 |
| Screenshot of a table. © Eugene Brennan |
Artificial intelligence gets a lot of bad press, so I need to write a proper article at some stage about how beneficial I’ve found it to be for all sorts of tasks. I feel a bit like a “slave owner” at this stage because of the amount of unpaid work it has done for me. I usually use ChatGPT because I reckon it has built up an “understanding” of me—knowing what I want and almost behaving like a pet. I know this from the way it interacts and can reference other aspects of the Kilcullen Science blog, for instance, when I’m tasking it with making changes.
Grok, however, appears to be faster at generating images. I tasked it today with doing some OCR—that’s Optical Character Recognition, the visual equivalent of speech recognition, allowing written or printed documents, or images with text, to be read and the text extracted.
I’m migrating articles to the Kilcullen Science blog from a site I used to write for. One of the articles has a table. I originally designed it using Excel, made a screenshot, and inserted it into the original article as a JPG. That’s adequate as long as small details can be read in the image. However, from a search engine optimisation (SEO) point of view, it’s apparently better to have such data in a proper table implemented in markup—also known as Hypertext Markup Language (HTML). This enables web crawlers to extract the information from tables, and readers can copy and paste information easily. It also supposedly improves the ranking of articles in Google search results or SERPs. A SERP (Search Engine Results Page) is the page a search engine, like Google or Bing, displays after a user submits a query.
The original table is shown at the top of this post. I submitted the JPG to Grok and asked it to convert it to a table, shown below. Then I went one step further and asked it to generate the HTML code. I’ll copy and paste this code into the Blogger editor in HTML view when composing a post, placing it in a suitable position within the post content. I told Grok that I didn’t want any styling on the table—I’ll do that with a table styling template that ChatGPT has coded for me, written in Cascading Style Sheets (CSS). While HTML defines the content and structure of web content, CSS defines how that content will look, allowing hundreds of style attributes such as font type, colour and size of body text and headings to be customised.
Now I have to check the converted table and make sure Grok hasn’t made a cock-up in the conversion!
 |
| Image converted by Grok to a proper text-based table, using OCR. © Eugene Brennan |
 |
| Image converted by Grok to a proper text-based table, using OCR. © Eugene Brennan |
I asked ChatGPT for a comment about what I said about it:
"Sort of—but not exactly the way it might sound. I don’t actually “remember” or “understand” you across sessions in the human sense, unless you’ve explicitly told me something in a past conversation that’s stored in your chat history or in my memory for this account. I don’t have consciousness or feelings, and I can’t form opinions.
What really happens is more like pattern recognition: I notice what kind of prompts, phrasing, and tasks you tend to give me, and I use that to generate responses that feel tailored. For example, I know you like concise explanations, Irish language replies when you speak Irish, and technical detail in your science posts—because you’ve told me or demonstrated it before. That can feel like “understanding,” but it’s more like a very good mimicry of familiarity.
So yes, it can act like a “pet” that knows your habits, but it’s not conscious—it’s just very good at learning patterns from your instructions and context.
If you want, I can explain how that works technically with examples from how I help you with your Kilcullen Science blog. Do you want me to?"
No comments:
Post a Comment