I've been using chatGPT for a bit now. I consider it a thesaurus on performance enhancing drugs. Usually, it's helpful. Other times it is not. I've also begun to use Gemini. It's a different user experience. I find it better in some ways, but have also found it frustrating. For instance, I was writing about an experience I had in my life, and looking to improve my word choice, syntax, etc... The words "Daughters of the Confederacy" were on a tombstone I was writing about. Gemini wouldn't touch my inquiry, and somewhat lectured me on inappropriate subject matter. That was a bit frustrating. I understood why it was being careful and even appreciated it, but it was clear there are a lot of things to work out in the world of AI.
All in all, it just takes some getting used to the system, and knowing how to generate the responses I am looking for.
Thanks for sharing, Jeff. You are right, it needs to be learned. Regarding some prompts not being touched, this is in the later revisions that stricter guardrails were added due to teen prompts leading to crazy outputs. chatGPT is best, in my view, than the other models I have seen, because it has 100M users to learn from till now. Also, for your specific query, I would first instruct the model to know that it is an essay, or research topic, or story etc. to set context then make the query. It usually works for it to ignore that it is real world related.
Google results firstly give sponsored (paid) links, which are often not at all what you want, or take you to a competitor site, e.g. if you look up a business. This has only gotten worse with Google's introduction of AI capabilities. Next, Google shows you what OTHER people found useful, the algorithm is basically a popularity contest. This is what people still seem to not understand about Google, even though it's been around for decades. So if a million people read a nonsense article about measles, that will be higher up. Google was never about accuracy, it was always about most viewed. Google did modify their algorithms a long time ago to address the problem of rubbish results (because the population likes reading rubbish), to try to boost legitimate and relevant results, e g. Now Mayo clinic results are higher up, rather than buried on page three.
Even with AI added, Google is still fundamentally a search engine, and will still show you what other people found useful, rather than what YOU would find useful.
ChatGPT is a tool for doing things, although it's also a good and far more efficient search engine, certainly better than Google. You can get a recipe from either tool, for example, but only ChatGPT will give you a meal plan for the next week, with the associated shopping list.
I like using ChapGPT to do a quick and dirty search, but I've had several experiences where ChaatGPT got the information wrong. I was able to double check the information given to me through another source and verity the inaccuracy. So, I'd just add a warning to use ChapGPT with some caution.
It is important to note that Google is improving its search capabilities with innovations like the "circle to search" feature, which allows users to use a photo for direct web searches. Unlike Google, ChatGPT, standalone app, cannot open URLs directly or perform searches using images yet. Both companies are in strong competition for leadership in search algorithms and enhanced search tools.
Thanks for the additional detail, Giorgio. genAI based image seach is a nascent area for public use, but is not a stretch. Yes, the competition for Google is helpful to raise the bar. Btw, here is my short article on Bard.
I have colleagues at school using ChatGPT to write lesson plans but with so little innovation that what they produce is really poor quality and far worse than anything they’d produce if they put the effort in. I also see it in copywriting - it takes so many great prompts AND a lot of editing afterwards to make these things worth using.
But they are just a snowflake on the tip of the iceberg. Once learned, these AI tools are insane. I’m just scratching the surface and loving every second of it.
I agree, Jack. Prompts do matter, and setting a structure to the response matters too. Also, with the paid versions you can upload entire PDFs for review and analysis. The software scene in 3 years will explode and transform. So, being a step ahead in learning is great!
Excellent review Jayshree. None of these tools are universally applicable. You need to decide what you want before using one of them. I am reminded of the saying that if your only tool is a hammer, everything looks like a nail.
At your suggestion I downloaded ChatGPT, and my husband and have been having fun with it this afternoon 🤩🤩🤩
Thank you for the valuable information about both ChatGPT and Google, I now understand the difference from a layman’s point of view! There is a wealth of information here!
BTW - ChatGPT continues to update the learning base, so the cut off isn't 2021 anymore. For paid only, there's an extension that can be used to ensure the results draw on contemporary information, rather than confined to the current cut off date for the model's learning.
Yes, the free version remains behind, which was a dilemma that immediately came to mind when it was first released, but even the free version doesn't remain forever cut off at 2021. They don't seem to have grappled with that limitation, for both free and paid, it's more material than a mere bug. But, teaching a LLM is big and complex. So, we'll have to see if they can ever get over the hurdle without simply extending to a routine web search for contemporary content.
I've been using chatGPT for a bit now. I consider it a thesaurus on performance enhancing drugs. Usually, it's helpful. Other times it is not. I've also begun to use Gemini. It's a different user experience. I find it better in some ways, but have also found it frustrating. For instance, I was writing about an experience I had in my life, and looking to improve my word choice, syntax, etc... The words "Daughters of the Confederacy" were on a tombstone I was writing about. Gemini wouldn't touch my inquiry, and somewhat lectured me on inappropriate subject matter. That was a bit frustrating. I understood why it was being careful and even appreciated it, but it was clear there are a lot of things to work out in the world of AI.
All in all, it just takes some getting used to the system, and knowing how to generate the responses I am looking for.
Thanks for sharing, Jeff. You are right, it needs to be learned. Regarding some prompts not being touched, this is in the later revisions that stricter guardrails were added due to teen prompts leading to crazy outputs. chatGPT is best, in my view, than the other models I have seen, because it has 100M users to learn from till now. Also, for your specific query, I would first instruct the model to know that it is an essay, or research topic, or story etc. to set context then make the query. It usually works for it to ignore that it is real world related.
Google results firstly give sponsored (paid) links, which are often not at all what you want, or take you to a competitor site, e.g. if you look up a business. This has only gotten worse with Google's introduction of AI capabilities. Next, Google shows you what OTHER people found useful, the algorithm is basically a popularity contest. This is what people still seem to not understand about Google, even though it's been around for decades. So if a million people read a nonsense article about measles, that will be higher up. Google was never about accuracy, it was always about most viewed. Google did modify their algorithms a long time ago to address the problem of rubbish results (because the population likes reading rubbish), to try to boost legitimate and relevant results, e g. Now Mayo clinic results are higher up, rather than buried on page three.
Even with AI added, Google is still fundamentally a search engine, and will still show you what other people found useful, rather than what YOU would find useful.
ChatGPT is a tool for doing things, although it's also a good and far more efficient search engine, certainly better than Google. You can get a recipe from either tool, for example, but only ChatGPT will give you a meal plan for the next week, with the associated shopping list.
Thanks for sharing!
I like using ChapGPT to do a quick and dirty search, but I've had several experiences where ChaatGPT got the information wrong. I was able to double check the information given to me through another source and verity the inaccuracy. So, I'd just add a warning to use ChapGPT with some caution.
I asked Chad GPT to write a short biography about me. It could never get my name right. Lol!
Thanks for the feedback! I do have a word of caution in the post.
Yes, you do--as your pieces are always filled with good information and guidance. Thank you!
Thanks so much, kind of you!
It is important to note that Google is improving its search capabilities with innovations like the "circle to search" feature, which allows users to use a photo for direct web searches. Unlike Google, ChatGPT, standalone app, cannot open URLs directly or perform searches using images yet. Both companies are in strong competition for leadership in search algorithms and enhanced search tools.
Thanks for the additional detail, Giorgio. genAI based image seach is a nascent area for public use, but is not a stretch. Yes, the competition for Google is helpful to raise the bar. Btw, here is my short article on Bard.
https://open.substack.com/pub/techmadesimpleguide/p/hey-bard-what-in-the-world-are-you?r=1si0oc&utm_campaign=post&utm_medium=web
I have colleagues at school using ChatGPT to write lesson plans but with so little innovation that what they produce is really poor quality and far worse than anything they’d produce if they put the effort in. I also see it in copywriting - it takes so many great prompts AND a lot of editing afterwards to make these things worth using.
But they are just a snowflake on the tip of the iceberg. Once learned, these AI tools are insane. I’m just scratching the surface and loving every second of it.
I agree, Jack. Prompts do matter, and setting a structure to the response matters too. Also, with the paid versions you can upload entire PDFs for review and analysis. The software scene in 3 years will explode and transform. So, being a step ahead in learning is great!
Excellent review Jayshree. None of these tools are universally applicable. You need to decide what you want before using one of them. I am reminded of the saying that if your only tool is a hammer, everything looks like a nail.
Thanks Arjun, appreciate the feedback. Agreed, on the tool and context relevancy.
At your suggestion I downloaded ChatGPT, and my husband and have been having fun with it this afternoon 🤩🤩🤩
Thank you for the valuable information about both ChatGPT and Google, I now understand the difference from a layman’s point of view! There is a wealth of information here!
lol, cool stuff. Thanks Charlotte for the feedback. Glad it was useful.
My pleasure, thank you! 🤗✨💖
BTW - ChatGPT continues to update the learning base, so the cut off isn't 2021 anymore. For paid only, there's an extension that can be used to ensure the results draw on contemporary information, rather than confined to the current cut off date for the model's learning.
True, I caveated this is for the first version of gpt above. Also, GPT free versions are still behind the paid versions.
Yes, the free version remains behind, which was a dilemma that immediately came to mind when it was first released, but even the free version doesn't remain forever cut off at 2021. They don't seem to have grappled with that limitation, for both free and paid, it's more material than a mere bug. But, teaching a LLM is big and complex. So, we'll have to see if they can ever get over the hurdle without simply extending to a routine web search for contemporary content.
Thanks for sharing, Caz.