Facebook Page Hurdles

booleanstringsBoolean Leave a Comment

I have recently started managing my AI image Facebook page, The Prompter. The experience feels like the 20th century. It is surprising, given the vast audience.

If I am logged in as both me and the page, FB goes into an endless loop wanting to redraw.

The page has two management suits, “Dashboard” and “Meta,” with somewhat overlapping functions.

Connecting Instagram to FB took days; it was failing with an alert and assured me the developers would be notified. Right.

I can only post to both FB and Instagram from “Meta.” I cannot drag the images into a box but have to select them.

The default for sharing is FB only, which I have to cancel. The default time and date is two days ahead. It offers to also share to FB groups but only lists a few.

Instagram accepts only specific aspect rations, so many images cannot go there.

Am I missing something? Why is this process so cumbersome?

If you have not seen my AI art created with Midjourney, please take a look at https://www.facebook.com/aibraingain. I have also uploaded the first “how-to” video on Youtube.

 

LinkedIn Engineering Compromises

booleanstringsBoolean Leave a Comment

Insufficient “processing power” is the reason why many improvements I had requested are not viable.

1. Why wouldn’t LinkedIn Recruiter search for a Boolean of school names, while LinkedIn.com can?

[Dan] This is possible, of course, but right now, it isn’t in the plan because it is really (really) expensive to build Boolean functionality outside of keyword searches.  This is true (the expensive part) for any field outside of keywords.  So right now, we allow keyword searches (as you, of course, know).  The cost (and therefore speed, and processing power… not just talking about engineering cost here) is substantial.

2. I cannot see an attached resume in Recruiter:

[Dan] This is a funny one, actually.  That guy Theo actually did a fairly creative use of the “featured” zone.  So we don’t actually “show” that as a resume.  We get resumes from Job Applications, and Post Apply Flow sharing. But that doc is actually in a very poorly adopted section called “Featured”.  We don’t show that in Recruiter, because it has VERY low liquidity.  So we don’t actually recognize that as a resume, despite the fact that it is.

Low liquidity?

3. The “Selection or Boolean” fields are confusing. I would separate them – give the user a choice of either Boolean or selections. Imagine searching for “current or past title” for this; I don’t think the UI should allow it:

[Dan] This idea would work if you 1) assumed our taxonomy was perfect and 2) we supported AND operators with “Selection”.  For 1), as an example, we don’t have “Jira Administrator” in our taxonomy. Now, does that mean we don’t let you search for this? Of course not! It’s a valid title. So we let you do Boolean to cover the first 85% of use cases well, but then allow keyword to let you get to 100% coverage. And because we don’t do 2), we would end up removing Boolean functionality if we made you do only Selection. The “Current Or Past” just tells us where to look. So in reality, I think the experience makes more sense now than what you are suggesting.  If we split them, it would not work as well, and not get as much coverage.

Have you looked at the hidden LinkedIn operators yet? They can do wonders; I wish that would be in the UI. 😊

[Dan] The article actually says this works but I don’t think it does. I tried it out (and had a PM on my team play with it too) for a few searches, and the results don’t match each time.  So I don’t think this would work better, because I don’t think it is actually doing what that article says it is.

Oh yes, it does.

Check out our latest class on LinkedIn Recruiter, incorporating all the feedback I got from LinkedIn Engineering (thanks, Dan!)

 

Three Amazing /describe Midjourney Examples

booleanstringsBoolean Leave a Comment

I am fascinated by Midjourney’s /describe command – kudos to its creators! MJ generates beauty out of very little.

In addition to transforming the input, the revealed prompts teach AI artists the “true” language. It was interesting to see words repeated, hashtags, and styles. And if you like a style you can carry it on with other scenes.

Below you will find three pairs of input and output images. I have noticed that MJ does text recognition as well, so it does in the first example.

 

silhouette of a man with a pipe and hat / private investigator

Please follow my AI art at The Prompter; thanks!

Image-to-Prompt in Midjourney

booleanstringsBoolean Leave a Comment

Midjourney has just released an amazing feature, invoked by the command /describe. You can upload an image and get four sample prompts to draw in Midjourney. This is “image-to-text” that creates only remote likeness to the subject. (I suppose photos of famous people and such may be better replicated). Instead, it creates four fantasies. Of course, you can edit the prompts as you wish.

Here is an example. I uploaded the above plain “Boolean” image, then ran the output through Midjourney. The results are interesting:

This feature adds a new layer of creativity. It also reveals some prompt secrets.

See (and hopefully like) my AI art at https://www.facebook.com/aibraingain/.

Google’s Bard: an Immature AI

booleanstringsBoolean Leave a Comment

I appreciate the “code red” that Google had experienced seeing ChatGPT success. However, Bard, Google’s response to AI, is quite underwhelming. Google should have given it time and effort to mature, I think.

The web is full of hilarious dialogs with Bard like

ChatGPT and Bing Chat seem to know more about Google than (its own) Bart. This is ridiculous!

At the same time, I do not feel like there is a solution of LLMs combined with web search yet. Due to Bing Chat’s hallucinations, I mostly Google, not Bing Chat, for recent news and facts. For text editing, research questions, anything language, my choice is ChatGPT and Grammarly. That said, Bing Chat is improving.

Exiting times!

The image above is made with Midjourney; I generate prompts with ChatGPT. Please see my portfolio at the Facebook page The Prompter.

 

Softly, Softly, Catchee Monkey. About Taming Bing

booleanstringsBoolean Leave a Comment


[ image generated with #midjourney. I do not mean that Bing is like a monkey, it is just an expression. See my portfolio on FB. ]

As we have all experienced, Bing Chat is “stubborn” and “lazy” (so is ChatGPT). After Bing’s volatile and “emotional” first day, it was told to say “no” more often than not.

If Bing responds to you with its infamous “I am an AI model” or any other objection, do not believe it until you have tried. There are things to do:

  1. Start a new chat and repeat the prompt – you might get lucky.
  2. Rephrase the prompt. (Some say saying “please” also helps!)
  3. Slowly talk Bing into doing the task it had rejected by taking small steps.

Here is an example of #3. I got a test list of LinkedIn profile URLs and asked Bing to generate a table based on them. It refused, telling me it has no access to the pages – only to search results and its training data.

I repeated the request with only one URL and got the beginning of the table – which Bing gladly generated. Then I asked to add one more URL, three more, and it worked. The result is scraped profiles for a list of LinkedIn URLs, where, unlike with SalesQL (which is a great contact finder and scraper), you do not need to log in.

I do not know how scalable this is, but perhaps we can “softly” teach Bing to create massive scrapes of profile lists. That allows to sort and filter by values unavailable on LinkedIn.

Here is another application: I have a list of candidates and want to know which ones work at midsize companies. I feed Bing a list of companies, starting with one, and increasing the number of fields by which I want to sort. Bing gets drawn into the task.

Bing Chat and ChatGPT plug-ins are promising aids in scraping public pages and even cross-referencing information on different sites.

As a reminder, registration is open for

Six-Day Sourcing Bootcamp with Irina Shamaeva,

where I will cover AI in recruitment as well all tools, techniques, and strategies relevant to sourcing. April 4-April 13, 2023. Seating is limited.

What March 2023 participants have said:

“Just wow! Very fortunate to learn from you. Thank you!”
“Having followed Irina for several years, it was very refreshing to be overwhelmed by her knowledge and the information she shared. Now I have to complete all the homework.”
“Irina you have an amazing knowledge ! I am impressed by seeing what you do in sourcing. Many thanks for the course!”
“I have been passioned about recruitment , have been working for past 14 years in that filed but your course showed me that I have gaps in sourcing and still much to learn from you and others! Homework is to be done but I need to be still studying your materials!”

GPT-4: Game-Changer for Image-to-Text?

booleanstringsBoolean 3 Comments

GPT-4 tells fewer lies. For example, it correctly lists Google search operators. What I find most interesting though at the first glance is its exceptional ability to interpret images based on URLs. There is no restriction for the creation time. I gave it the image above and it produced a perfect output:

I would do face recognition, too, and tell you the person’s bio – try it on someone’s photo.

OK, it was wishful thinking for now! After a few colleagues’ comments and a few more tests, I realize that it is hallucinating based on the image URL. So what does multi-modal mean?

P.S. Please follow my Midjourney AI Art at my new page The Prompter. I use a ChatGPT-based prompt generator. 🙂

Defeat Continuous Scrolling

booleanstringsBoolean 2 Comments

Google has been rolling out continuous scrolling, and it’s a challenge. Seeing all of your results now requires a lot of scrolling down and pressing the “more results” button.

There is a preference, but it does not influence Google’s behavior (at least for me):

The change is unfortunate. Chrome extensions and other tools will break over the scrolling format.

However, there is a workaround. Add &start=<any number> to the end of your search URL, and the familiar view with numbered pages comes back. Following the link, you will arrive at the second page and then all tools, including Instant Data Scraper, will behave again.

So, to view or collect all results:

  1. Get results from the first page
  2. Use &start=1 to get to the second and following pages.

Note that the “include omitted results” option is now hard to reach:

Therefore, remember to add &filter=0 to your search URL to get more results from the same sites.

Join us at the renewed Boolean Strings Basics & Beyond on March 1-2, 2023. We will cover principles that are staying and new developments that change the best ways to Google, like the one above, as well as touch on Google vs. ChatGPT competition.

As a reminder, I am holding a Six-Day Sourcing Bootcamp with Irina Shamaeva March 7-16. I will cover all things sourcing in this course. It is filling up, so if you are interested, I recommend signing up.

Bye, #ChatGPT DAN!

booleanstringsBoolean Leave a Comment

There is no need for “DAN” Do Anything Now jailbreaking prompts to try and access the web anymore. The New Bing Chat has changed that. 
The new Bing does not look like “the search engine Bing combined with ChatGPT”  – it is “Bing Chat,”
ChatGPT with web access
(in a different tab even). That is excellent news!
Here is an example of what you can do as soon as you get access to the new Bing.  Prompt:
Be cautious, though! ChatGPT always starts with minimal effort. Tell Bing Chat to be thorough. Tell it to digest the pages fully; otherwise, it may begin to hallucinate. There are still potential limitations that we will discover by practicing.
What are your experiences with the new Bing?

ChatGPT Will Not Replace You If You Embrace It

booleanstringsBoolean Leave a Comment

My recent LinkedIn poll, “Recruiters: are you already using #ChatGPT in your work to write emails to prospects and other content?” shows an impressive rate of acceptance and interest in ChatGPT.

Our Facebook group ChatGPT, Artificial Intelligence, Prompts, Insights for Recruitment amassed 5.6K+ members in no time; I have never seen such growth. The LinkedIn group with the same name is exploding as well.

Attitudes vary a lot, though. Answering “What is your interest in ChatGPT?” someone joining the FB group said, “Know thine enemy.” (lol)

Some skepticism arises from unimpressive initial uses.

  • If you tried ChatGPT and it produced a generic, mediocre piece of writing, you may have voted “It is not good enough” in the poll. But rewriting your prompt and giving it additional direction will help!
  • ChatGPT sometimes states the wrong facts. It is not a good reason to stay away from it! But remember to verify its output.

On the other hand, some enthusiasts believe they will now automate any writing. Or they worry about their jobs. Not so fast! If you spend at least a few hours with ChatGPT, you will learn that getting the desirable, well-written piece requires a lot of effort, guesswork, testing, and (yes) Googling. Just as there are no useful Boolean Builders, there is a limit to lists of universal ChatGPT recipes anyone can create. Using ChatGPT is more of “know-how.”

We have pushed ChatGPT’s tires a bit, and I will be sharing what I have learned in a webinar. The proceeds will go to a good cause.

Pease join me at a 90-minute webinar Revolutionizing Recruitment with ChatGPT and AI (Ukraine Benefit) on Wednesday, February 15th! Some of the topics I will cover are:

  • Introduction to ChatGPT and its capabilities
  • ChatGPT vs. Google
  • How ChatGPT can be used to streamline and enhance various aspects of the recruitment process
    • Research the market
    • Create Boolean Strings
    • Manipulate text data
    • Craft targeted LinkedIn messages and emails to passive candidates
    • Improve your writing
    • Generate interview questions and assessment tests
  • Brainstorm with ChatGPT
  • ChatGPT for everyday use
  • Limitations and how to overcome them
  • Competing landscape
  • Image AI
  • Q&A session