Just the beginning, lots more to come 😎
1 Like
Here’s how to access ‘AI’ in RW/Stacks 😜:
In saying that…it would be nice if we could assign shortcuts to add particular stacks (e.g. our top 5 most used/favourites).
5 Likes
here is a quick demo of accessing some AI assistance in RW/Stacks (using Raycast and a chatgpt extension):
10 Likes
I finally found some time today to get a play with this AI content creation nonsense. I asked Bard to create blog posts on various topics about which I’m fairly knowledgeable. There is no doubt the output was accurate in most cases and had a good reading flow to it. But in all cases it was so dull. None of the articles created generated any sort of enthusiasm for the subject. And while mostly factually correct they were not in the slightest but engaging.
I then spent a few hours reading AI generated content that other people were raving about, saying it was going to change everything, and it was the same. Accurate, but dull and uninspiring.
If what I read today is typical of the standard of AI produced content, copywriters can sleep easy. They ain’t gonna be out of a job anytime soon. Not the good ones anyway.
I personally would never populate a website with anything AI created for me today. Not if I wanted to actually engage and retain the visitor.
In general, I’d agree (as would Chat GPT). At the moment LLMs are much better used to generate a ‘copy platform’, pulling together all the elements that need to be included, for a writer to rework. But there are a few observations I’d throw in here from my experience of Chat GPT already.
It’s really, really good at writing in a positive but impartial and unthreatening way. Much better than most human beings, in fact. It has a lot to teach us about tone of voice.
It doesn’t waffle. Human writers often repeat things in different words. Chat GPT packs a lot of information into relatively few words, and moves forward in a coherent way. It has a lot to teach about concision too.
Its summaries are also worth studying. When it breaks things into numbered points, there is more overlap between them than we’re used to, and the overall effect of this is to give the sense of different facets of a single whole, rather than different points taken off in different directions.
Never ask LLMs to write in the style of somebody, except for amusement. I witnessed an object lesson of this this morning: a facebook friend who is a pensions industry expert and has a successful blog asked Chat GPT to write a post in his style. It rambled on, just as he does.
The thing about LLMs is that if you ask them really interesting questions, you can get really interesting responses. The kinds of things that you would never have come up with. These make excellent discussion points, or starting points for interesting pieces.
1 Like
Interesting points.
However, what’s AI doesn’t do is potentially the thing(s) that make articles more engaging to humans. A la your friends rambling articles.
I have to admit, I am warming to this.
Already today two of my clients have asked me to produce monthly blogs using AI and add them to their site. For a nominal fee ;-)
I’ve told them exactly how they are produced, said i will do a little bit of editing to personalise it, and create a nice graphic, but otherwise they’ll be getting the generic content the AI engine creates. They both said something bland and generic is better than nothing.
So now that it’s an earner, I’m all for it.
:-)
4 Likes
Seems I have to add some AI driven blog post creation in Volt CMS 2.
🤓
3 Likes
I wonder if this will kill blogs long-term? Why do I need to go to a blog when I can ask AI myself? I imagine those using this for blogs need to know if the AI is plagiarising. Also, who has the copyright for the content created? More than one person can ask an AI the same question and get the same content.
Another issue is with what AI produces. I imagine it’s a dream come true for those pushing harmful and purposely false information.
What about the question to the quality of AI’s source for producing its content? If it’s using the internet, that’s probably the worst place to feed an AI. Very little public accessible information on the web has any form of real evidence to substantiate it.
1 Like
Just an aside: Does the AI produce a list of footnotes and links to the content where it gets its information?
Yes, of course ChatGPT would. However, keep in mind that unless you are using v.4 with plugins (that allow it to access the Internet), it is not capable of conducting independent research or accessing external sources beyond its knowledge cutoff date in September 2021. Therefore, any citations it provides would be based on general knowledge up to that date.
And beware of the B.S.
ChatGPT might present something like it’s been in Encyclopedia Britannica since the first edition of 1768, while it is in fact a total fabrication. For example I asked ChatGPT (using GPT-4 mind you) how to replicate an ETF not available for private investors in the EU. Basically it’s a kind of question like: “What ingredients do I need to make my own Oreo’s?” ChatGPT came up with non-existing ETFs, with ticker symbols, ISIN numbers and all (what made it easy to verify btw). Caveat scriptor! Or: cave scriptor? 😉
1 Like
I have a wee house in Spain that I rent out as a holiday let when I’m not there, via a site I created last year. The content on the site is all post July 2022, it’s all written by me and it’s all only on that site, nowhere else. Indeed, there is almost no reference to the house anywhere online other than that site.
I asked Bard to write a review of the house. It more or less lifted my content, pretty much word for word, but rearranged it into different chunks. It also added in lots of additional info (all incorrect), I suspect lifted from other sites about similar houses. For instance, my house is two bedroom with one bath, but the review said two bedroom, two bath, one en-suite! It also said there was a swimming pool. Which there isn’t. Sadly.
I’m not sure how all this works about copyright.
I doubt it will kill blogs, but it will drive better writing (just as smartphones’ video-making tools have pushed higher standards of editing, as content creators seek to stand out). What LLM’s do is put more emphasis on what the author asks the system, and the quality of the conversations they have with them. And as LLMs acquire more quality data, it’s possible to ask questions that would previously have involved weeks of trying to identify research. For instance, we can ask a question like: “how much of the US’ wealth is in property, and how much is equities?” and if the data is there, somewhere, available to the LLM, it will come back to us with a summary. In about ten seconds. This kind of thing is incredibly frustrating and time-consuming to research, even with Google — you have to comb through dozens of websites, and you discover that what they have is not in the form you need it, or out of date, or in percentages rather than amounts, etc. LLMs now make it possible for the individual blog writer to come up with stories that are as insightful as, for instance, those of a Pew or a YouGov with huge research capabilities (and, as an aside, there’s big opportunity here for those who can make good infographic and table tools for blog writers).
If you can ask interesting enough questions, you could easily populate a blog just with Chat GPT/Bard answers. And it’s not that anyone else couldn’t get those answers, but by no means everyone can ask the right questions. But those answers would be much better re-written, with some personal commentary too. That’s where human beings will always add the value: in curating and explaining content.
2 Likes
I have to agree. I’ve 100% turned around on this. Yesterday, asking simple questions I got pretty piss poor blogs, which formed my initial post above. But since then I’ve better learnt how to use it, and what I’m getting back is getting better and better. And I’m less than 24hrs in. The blog articles created need editing and personalising, but that’s easy.
3 Likes
Also bear in mind that Italy “banned” ChatGPT a few days ago. More countries will very likely follow.
1 Like
I do think it needs more thinking through before it’s let out in the wild.
I’ve recently found myself utilizing the service quite frequently, so I’ve made the decision to fully commit by subscribing to the Plus account at $20/month. While I initially used it for web design-related tasks, I’ve discovered that it’s incredibly versatile and useful for a wide range of applications. In fact, I now use it extensively for marketing support in my full-time 9-5 job. Additionally, it has become a valuable tool for composing and revising my emails, as well as crafting responses to forum posts.
2 Likes
No, but I have seen it use Bible versus.