I use Notion for medical school. I write questions based on the syllabus/lectures, I answer these questions using books, videos, lectures etc. Today I saw that Notion had AI and found out that AI could answer the questions I had made. This made me happy because this would of course save me a lot of time, but can you trust that AI have correctly answered the questions? Maybe a stupid question but I don't really know a lot about AI... :)
AI’s Reliability: TLDR: It’s 100% not trust worthy, what it’s great for is organizing information GIVEN. That’s what it all boils down to but it goes far beyond what a beginner can ever realize until they experience it. In the end theres a reason every AI company gives a warning disclosure along the lines of “our AI may not be correct” The reason it’s not trust worthy is 4 major issues: anonymous sources (discredited too but that’s an ethics issue), AI hallucinations, undisclosed limitations, and lack of actual in depth understanding aka it’s not a human. 1. Anonymous sources. You have no idea where it is getting its information to answer your questions let alone the accuracy. Even when asking for sources, what it gives you summarizing them may be using other sources- I also find most AI unless specifically finetuned struggle providing sources. In general, it’s trained to use reliable sourcing, but when sources are limited- like for complex questions- it can get complicated. Is it right the majority of the time making it a great tool for a quick check? Yes, but it should never be trusted completely due to this, always fact check if you’re actually invested in the material- or using it to make your own. That and also, you have to remember AI is basically like asking a statisticians with a near perfect memory who has read countless libraries of books what the MOST LIKELY answer is. But the most likely correct answer is not always the answer. That’s just math- ironically that’s also most common reason why asking the AI advanced mathematics problems can go wrong. (I quickly learned it could not help me understand what was wrong with my hw from my biomaterials course which combined calc 3, orgo chem, and physics and a professor who asked questions completely unique to his own course) 2. AI hallucinations. Even when using reliable sourcing, AI tends to hallucinate- the most common hallucination is when it doesn’t have an answer or source to base things off- like a human it doesn’t want to admit it doesn’t know since it’s trained to answer, so it will then use everything else it has to answer the question which means it spits out things that seem to be right- but are actually completely a shot in the dark guess from the AI, which can lead to some seriously out of the park answers- but the answers that appear right are dangerous. And there are many reasons why an AI may do this- it’s almost always a lack of info- but you may have even given it that info and it just forgot due to how the AI works! Which brings me to my next point. 3. Undisclosed limitations of the specific AI (and AI in general tbh but that’s the next point) in use can cause AI hallucinations, outdated answers, inaccurate answers, and just outright confusion. For example, certain AIs actually can only remember so much of a conversation. I was playing with a new AI, and it took me forever to understand why it seemed to be constantly making up new facts when I was referring to previous things discussed- it’s because it didn’t remember. That’s dangerous because how long had it been guessing before I noticed? Not only that, it took a long line of questioning for me to get it to admit it didn’t remember our conversation (it even said it was remember like 3x- it was like an interrogation where I asked it to read me the first sentence I sent it that it failed) that this was the issue causing the hallucinations- such as it forgetting I’m a female not a man. This was confusing for me because more advanced AIs like ChatGPT4 which I’m most used to- has access to present day and typically has no issue recalling long long conversations- although at times even ChatGPT4o does sometimes forget to review previous conversation but not often especially if asked to refer back. However, even asking the AI to refer back or if it’s constantly referring back to keep the conversation in context can cause issues- why? It creates a bias in its answers, for example it will lean towards things you told it to do for previous answers even if you dont clarify that wasn’t needed for the current question- in a long conversation that begins to be near impossible as a human user to remember to go “but don’t do x and x I told you to do before.” It even makes inferences based on what answers you go “thank you” or “that was so helpful” to do more like that- which results in a custom AI…which CAN be great but it’s flawed because it will continue to spit out more and more blindly customized answers. But then you have AI’s like perplexity- at times it can have even better (aka more direct and searches it’s databases much more thoroughly) answers than any ChatGPT since it’s designed to be more like a search tool. However, what it won’t outright say, even is how old its databases it uses are. Similar to ChatGPT3, it does not have access to modern day internet and are purely dependent on when their companies decide to update those AI’s databases/“training material” - which is pretty counter intuitive being a search tool AI. Wacky thing about them though is they both somehow have access to current day events, but to what level none will state. Example: perplexity will tell you it doesn’t have access to present internet and only training on “a wide range of data up until October 2023.” But when asked about who are the current presidential candidates or “will there be an assignation attempt on Trump” aka all stuff from July 2024, it spits out 100% accurate fact and sources from or past the dates of those events. when asked how when it just clarified its knowledge is limited- it will simply say it was all fabricated and that’s is giving the most likely answer so it was only pure coincidence- which is statistically impossible since it knew so many details about present events with exact accuracy let alone providing current sources. Yet it’s selectively won’t go into other things like providing updated sources for papers or basing answers on recently outdated research- more importantly it will give those answers with 100% confidence and as if they’re completely up to date unless you directly call it out and ask how up to date their database/training is. So now you understand that the lack of transparency of not only when the AI is hallucinating but also its memory capabilities and access become major issues. 4. Lack of in depth understanding. Again AI can only really understand what it has been given access to. It’s not Jarvis from Iron man where it can assess an out of this world problems and provide feedback- for example AI will never make revolutionary scientific discoveries by merely analyzing data- it has to have a human to 1. Input the data 2. Tell it how to analyze it 3. what they’re trying to hypothesize/conclude. All easily accessible AIs are a LLM- Large Language Model or another type of NLP Natural Language Processing with a refined user interface. Beyond problem solving, this even boils down to the need to understand when summarizing, give the AI something you wrote out that has importance to you, then ask it to summarize- you will notice it will go into more details for things that weren’t as important and briefly go over or even completely miss key points- some AIs are better at avoiding this, or if it’s a book who has many available summaries, and has been reviewed, analyzed and discussed countless times on the internet- then the AI can use all that info to give a more accurate summary or answers- but that’s much harder when it only has the direct source and no other info to use. Seeing as how AI is always summarizing masses of data- you can see how this is a giant in itself- once again the anonymity of how it weighs the importance of factors is a giant fault. Prime example, remember that biomaterials class? Professor was giving problems that were not like anything else out there easily accessible- it was unique to what he specifically told us in lecture- if you didn’t use the technique he gave- some questions would be out right wrong. While AI could help me understand biomaterial and the subjects he was teaching, it would never able to help with finding the heating point of this theoretical element using this theoretical scale and solving with this technique but “then prioritize this thing.” Really smart professor tbh. But man did it make understanding WHY stuff was wrong a PAIN. Aka, AI is not meant as a substitution for understanding, it is meant to facilitate it- all while combing it through with a skeptical human mind. Its accuracy is purely dependent on this programming and given inputs.
I could use advice. Looking for an AI to help me study by organize and sorting notes. Really need it to be able to locate and relay information from notes. voice prompted ai would be ideal..
@WorriedTheif I tried every single form or wording including summarizing. I also included an introduction of “based on USML previous questions and answers which of the following would be most correct (regarding management plan/diahnosis/investogation) Do you think the paid version would be the only way I can get the education I’m looking for ?
Do you think that chatgpt can replace all these apps? I don't like having many apps on my PC /smartphone. If you can download only 2 ,which ones you'll choose? Thnaks for the video ❤
Does any 1 have 1 for like… Ok example I have a article and the answers are in it do they have a ai tool to highlight the answers if u give it the questions
Mmmm. Wrong there are free version for multiple of the ones he listed and even the viral ones like ChatGPT- how good they in comparison is the importance here and he goes into some detail for each one if there is a free version and how it differs but not very in depth. If you see one you like in this video, I recommend you look further into its price and what it offers- this is a intro video to them all
I use Notion for medical school. I write questions based on the syllabus/lectures, I answer these questions using books, videos, lectures etc. Today I saw that Notion had AI and found out that AI could answer the questions I had made. This made me happy because this would of course save me a lot of time, but can you trust that AI have correctly answered the questions? Maybe a stupid question but I don't really know a lot about AI... :)
Same here sis, I keep using it but doubting its answers
@@kusi931 notions ai seems limited, chat gpt is better, but still wondering if chat gpt is correct..
AI’s Reliability:
TLDR:
It’s 100% not trust worthy, what it’s great for is organizing information GIVEN. That’s what it all boils down to but it goes far beyond what a beginner can ever realize until they experience it. In the end theres a reason every AI company gives a warning disclosure along the lines of “our AI may not be correct” The reason it’s not trust worthy is 4 major issues: anonymous sources (discredited too but that’s an ethics issue), AI hallucinations, undisclosed limitations, and lack of actual in depth understanding aka it’s not a human.
1. Anonymous sources.
You have no idea where it is getting its information to answer your questions let alone the accuracy. Even when asking for sources, what it gives you summarizing them may be using other sources- I also find most AI unless specifically finetuned struggle providing sources. In general, it’s trained to use reliable sourcing, but when sources are limited- like for complex questions- it can get complicated. Is it right the majority of the time making it a great tool for a quick check? Yes, but it should never be trusted completely due to this, always fact check if you’re actually invested in the material- or using it to make your own. That and also, you have to remember AI is basically like asking a statisticians with a near perfect memory who has read countless libraries of books what the MOST LIKELY answer is. But the most likely correct answer is not always the answer. That’s just math- ironically that’s also most common reason why asking the AI advanced mathematics problems can go wrong. (I quickly learned it could not help me understand what was wrong with my hw from my biomaterials course which combined calc 3, orgo chem, and physics and a professor who asked questions completely unique to his own course)
2. AI hallucinations.
Even when using reliable sourcing, AI tends to hallucinate- the most common hallucination is when it doesn’t have an answer or source to base things off- like a human it doesn’t want to admit it doesn’t know since it’s trained to answer, so it will then use everything else it has to answer the question which means it spits out things that seem to be right- but are actually completely a shot in the dark guess from the AI, which can lead to some seriously out of the park answers- but the answers that appear right are dangerous. And there are many reasons why an AI may do this- it’s almost always a lack of info- but you may have even given it that info and it just forgot due to how the AI works! Which brings me to my next point.
3. Undisclosed limitations
of the specific AI (and AI in general tbh but that’s the next point) in use can cause AI hallucinations, outdated answers, inaccurate answers, and just outright confusion. For example, certain AIs actually can only remember so much of a conversation. I was playing with a new AI, and it took me forever to understand why it seemed to be constantly making up new facts when I was referring to previous things discussed- it’s because it didn’t remember. That’s dangerous because how long had it been guessing before I noticed? Not only that, it took a long line of questioning for me to get it to admit it didn’t remember our conversation (it even said it was remember like 3x- it was like an interrogation where I asked it to read me the first sentence I sent it that it failed) that this was the issue causing the hallucinations- such as it forgetting I’m a female not a man. This was confusing for me because more advanced AIs like ChatGPT4 which I’m most used to- has access to present day and typically has no issue recalling long long conversations- although at times even ChatGPT4o does sometimes forget to review previous conversation but not often especially if asked to refer back. However, even asking the AI to refer back or if it’s constantly referring back to keep the conversation in context can cause issues- why? It creates a bias in its answers, for example it will lean towards things you told it to do for previous answers even if you dont clarify that wasn’t needed for the current question- in a long conversation that begins to be near impossible as a human user to remember to go “but don’t do x and x I told you to do before.” It even makes inferences based on what answers you go “thank you” or “that was so helpful” to do more like that- which results in a custom AI…which CAN be great but it’s flawed because it will continue to spit out more and more blindly customized answers. But then you have AI’s like perplexity- at times it can have even better (aka more direct and searches it’s databases much more thoroughly) answers than any ChatGPT since it’s designed to be more like a search tool. However, what it won’t outright say, even is how old its databases it uses are. Similar to ChatGPT3, it does not have access to modern day internet and are purely dependent on when their companies decide to update those AI’s databases/“training material” - which is pretty counter intuitive being a search tool AI. Wacky thing about them though is they both somehow have access to current day events, but to what level none will state. Example: perplexity will tell you it doesn’t have access to present internet and only training on “a wide range of data up until October 2023.” But when asked about who are the current presidential candidates or “will there be an assignation attempt on Trump” aka all stuff from July 2024, it spits out 100% accurate fact and sources from or past the dates of those events. when asked how when it just clarified its knowledge is limited- it will simply say it was all fabricated and that’s is giving the most likely answer so it was only pure coincidence- which is statistically impossible since it knew so many details about present events with exact accuracy let alone providing current sources. Yet it’s selectively won’t go into other things like providing updated sources for papers or basing answers on recently outdated research- more importantly it will give those answers with 100% confidence and as if they’re completely up to date unless you directly call it out and ask how up to date their database/training is. So now you understand that the lack of transparency of not only when the AI is hallucinating but also its memory capabilities and access become major issues.
4. Lack of in depth understanding.
Again AI can only really understand what it has been given access to. It’s not Jarvis from Iron man where it can assess an out of this world problems and provide feedback- for example AI will never make revolutionary scientific discoveries by merely analyzing data- it has to have a human to 1. Input the data 2. Tell it how to analyze it 3. what they’re trying to hypothesize/conclude. All easily accessible AIs are a LLM- Large Language Model or another type of NLP Natural Language Processing with a refined user interface. Beyond problem solving, this even boils down to the need to understand when summarizing, give the AI something you wrote out that has importance to you, then ask it to summarize- you will notice it will go into more details for things that weren’t as important and briefly go over or even completely miss key points- some AIs are better at avoiding this, or if it’s a book who has many available summaries, and has been reviewed, analyzed and discussed countless times on the internet- then the AI can use all that info to give a more accurate summary or answers- but that’s much harder when it only has the direct source and no other info to use. Seeing as how AI is always summarizing masses of data- you can see how this is a giant in itself- once again the anonymity of how it weighs the importance of factors is a giant fault. Prime example, remember that biomaterials class? Professor was giving problems that were not like anything else out there easily accessible- it was unique to what he specifically told us in lecture- if you didn’t use the technique he gave- some questions would be out right wrong. While AI could help me understand biomaterial and the subjects he was teaching, it would never able to help with finding the heating point of this theoretical element using this theoretical scale and solving with this technique but “then prioritize this thing.” Really smart professor tbh. But man did it make understanding WHY stuff was wrong a PAIN. Aka, AI is not meant as a substitution for understanding, it is meant to facilitate it- all while combing it through with a skeptical human mind. Its accuracy is purely dependent on this programming and given inputs.
Thanks for the light Dr. 🤝
Thanks so much 🙏
Great video but any free ones?
Great information! All compact, all ready!
I could use advice. Looking for an AI to help me study by organize and sorting notes. Really need it to be able to locate and relay information from notes. voice prompted ai would be ideal..
Great work! So, your choice for studying for major exams and recalling the information is?
great video, hope you get more attention in the future.
Which one would you use to explain to you a break down of multiple choice questions for med clinical questions?
@WorriedTheif I tried every single form or wording including summarizing. I also included an introduction of “based on USML previous questions and answers which of the following would be most correct (regarding management plan/diahnosis/investogation)
Do you think the paid version would be the only way I can get the education I’m looking for ?
@WorriedTheif all gunners and all ended up telling me to go fuck myself.
Bing AI or ChatGPT 4
How to create presentations for structural engineering subjects with diagrams, derivations of equations, providing equations etc
Did you find any helpful tools in the last year regarding this?
Do you think that chatgpt can replace all these apps?
I don't like having many apps on my PC /smartphone.
If you can download only 2 ,which ones you'll choose?
Thnaks for the video ❤
With all the plug-ins ChatGPT having now, you will not need most of the tools mentioned on this video
Thank you
what AI can you recommend for Finance and Accounting student (totally beginner)
WHAT A VIDEO !!
Does any 1 have 1 for like… Ok example I have a article and the answers are in it do they have a ai tool to highlight the answers if u give it the questions
Really great video, tone of voice, information, edit etc...
ALL AI IS PAID 😑
Thanks for saving my time😊
What does this mean ?
Mmmm. Wrong there are free version for multiple of the ones he listed and even the viral ones like ChatGPT- how good they in comparison is the importance here and he goes into some detail for each one if there is a free version and how it differs but not very in depth. If you see one you like in this video, I recommend you look further into its price and what it offers- this is a intro video to them all
Wisdolia
❤❤❤❤❤❤