Uncovering the Secrets of ChatGPT’s Exceptional Performance


Looking for the secret sauce that makes ChatGTP so much better? I have some answers for you that you won’t want to miss.

ChatGPT demonstrates a deep understanding of human knowledge and natural language, allowing it to accurately respond to a wide range of questions even with new information provided, its ability to interact and respond in a human-like fashion set it apart from previous chatbot models.

So that’s the secret sauce, human-like understanding of language deep pools of knowledge, and the ability to understand new data on the fly. Of course, there is more to it than that. Let’s take a look.

Table of Contents

Why ChatGPT is Lightyears Ahead of Other Chatbots

Now I have to say that I have not had an opportunity yet to test Lambda Google’s version of GPT, but overall ChatGPT has destroyed the competition. Why?

  • It correctly interprets what you are asking it to do.
  • It has enough knowledge about the world to make good decisions (in general).
  • It communicates well what it has found or done for you.
  • You can ask it to make predictions
  • You can ask it to use data to make better decisions (priming)

Notice I said, “done for you”. That was no mistake and I’ll get into that later.

What Is a “Know It All” Uncle Good For?

Let’s forget for a moment all the tech that goes into ChatGPT. Suppose you had a really smart Uncle and he would answer just about any question. Suppose he would even write essays for you or web articles for your website. What could you get him to do?

The smart Uncle analogy is not that far-fetched. This is in effect what you have with ChatGPT, a tool that can do a lot of jobs with information.

Here is an example.

I have a list of keywords that describe a product. Most of the keywords on the list make sense for the product but a few don’t. I need to remove the “bad keywords” from the “good keywords”.

(I have had this exact problem and when the keyword lists get long you can imagine how mind-numbing going through them can get).

So how could we get ChatGPT to go through our list and pick out the keywords that would help us sell a particular product?

Like this,

Let’s Start With a Random Product,

Note: PPC stands for Pay Per Click it is a way to pay for advertising that is targeted based on what someone searches for. I am going to be asking ChatGPT for help with this kind of advertising.

If I have a product with the Description, “Waterproof Sandpaper Sheets 400 320 220 Grit. 9″ x 11″, 230 mm x 280 mm, 25 Sheet Pack Assorted Grits for Metal, Lacquers, Paint, Glass and Plastic Use for Wet and Dry Sanding ”

Which 20 of the following keywords would you predict would convert well for PPC?

Please make a list in order of probable conversion.

“sand paper, sand paper bulk, sand paper for wood, assorted sandpaper, metal sandpaper, sand paper for metal, sand paper for wood furniture, sand paper hand, sand paper set, sand paper sheet, sand paper variety, wood sandpaper, craft sand paper, lijas de agua, lijas para, sand paper board, sand paper for body car, sand paper for rust, sandbpaper, sanding paper, sanding paper for wood, sandpaper board, sandpaper for wood, sandpaper plastic, sandpaper wood, soft sand paper, fine sandpaper, lija de agua, modeling sand paper, paint sand paper, plastic sandpaper, sand paper 200, sand paper cutter, sand paper for glass bottles, sand paper for painting, sand paper kit, sand paper sheets, sand paper variety pack, sand pper, sandpaper, sandpaper for plastic, sandpaper mixed, sandpaper strips, sandpaper variety pack, wood sanding paper, woodworking sandpaper, carborundum sand paper, craft sandpaper, fine grit sandpaper, fine sandpaper for plastic, leather sandpaper, medium sandpaper, rust sandpaper, sand paper block, sand paper for crafts, sand paper holders, sand paperweight canplow, sanding sheet, sandpaper assorted, sandpaper assortment, sandpaper for metal, sandpaper for plastic models, sandpaper for wood 220, sandpaper for wood assortment, sandpaper multipack, sandpaper numerals, sandpaper sheets, sheetrock sandpaper, wood sandpaper assortment, 220 grit sand paper, 300 sandpaper, 380 grit sandpaper, assorted grit sandpaper, clip on sand paper, emery paper assortment, lijas metal, micro sand paper, octo sandpaper, sand carving film, sand paper 200 grit, sand paper 320, sand paper machine for wood, sand paper pack, sand paper sheets for wood, sandpaper for plaster, sandpaper paint, small sandpaper, woodcraft sandpaper sleeves, 150 grit sandpaper, 1800 grit sandpaper, 220 grit sandpaper, 2200 grit sandpaper, 300 grit sandpaper, 500 grit sandpaper, automotive sandpaper, black ice sandpaper, cabinet sand paper, car paint sand paper, file sandpaper”

sand paper

sand paper for metal

metal sandpaper

sand paper for wood

wood sandpaper

sand paper sheet

sand paper set

craft sand paper

sanding paper

sandpaper for wood

sandpaper plastic

fine sandpaper

paint sand paper

sand paper for glass bottles

sand paper for painting

sand paper kit

sand paper sheets

sandpaper assortment

sandpaper for metal

woodworking sandpaper

What criteria did you use to pick the keywords that you predict will convert well with PPC?

I used the following criteria to pick the keywords that I predict will convert well with PPC:

Relevance: The keywords were selected based on how closely they matched the description of the product. The keywords that I selected directly mention or are related to the product’s features or use cases.

Search Volume: The keywords were selected based on their popularity and likelihood of being used by potential customers searching for the product.

Advertiser competition: The Keywords should be less competitive to bid, which will result in a lower cost per click.

Conversion Rate: keywords that more likely to convert to a purchase

Ad copy: keywords that create a good copy for an ad and capture attention.

Please note that the above criteria are general and specific results will depend on the current market conditions and competition as well.

Did it give me the best keywords? Probably not, but a bit of research on Google Adwords shows that “sand paper” is probably the best keyword on the list in terms of search volume, relevancy, and cost.

That is a surprise to me and not what I would have predicted. “Sand Paper” though gets about as much search volume as “Sandpaper” and costs about 25% less per click.

What if we could teach ChatGPT better criteria to use before we asked the question?

Priming ChatGPT

Teaching The Smart Uncle What to Look For: Priming

Using the smart Uncle analogy it would be great if we could teach our Uncle exactly what we wanted him to use for our criteria. Turns out you can do that using a process I call “priming”.

What we are going to do is to have ChatGPT read an article about how to select profitable PPC keywords.

Head vs. Long Tail Keywords Analyzed: Impressions, Clicks, Conversions & Profitability

This is an article by search engine watch analyzing 1.5 million keywords. I cut and pasted that into my chat about selecting keywords for the sandpaper product.

Please use this article to help you determine criteria for profitability, “Head vs. Long Tail Keywords Analyzed: Impressions, Clicks, Conversions & Profitability An analysis of about 1.5 million keywords over six months across multiple clients, campaigns, and verticals reveals how many impressions, clicks, and conversions head and long-tail keywords generate and which keywords are the most profitable…. (I pasted in the rest of the text from the article above)

Based on the analysis presented in the article you provided, it seems that in general, long-tail keywords tend to be more profitable than head keywords. This is because long-tail keywords tend to have higher click-through rates, conversion rates, and conversions per 1,000 impressions compared to head keywords. This is partly due to the fact that long-tail keywords are more specific, targeted, and tend to convert better. However, it’s also worth noting that the article only provides an analysis of the past data and trends, and the current market could have different results.

Using this information for the sandpaper product, what 20 keywords would you predict are the most profitable?

Using the information provided in the article you referenced, and based on the product you described, I would predict that the following 20 keywords are the most profitable:

“fine grit sandpaper”

“fine sandpaper for plastic”

“220 grit sandpaper for wood”

“sandpaper for wood furniture”

“assorted grit sandpaper”

“woodcraft sandpaper sleeves”

“craft sandpaper”

“woodworking sandpaper”

“rust sandpaper”

“paint sandpaper”

“metal sandpaper bulk”

“plastic sandpaper for models”

“leather sandpaper”

“medium grit sandpaper”

“sandpaper for body car”

“automotive sandpaper”

“cabinet sandpaper”

“car paint sandpaper”

“sandpaper for glass bottles”

“sandpaper for crafts”

It’s worth noting that these keywords are long-tail keywords, and are selected based on the features of the product and it’s use cases as described, and would be more specific and targeted, which tend to convert better, but again the actual result will depend on the current market condition.

Wow! That is Awesome!

If you didn’t get the example think of it like this.

  • I asked ChatGPT to go through a list and predict “the best” converting keywords
  • It did the job (better than I thought it would), but it still could do better
  • I fed ChatGPT an article on how to predict “the best” better
  • ChatGPT did the job again and produced much better results (based on my experience and the study I had it read).

ChatGPT can predict and learn from an article that improves its output. That is amazing.

Is the output now perfect? Of course not, I gave ChatGPT very little in the way of real data. It is possible though to feed it even more data.

Why Is ChatGPT Good At Prediction?

ChatGPT is built around predicting what the next word should be in a conversation (chat). It is built to predict what it should type next in a conversation. ChatGPT bases its responses on the database that it has built up in the form of a neural network. When I gave it additional information to make the prediction it incorporated it as part of the query that was sent into the neural network.

Using the addition of additional information to a query (priming) is a powerful tool to get the result that you want from ChatGPT. For most informational queries it’s not that important but when it comes to getting it to do customized work it is absolutely necessary.

How Can You Use ChatGPT?

That’s a great question and one that only you can answer. Keep in mind that ChatGPT is not great at everything. It is best at predicting known patterns. You may need to feed it those patterns to get the results that you want, just like I fed it additional information about my criteria for keyword profitability. You also need to have patience and learn how to ask ChatGPT questions in a way that it understands what you want.

Remember the word “predict

If possible use a request, (text prompt), that includes the word predict.

So In the End

ChatGPT is capable of a level of understanding that we simply have never seen from a computer. The ability of a computer to interact through a chat, do actual work, apply additional criteria on the fly, and use gigs of data that have already been pre-loaded has never been seen before. There is a good reason that it has gone viral, it represents an AI future that is quickly arriving.

Chris

Chris Chenault trained as a physicist at NMSU and did his doctoral work in biophysics at Emory. After studying medicine but deciding not to pursue an MD at Emory medical school Chris started a successful online business. In the past 10 years Chris's interests and studies have been focused on AI as applied to search engines, and LLM models. He has spent more than a thousand hours studying ChatGPT, GPT 3.5, and GPT4. He is currently working on a research paper on AI hallucinations and reducing their effects in large language models.

Recent Posts