the forums now where the sentence structure is much better but the word choice is less technical so the whole post is much less understandable in the context of what these forums are about.
If so I would say claims for next year are premature maybe five years.
If so I would say claims for next year are premature maybe five years.
The Gartner Hype Cycle.
Companies in this space have followed the classic Silicon Valley marketing playbook. I can't count how many times I've been annoyed by someone in my field (education) claiming that it is a "gamechanger" and that "the university is dead". Almost without exception, they have an AI product to hawk.
Most interestingly for me, I read somewhere (I want to say Ars Technica, but I can't find the article now) that the engineers behind ChatGPT were initially surprised at how competent it was at translation.
