Artificial Intelligence Is Lying to You! Here's Why...
Vložit
- čas přidán 4. 05. 2024
- With the world of artificial intelligence advancing so rapidly, it's important to highlight some of the things that are going wrong and how we can fix them! In today's video, I explore the notion of 'hallucinations' in artificial intelligence systems and how, by using a process known as retrieval augmented generation, we can fix such issues! A big shoutout to www.marqo.ai/ for helping fix these problems in the world of AI!
Marqo GitHub: github.com/marqo-ai
My GitHub: github.com/ellie-sleightholm
Delighted to be delivering tech and AI videos on Marqo's channels. Follow them below so you don't miss out on a video:
📸 Marqo Instagram: / marqo_ai
👨💼 Marqo LinkedIn: / marqo-ai
🐦 Marqo Twitter / X: / marqo_ai
▶️ Marqo CZcams: / @marqoai
🎵 Marqo TikTok: / marqo_ai
🔗 My Social Media 🔗
📸 Instagram: / ellieinstem
🎵 TikTok: / elliesleightholm
👩💻 GitHub: github.com/ellie-sleightholm
🎶Spotify: open.spotify.com/user/1162825...
▶️My Coding CZcams: / codeofthefuture
For those of you that are new here, hi there 🌞 my name is Ellie and I'm a Part III Mathematics Graduate from the University of Cambridge! This channel is where I nerd out about maths, tech, AI, physics and space so if that sounds like something you're interested in, click the subscribe button to follow along ☺️
Subscribers: 59,676 - Zábava
Yes ! The same thing happened with me : I asked chatgpt to calculate an orthocenter of a triangle given 3 points and it kept turning the denominator 0 and somehow produce -21/247 as the x - coordinate when I double checked the solution it had ignored its own formula and had performed addition instead of multiplication. When I pointed out the mistake it prompted: “yes it appears that I have made a mistake: recalculating” after performing the same operation again it made the same mistake when I pointed it out again it proceeded to go into a loop and started prompting nonsense and repeating the same prompt again and again.
the “yes it appears I have made a mistake” bothers me because you could essentially convince the AI it was wrong without having a valid reason.
I kept out some of the specifics if the problem I can reply on this comment if you(reader) need further info on my experience.
Interesting video. I'm a machine learning engineer in the US. Yes, LLMs do hallucinate, and it's a problem. There is research showing good results when an LLM is combined with an RL, though. I did a video about Google DeepMind's Alpha Geometry which initially scared the bajeezus out of me with how "AGI"-like it is. And there's other interesting work when combining LLM and RL together to control nuclear fusion in tokomaks. Anyway, an LLM is a tool. Like with any tool, it can be misused, like for example, using it for academic publications. :-) Nice meeting you and coming across your channel. Hopefully your new friend, Rae.
Is being sn engineer.mostly.fun and enjoyable if Inmay ask or donyou sot a t a desk in fornt of.s comouter all day every day which doesn't dound fun to me. Thanks for your time.
Hi, Ellie! I’m a fellow European, currently a senior in high school from North Macedonia. I’ve also been a subscriber for about a year, and I want to try putting into words how much of an inspiration you have been to me. I’ve always been fascinated by mathematics and physics, and just recently I finished a Python programming course and did some projects with it. Your videos are super informative and helpful for me, as I’m tackling advanced college math. I recently got admitted to a university in the USA as an AI major, and I can’t stop thinking of the influence your videos have had on me. Keep doing what you’re doing, and one day I hope to become as hard of a worker as you are! ❤
This comment has made my year 🥹 thank you so much for this and thank you SO much for supporting my channel, I can’t tell you how much it means to me!🫶 Huge congratulations, that’s amazing!! If you follow me on instagram, send me a message - I’d love to follow your journey 🥰 thank you again 🥹🫶
Very informative. You've elaborated the issue very well. ❤
Thank you so much! Stay tuned for more content coming soon 👀
@@EllieSleightholmThanks for the video and info hope you can ahare more astropnusics and neuoriscience topic videos as well when you can. Do you still mainly do research? And off topic, whonwrote the musical interlude at 3:00? I like it. Thanks very much.
chat gpt trying to calculate the concentration of something: "55.845g/mol×98=5471.91g/mol"
Very nice Ellie, ur way of explanation is nice
thank you!!
I follow you from a long time mam your content is high yield keep doing your job like that and help people like me....thank you so much
Thank you so much for such a lovely comment ☺thank you for supporting my channel 🫶
Thank you so much mam
I discovered the topic of "intelligence", thus, AI, without my discovery is simply "Artificial Idiot" or "Artificial Instinct"...
❤ thank you for sharing this information
Glad it was helpful!
Thanks PROF YOUR videos are awesome.
Thank you for supporting my channel ☺
At the end of the day, it all comes back to one thing: data, data, more data. Doesn't matter how good your RAG algorithm is if you don't have good data to feed it. Implementing RAG into any model is trivial. Finding large authoritative datasets that are more accurate than just scrawling the web is the hard part.
RAG is awesome to me but ensuring the quality of your data is good is such a massive part too, for sure! I'm planning a video on the importance of data and will be showing examples of bad vs good data with the RAG implementation.
@EllieSleightholm what can be done to prevent false data from users being put into it. IE a spam bot attack? (If I understand correctly openAI does something like this.)
@@EllieSleightholmThanks for sharing Ellie. I hope you can PLEASE respond to my question about math work/lofe balance and pay and vacation when you can. I would really appreciate it.
Awesome!
Actually I am interested in your intro music
ChatGPT has lied to me many times, every other conversation. Often, false information is presented regarding completely uncontroversial issues.
As an example, I was having a conversation regarding Edward I of England and I pointed out that he executed William "Braveheart" Wallace. ChatGPT corrected me and said it was Edward II and claimed he was king during Wallace's rebellion. I then asked the question about the date of Braveheart's execution because I knew ChatGPT was wrong. I got the answer that he was executed in August 1305, two years before Edvard II became king. When I pointed out that the dates didn't match, ChatGPT corrected his fictional story regarding Edward I and his son, with the usual apology.
I always get an apology for the false claims, but never an explanation as to why they happen.
I've found that ChatGPT lends itself primarily to surreal idea generation. Nothing it claims can be taken as fact.
I once had it make up a software library that didn't exist.
informative video thanks 👍
Thank you!
Indeed resolves a catastrophic error in Chat AI engines.
Do people have examples of incorrect answers? Very useful to show my pupils when discussing the issues of using AI for their work. Can be fantastic when used carefully. But many of pupils aren't fully aware of the pitfalls.
You can check the comment I made where I faced this issue
I'm barely realizing the I.T. factor given the A.I. amongst it
I may or may not have once convinced chatgpt that 2+2=5
Btw, love your videos
I asked chatGpt to write a song about Adolf Hitler. It refused doing so stating "I'm sorry, but I can't assist with that request." It can write other songs though. So the people behind has equipped ChatGpt with a "sense of moral" which I see as a big problem. Because who is to decide the morals of future AI-systems that actually may have a big impact on the world?
Why is E=mc2 and not E=mc3? Why not mc cubed?
U have to start with initial formula then after derivation you will get E= mc ^2 not E= mc^3
It's just a unit conversion factor. c is the ratio of of your unit of length to your unit of time so you can use it to convert between those two units. Energy is kg*m²/s² in SI base units. There are such relationships between all the units but not all conversions have a physical meaning. There's not really anything that special about the modern interpretation of E=mc² since we did away with the concept of relativistic mass as being overly confusing. What was originally meant was E=𝛾m₀c² which is equivalent to E²=(m₀c²)²+(pc)². We no longer define m=𝛾m₀ where m₀ is rest mass so people now interpret E=mc² as rest energy rather than combined rest and kinetic energy.
@@protocol6 fuck this. I hate physics, thank G-D I'm just in 8th grade not learning this stuff.
@@protocol6 chat gpt 🤣
You might not be inclined to answer this, but do you like to travel ? Did you ever think of visiting India, if you ever do plan to come here I'll do my best to assist you in every possible way I can ! Namastey.
Is Marqo someone whomsponsored your videos now? Just curious.
I asked a I to write me a report and all the information and it was wrong
You're not just intelligent, you're also very beautiful. I hope someday we can be friends. I really admire intelligent people.
The fisherman😂
🤗
🤗
X⁰=1????
Ai is so crazy man
mind blowing!
From chat gpt through make provide video in hindi because good content in hindi viral easy please soon 🔜 think and one makes in hindi to translation video like as dhruv rathi ❤❤❤❤❤❤❤❤
Ellie, are you implying that the AI chatbox has the potential to be a new-generation propaganda machine?
GIGO.
Garbage In. Garbage Out.
Yep, you know it! Cool that we can use RAG to help rectify that!
Its rather easy. I am sure machine learning systems are much faster than humans. They may be much cheaper than humans. And they may make less mistakes than human.
But in the end... you can not trust them more than a human cause they still make mistakes and will always do. Cause ANY computer system makes mistakes.
And again (like ALWAYS!) we are creating a technology and industry doesnt give a sh** how to deal with the issues that HAVE to be expected. Of course not. Our economical system is created in the way that companies try to externalize every issues unless they are somehow forced to face the issues.
In my opinion this means automatically that economy is sort of infantile cause it doesnt care for the consequences of its actions except makeing money and following laws (which already often is bend or even broken).
And we still think this is the best possible way to run our society?
Sorry - but this species deserves to fail. We are so dumb on a social level it almost hurt. Mail goal of companies is profit. Not making lifes better, helping people or maintaining our society. These are at best secondary goals. If something creates revenue in the short terms but will destroy our society on the long run... we will do it.
Anyway - thanks for your video 🙂
poa pratensis...
Ai is too much wrong 😅😅😅😅. We know it what you recommend to use