If you’re a parent, or have never grown up, I expect you’ll likely have been cursed like me with the regular earworms of Disney songs.
Take bibbidy bobbidi boo in which the Fairy Godmother automates the manufacture of a carriage for the ball through magicking.
You’re welcome.
Ah, sorry that’s another Disney tune.
Anyway, awkward earworms into segue, you may have heard ChatGPT-4 has been revealed.
Just over 3 months after ChatGPT-3 rocked the world with generic content, lies and gaslighting.
It will be interesting to what kind of automated magic we can come up with, and I will be trying the same ideas I did in December.
A key change is the addition of image input to text output.
There are some brilliant applications that immediately spring to mind such as automated alt-text for images, useful in both social media posts and presentations for better accessibility.
I hear from my clever tech wizarding friends that subsequent versions will feature video and audio input.
Aside from the general improvements that likely correlate with Moore’s Law, OpenAi claims GPT-4 will be more creative and less likely to tell porky pies.
However, it will still work in the same way, by parsing data from established sources and reforming in a way that matches its estimation of your intent.
Funnily enough, it’s already in the wild, until now under a cloak of invisibility: BingChat and Stripe use it.
You may have heard BingChat is certifiable, so let’s not hold out too much hope for GPT-4 being true and sane.
I expect 4 will lack specific situational insight in the same way as GPT-3.
For example - with vacancy advertising, the output may look and feel like an advert, but will lack the context that makes a vacancy unique, without the unique voice that speaks to the unique aspirations of the reader.
However, prompt engineering is a burgeoning skill.
It’s entirely possible to input the criteria that spell out content for these unique qualities.
I came quite close with a Streetfighter 2 homage that featured Blanka welcoming new career challengers and electrifying them with a levelled-up career.
My effort was merely an illusion of creativity with little substance.
Even if you have the skill to successfully prompt engineer an advert, it’s more enjoyable to write an advert whose flaws will be more human and engaging.
In the 3 months since my last article on the ChatGPT in recruitment, my understanding of its application has developed.
Initially, we all looked at it as a content generation tool, which indeed is its point as a Large Language Model.
However, I now see it more as an ideation, confirmation and error-checking tool, with the caveat of its unreliability.
Here are some suggestions you can try, which may conjure further thoughts on effective use:
Vacancy checking
1/ Go on to ChatGPT
2/ Type "What should this job title be:" and paste the responsibilities section from your problematic vacancy.
If the job title produced is not close to the job title expected, you may have a problem with
3/ the wrong job title
4/ unclear job responsibilities
I've tried this with 30 UK job descriptions, including few a ‘original vs updated by me’ comparisons, and the results were consistent.
It tracks with one of my first steps in partnering on a 'problem' vacancy.
Is it acceptable for your vacancy to be held back by misrepresentative documentation?
Candidate requirement checking
Try the same for “what jobs would this person be suitable for:”
Responsibility and candidate requirement generation
Do the opposite for the job title you are recruiting for, and compare the results with your own requirements.
“What skills, responsibilities and experience should the vacancy of a Prompt Engineer have?”
New role design
Map out the problems your new vacancy solves, with as specific data as possible. Do the above steps with this in mind.
Interviews
“Give me a list of competency questions based on these responsibilities and candidate requirements”
Risky to do if your job description and definition of ‘good’ in a candidate are flawed – imagine failure cascading through your process from this bad start.
If the output isn’t satisfactory, iterate. “Try again” or “Try again with <additional criteria>”.
While none of these suggestions is error-proof, they are great for idea generation as well as a starting point when you don’t know where to start.
Think about the questions you ask of your own recruitment process, and try asking ChatGPT instead.
Considering ChatGPT works by parsing existing content that 'is out there' and reforming it as a best guess for your intent, it's a good mirror for your own work.
You can try them out now on ChatGPT-3, and I imagine the results will be better with 4, which you can already access if you subscribe to ChatGPT Plus.
This newsletter will be outdated soon. ChatGPT-4 has been live behind closed doors for a year. What sorcery do you think They are working on now?
I typically take a stock take on where we are with AI once a year, given how quickly things change. I can’t wait for this time next year.
As Arthur C Clarke once said, “any sufficiently advanced technology is indistinguishable from magic.”
Thanks for reading.
Regards,
Greg
p.s. While you are here, if you like the idea of improving how you recruit, lack capacity or need better candidates, and are curious how I can help, these are my services:
- commercial, operational and technical leadership recruitment (available for no more than four vacancies)
- manage part or all of your recruitment on an individually designed basis for one client (no availability until June)
- recruitment coaching and mentoring (no capacity until May)
- recruitment strategy setting
- outplacement support
Just hit reply to check if my approach is right for you.