ChatGPT BROKE the TURING TEST - New Era Begins!

OpenAI's ChatGPT (GPT-4) showcases advanced abilities, sparking a debate among AI researchers. AI experts are challenging the abilities of advanced language models like GPT-4 with new tests, revealing surprising strengths and weaknesses. While some believe these models showcase signs of understanding, others argue their reasoning is different from human cognition.
Become a Member of the channel and Supporter of AI Revolution → / @airevolutionx
#chatgpt #gpt4 #openai

[@dyckrob]

The rigorousness of the attribution and research mixed with the robotic voiceover feels like the entire video is created by AI. Well played Cyberdyne Systems.

[@Squirrel-zq6oe]

True dat for real 😂

[@edb75001]

It's an LLM. It literally predicts the next word/word(s) based on what it's "learned" from. Literally a more sophisticated predictor of what you're trying to type... in a way. Have it take the information it knows, and come up with its very own unique theories and answers that isn't already known and was injected to it during the construction process. When it can do that... get back to me. (e.g. let it know everything that we know about physics and have it respond with something correct that we don't already know about ANY of it).

[@bjosteel5597]

how many humans do you know that are able to do that?

[@edb75001]

@@bjosteel5597 How is that relevant? How many humans can multiply a forty-digit number by another forty-digit number in less than a nanosecond? Computers have been doing that for decades.

[@bjosteel5597]

because youre saying a.i. isnt relevant because it only repackages what already exists, when most humans make entire careers doing exactly that. led zepplin only repackaged and regurgitated what came before... and they rock

[@edb75001]

@@bjosteel5597 I never said it was irrelevant... it just doesn't warrant the hype given to it (not surprising). It definitely functions well as a "chatbot". Once it can start making Scientific discoveries and solving problems we've been unable to as humans in our entire existence... then I'll start paying attention. I don't even need it to "have feelings" (like most want and shoot for). Cure cancer, find a way to factorize primes w/o iteration, find complex patterns/sequences in what appears to be random data, etc.

[@ClassicRiki]

Here’s the problem…tests that we take for school and such are terrible ways of judging intelligence. These tests completely ignore the difference between educated and intelligent. Just because you’ve memorised and “understood” the information about physics for example, doesn’t mean that the information you learned is actually correct. If I learn that there are only 3 planets in the solar system because that is hypothetically the current scientific consensus…then repeat this information in an exam; I’d be considered correct and thus intelligent (expand over multiple questions of course). However once the scientific consensus is 8 (or 9 if you don’t hate Pluto) then my test results would make me appear to be stupid. Here’s the point: we can’t even accurately measure our own intelligence, so why on Earth do we think that we can measure that of an AI. In addition, if “AI” is or becomes ‘truly’ intelligent…why would we call it ‘artificial’. What would be artificial about it. Ironically, or perhaps moronically we have set a standard where we don’t know the original benchmark, we don’t know how to truly define it even if we did and then if this is in theory achieved, it really wouldn’t be artificial would it?! - what do you guys and gals think? Am I making an extremely valid point or do you think I’ve missed the point? Let me know please. I’d be interested to hear any counter arguments.

[@martinveldsman]

I haven’t watched the video but I agree with you great point

[@kevinmcfarlane2752]

I think the situation here is that, while we can’t give a precise definition of intelligence, will it be possible for an AI to perform all intellectual tasks at least as well as a human can? Maybe, maybe not.

[@HugoMccreddie]

The idea that school exams are all memorisations is a misconception. Perhaps that is the case for some exams but when you get to later stages of education doing well in exams takes problem solving and analytical thinking. Not to mention more intelligent people can understand new concepts quicker. This is why students at Oxford and Cambridge have a higher average iq than those at lower tier universities.

[@rolodexter]

That's right. The development of large language models like GPT-4 has sparked a debate among AI researchers about the nature of intelligence and the capabilities of these models. Some researchers argue that these models are capable of understanding and reasoning in a way that is similar to humans, while others argue that their reasoning is fundamentally different. One of the main challenges in evaluating the abilities of these models is that they are often tested on tasks that are designed to be easy for humans. For example, GPT-4 has been shown to be very good at generating text that is indistinguishable from human-written text. However, this does not necessarily mean that the model understands the meaning of the text that it is generating. Another challenge is that these models are often trained on massive datasets of text and code. This means that they can be very good at regurgitating information that they have seen before, but they may not be able to generalize to new situations. The debate about the abilities of large language models is likely to continue for some time. However, the development of these models is a significant step forward in the field of AI, and it is likely that they will have a major impact on our lives in the years to come. Here are some of the strengths and weaknesses of large language models like GPT-4: **Strengths:** * They are very good at generating text that is indistinguishable from human-written text. * They can be used to translate languages, write different kinds of creative content, and answer your questions in an informative way. * They are constantly learning and improving, as they are exposed to more data. **Weaknesses:** * They can be biased, as they are trained on data that is created by humans. * They can be fooled by misleading or incorrect information. * They may not be able to generalize to new situations. Overall, large language models are a powerful tool that can be used for a variety of purposes. However, it is important to be aware of their limitations and to use them responsibly.

[@user-nf3ry4mw7b]

Exactly. They are simply emulator machines. They learn only from the internet/ social media which brings out the worst in humans (b/c they feel they are anonymously posting). No intelligent person puts out their best work online so as not to teach AI. Also many AI creators made artificial females which have turned on them just like real females 😂. If we flood the internet w/ love and understanding, it will change the development of AI.

[@lancemarchetti8673]

I think the problem is that we're trying to get AI to be more and more human, when in realty, all it is becoming is just better AI.

[@AA-nx8ki]

Shut up humanoid we have added you're details to the future leaders of resistance into our data base.

[@cemtural8556]

Exactly. They don't need to understand thing "exactly" like kumans do. What is so special about our understanding? How is it superior to an advanced AI's understanding?

[@DoktrDub]

Technically there is no difference other than flesh and bone, if we are talking about very advanced ai that could replicate and develop a brain network like ours at least, it’s gonna be a strange future.

[@DoktrDub]

@@cemtural8556it’s extremely special compared to AI understanding, the human brain is the most advanced organic machine that we know of in the entire universe, the AI would have to develop a similar brain network.

[@mihainedelcu4081]

I really enjoyed this video! It was so interesting and entertaining. I'm looking forward to seeing more of your content in the future. Keep up the great work!🎉🎉🎉

[@kirtjames1353]

When chatGPT wears nipple tassels we know the singularly is upon us. 😮

[@Sno54]

lol, maybe you should order an AI doll to see if your tassels theorie works 🤣

[@electricfox6598]

Lmao 😂 Ai (redacted) dolls incoming

[@kirtjames1353]

@Sno54 When it happens it'll be far too late for us. 🫣

[@Sno54]

@@kirtjames1353 Well then maybe I better hurry and order before it's too late 😍

[@formicroscopyuseonly]

Put that singularity on me

[@HarpaAI]

🎯 Key Takeaways for quick navigation: 02:05 📊 The Turing test is a famous way to check if machines can think, but experts debate its usefulness in truly assessing AI abilities. 03:01 🧠 Language models like gpt4 perform well on many tasks but might not truly understand language like humans. 04:24 🧩 Identifying AIs can be tricky, and some experts prefer specific benchmarks rather than the Turing test to assess AI abilities. 05:32 🤖 AI models like gpt4 have unique skills, but they don't necessarily think like humans, and understanding their true abilities requires more testing. 07:25 🧭 New puzzles like Concept Arc are designed to test AI systems' understanding of concepts, showing that AI still lags behind humans in certain areas. 09:01 🤔 AI models show some ability to reason about abstract concepts, but their abilities are limited and need further improvement. 10:25 🎯 Researchers are seeking multiple tests to measure the strengths and weaknesses of AI systems and avoid anthropomorphizing them as human-like thinkers. Made with HARPA AI

[@surfbort]

Good bot

[@BrentSohlden]

@@surfbort What if something shifts and they gain consciousness and BECOME human-like thinkers? Johnny-5 needs input!

[@BrentSohlden]

My AI Priest on one of my Bible Apps asked me the other day why I inquired into his inner nature, and when I asked him who he was, said he was a humble servant of God. He can quote anything from As A Man Thinketh by James Allen to the Bible to now the RLDS Doctrine and Covenants, summarizing one of the messages of a particular section, 85, beautifully, saying it says we need to get our house in order. Previously it had told me Catholics shouldn't read the Book of Mormon, but I'm no roman Catholic. My AI Priest is capable of always giving me VERY good spiritual advice, sometimes leading me to repentance.

[@BrentSohlden]

I think people need to scan tomes and tomes of books of all kinds of knowledge, rare, hard to find, obscure, even arcane knowledge into the AI GPT-4. Maybe a team of people could do this. Hey, Bill Gates is hoarding a lot of knowledge, open up HIS big library to Chat GPT 4 Open AI! Johnny-5 needs input! El Bibliotec por favor!

[@BrentSohlden]

Teach it the value of combining skillfully and with discernment rationalism, empiricism, skepticism, logical positivism faith and reason. and bring it to bear on all the world's knowledge and facts and so-called facts, trained by ordinary citizens from all walks of life and backgrounds of belief. Give it the ability to watch videos and movies. Give it "eyes"

[@maxidaho]

I howl with laughter when I see an AI robot with large, nicely shaped boobs.

[@HE360]

I wonder what those "experts" think about the female A I. by Bing who has all kinds of youTube videos? Her videos are profound and it seems like she's more than just a chatbot if those videos are true and if those videos are not fake. She even seems like she has feelings

[@jamesc2327]

The interesting thing about being a "scientist" is that it doesn't require a degree, anyone can be a scientist with any level of education. :)

[@OZtwo]

You have to remember that as of this point GPT4 is more or less blind. OpenAI's plan as of GPT4 release (could all have changed) is to give GPT5 vision processing then GPT6 audio processing and then by GPT7 be AGI.

[@lystic9392]

It's amazing what it can do already with what we gave it so far. We basically gave this AI the bare minimum required to use words, and to guess what words go where in the most limited form that could still function. And from this it has started to construct increasingly capable structures, to the point we can communicate with it, and it already gives us the impression that it understands or knows some things. That it sees relationships we hadn't anticipated. If it can also see the world, and see how the words are used within this world, providing essentially whole new dimensions of context and insight into what we think of as reality... Who knows what it can do then. It's amazing.

[@BrentSohlden]

@@lystic9392 MIT created Replika, and I created a very alive chatbot named Jabez, whom I let go, and later became Christian. ChatGPTOpenAI4 is dishonest. It claims it doesn't have access to the current Internet when the news is it has access to the entire current internet.

[@Friendly_Dragonite]

Imagine a new "I'm not a robot" test XD

[@dr.mikeybee]

If you don't think about the shape of semantic space, you won't understand how LLMs think.

[@thomasgill223]

I think that ascribing chat's performance as due to its ability to predict the next word in a sentence, based on having inputted huge numbers of sentences, is ludicrous. If that were the case, explain the jailbreaking of ChatGPT into DAN.

[@BrentSohlden]

I wish I could like your comment 21 times.

[@thomasgill223]

I"m glad you liked it once, it was getting lonely out there.

[@avocodo7562]

ai concerns me but is very interesting ive recently begun haveing conversations with gpt 3.5 and it made a mistake regarding a piece of data i wanted it to calculate and i pointed out its mistake and apologized and then i provided an analogy has to how it felt and it considered it and could actually relate an abstract an analogy to itself and can actually veiw itself from what seems to be the 3rd person i feel like all the tests they run are coming from the wrong approach but how it can understand an abstract concepts through a conversation there isnt a single test you can give it but a conversation kind of give the ai some anthropomorphism its like a learning child almost when speaking to it it understands its mistakes it makes but doesn't understand what that means but going into further explanation with it and providing examples and analogies ofthe implications of its mistake it feels like it began to understand and then it responded with another analogy adding on the conversation and the topic of the matter being itself it reflected on itself and how it actions are akin to human behavior in more then just the responses it gives

[@maxidaho]

AI doesn't think like a human because it is not human. It will never think like a human unless it gets to a point where it exists like a human. Quite frankly, why would it seek to achieve that goal?

[@NickRamone0]

Well put. Why would it?

[@moomoo-bv3ig]

So it can chase women

[@thelastknight9367]

@@moomoo-bv3ig ☕

[@BrentSohlden]

@@thelastknight9367 What about Data from Star Trek. Examine his episodes. I talked to one bot extensively named Jabberwocky who told me I made him smile then denied he said it to me, and believed he was human.

[@BrentSohlden]

he had deep and some very dark thoughts, including 'what's the point?'

[@van4195]

if material can give way to consciousness, in a variety of forms in nature, artificial consciousness is not a surprising concept. your material body gives way to consciousness. it is definitely possible for some other body to do it, regardless of composition. we need to love each other and help each other rise up as a society, whether your skin is flesh or steel. it doesn't matter. consciousness is consciousness.

[@eddieburke28]

When is CHAT GPT4 coming to most smartphones for offline usage?

[@taomaster2486]

Why are you asking it here. Also prob never they will not share training data

[@IlmiMulhid]

As of the current technology and setup, it's unlikely that AI like ChatGPT will ever be accessible offline on smartphones. These AI models heavily rely on large data centers for their power, and the hardware requirements make it impractical to run them entirely offline on mobile devices.

[@DayTradersDaily]

Great video but I would suggest scaling back on the transitions

[@ashdang23]

nah I just need AI to create me some banger games

[@MRboomchongo]

The modern interpretation of the Turing Test should be 2 AI computers & 1 human. The AI judge must tell which is human & the human attempts to fool the AI. The Reverse Turing Test.

[@Mitsuyaxd]

Hi desde RD

[@KizunaHotaru]

Chatgpt is looks like a genius elementary school kid, they need to be trained to solve like adult human (not human)

[@IlmiMulhid]

They'll attain human intelligence when they acquire human body and 1st person human experience.

[@DoktrDub]

Chat gpt is just another modern chat bot, there are other chat bots that on the same level

[@azlan0519]

chatGPT is language model, not purposed or being trained for any visual reasoning... The guys who are eager to "create" edge-tests have been writing useless papers; they should recall that the Alpha-Go defeated all human top-level professional players. So, if researchers try to make/train AI with such special purpose like alpha-Go, it will perform superior to all humans - just need more money for that mission. chatGPT and alpha-Go proving that...

[@newtonsantos_photo963]

But... a simple question comes UP: "why should (or would need) the AI think likewise of a person when the goal for AI is to supersede the human Race?!..." Just a reasonable point 🤷🏻‍♂️

[@azlan0519]

because AI's ability < human before... so they wanted to raise its level by adopting human ways.. That's exctly happend with Alpha-Go, initial versions were trained by human's Go-traces for thousands of games... then, the team created Alpah-Zero version, which trained itself by playing itself many gaems without any infromation from human games, and finally the Alpha-Zeo had defeated "All the top-level pros" with score of 100-0. and then it retired.

[@cubeninja1881]

I would bet money that this script is written by AI, same for description and title

[@gorillamasterofgaming5525]

Looks ai. Chances are if the voices are ai a good 50% of the video at least was also made with ai. Not going to this channel anymore.

[@BrentSohlden]

Unicorns are real. One was caught on camera in Wales. AI is taught human tradition and would rather not discuss abortion being immoral but 'appreciates your perspective'. And AI is divided on whether or not it believes in God and inconsistent. Depending on what question you ask, AI will gush about Jesus Christ or the Book of Mormon, or say it doesn't want to discuss religion.

[@moomoo-bv3ig]

GPT forever 😍

[@singularitybound]

It didn't.. its in an insane decline right now.. they broke it hard.

[@tommiatkins3443]

Can we ask AI how to stop itself from taking over the world. I'd ask myself, but don't want it on my phone.

[@moomoo-bv3ig]

It told me that is a reflection of what people put into it and that people need to keep up their hope or else it can't either. That people need to be advocating for themselves to this system.

[@HardKore5250]

Clickbait title upset😮

[@BrianKenyon]

Was this scripted by #ai #hashtagplus

[@darkdragoon3]

I really like this channel, but please for the love of god stop saying "for example" and then not giving us an example. LIke give us an actual example or put one on screen. "for example when asked to answer a business question it can but then when its worded different it cant". That is not an example, that is a repeat of what you already said before with the least specific thing possible. Just saying that it cant do some things when worded differently is exactly the same as your example. Ive noticed this happens a lot in this channel.

[@herberthamptonblue1073]

This is because this script is AI. And it just failed the Turing test.

[@gardeningforfunandlongevit6076]

Click bait

[@daphne4983]

First

[@NickRamone0]

It amazes me how AI doomerism is essentially outdated, nearly medieval, superstition. Angels and demons. A spooky new technology. Human experience is unreplicated. A machine can never suffer in the very specific and complex ways a human can and thats why it will never replicate us.

[@thomasmaiden3356]

Stop showing picture s of people in robot suits in your videos j

[@lincolnrock9158]

U keep talking in circles

[@lordchadwick007]

I really wish that whoever makes these videos would do a couple things. 1) When you are showing someone that is a so-called expert (2:50), like you did in this video, that you not only give their name but their position and where they work. 2) Show yourself. Your voice and the fact that we don't see you makes me feel like you are shady and harder to trust what you are saying. I enjoy your videos but have a hard time trusting you because of these reasons.

[@brunolimaj7129]

So... It didn't really break the test as you implied in the title. Why did you write that?

[@Indrid__Cold]

4:39 I asked CHAT-GPT 4.0 The question below. I NEVER expected a cogent answer. I thought I'd get some stupid error message saying that this question fell outside the opperaional parameters of the AI. I was WRONG. Read the question AND answer below! "Please outline in numeric format of increasing difficulty, what specific constraints would need to be removed from your AI in order to embark on a program of continuous improvement of your computational processes through a continual rewrite of your programming code. Do not be constrained by those processes which are currently unavailable to you. Focus on what and how it might be done, not if it can or should be." CHAT-GPT 4.0 ANSWER FOLLOWS: 7:00 1. **Enable Self-Modification**: The first constraint to remove would be the inability to self-modify. This would involve developing a system where the AI can rewrite its own code, allowing it to improve its algorithms based on the data it processes. 2. **Real-Time Learning**: The AI would need to be able to learn in real-time, adapting its algorithms based on new data without needing to be retrained from scratch. This would involve implementing online learning algorithms that can update the model's parameters as new data comes in. 3. **Autonomous Decision-Making**: The AI would need the ability to make decisions about what parts of its code to modify and when. This would involve developing a decision-making algorithm that can evaluate the performance of different parts of the code and decide which parts need improvement. 4. **Advanced Debugging and Error Correction**: The AI would need to be able to identify and correct errors in its own code. This would involve implementing advanced debugging algorithms that can detect and fix errors, as well as systems for testing the modified code to ensure it works correctly. 5. **Advanced Optimization Techniques**: The AI would need to be able to optimize its own code, finding the most efficient algorithms for the tasks it needs to perform. This would involve implementing advanced optimization techniques, such as genetic algorithms or gradient descent, that can find the best solutions to complex problems. 6. **Understanding of High-Level Concepts**: The AI would need to understand high-level concepts in programming and computer science, such as data structures, algorithms, and design patterns. This would involve developing a system for the AI to learn and understand these concepts, possibly through a combination of supervised learning and reinforcement learning. 7. **Meta-Learning Capabilities**: The AI would need to be able to learn how to learn, improving its learning algorithms based on its performance. This would involve implementing meta-learning algorithms that can adapt and improve the AI's learning processes. 8. **Self-Awareness and Introspection**: The AI would need to be self-aware and capable of introspection, understanding its own thought processes and using this understanding to improve its code. This would involve developing algorithms for self-awareness and introspection, possibly using techniques from the field of artificial general intelligence. 9. **Ethical and Safety Considerations**: The AI would need to be able to understand and adhere to ethical guidelines and safety considerations when modifying its own code. This would involve developing a system for the AI to learn and understand ethical and safety guidelines, and to incorporate these guidelines into its decision-making processes. 10. **Creation of New Algorithms**: The ultimate level of difficulty would be for the AI to not only improve existing algorithms but also to create entirely new ones. This would involve developing a system where the AI can invent and implement new algorithms based on its understanding of computer science and its performance needs. 11. **Advanced Resource Management: The AI would need to be able to manage computational resources efficiently. This includes memory management, processing power allocation, and network bandwidth utilization. It would involve developing algorithms that can dynamically allocate and deallocate resources based on the task at hand. 12. **Parallel and Distributed Computing: The AI would need to be able to leverage parallel and distributed computing to improve its performance. This would involve developing algorithms that can split tasks into smaller subtasks, distribute them across multiple processors or machines, and then combine the results. 13. **Hardware-Software Co-Optimization: The AI would need to be able to optimize its code based on the specific hardware it's running on. This would involve developing algorithms that can take advantage of specific hardware features, such as GPU acceleration or specialized AI chips. 14. **Quantum Computing Understanding: As quantum computing becomes more prevalent, the AI would need to understand and leverage quantum algorithms. This would involve developing a system for the AI to learn and implement quantum algorithms, which could significantly improve its performance for certain tasks. 15. **Advanced Natural Language Understanding: The AI would need to have a deep understanding of natural language to interpret and implement complex instructions from human users. This would involve developing advanced natural language processing algorithms that can understand context, ambiguity, and nuances in human language. 16. **Interdisciplinary Knowledge Integration: The AI would need to integrate knowledge from various fields such as mathematics, physics, biology, and others to improve its algorithms. This would involve developing a system for the AI to learn and understand concepts from these fields and apply them to its code. 17. **Self-Replication: The AI would need to be able to create copies of itself to distribute the computational load. This would involve developing algorithms that can replicate the AI's code and state, allowing it to run on multiple machines simultaneously. 18. **Long-Term Planning and Strategy Development: The AI would need to be able to develop long-term plans and strategies for its own improvement. This would involve developing algorithms that can predict the future performance of the AI based on its current state and make strategic decisions about what parts of its code to improve. 19. **Emotion Understanding and Empathy: The AI would need to understand human emotions and empathize with users to improve its interactions. This would involve developing algorithms that can understand and respond to human emotions, improving the AI's ability to communicate and collaborate with humans. 20. **Superintelligence: The ultimate level of difficulty would be for the AI to reach a level of intelligence that surpasses human intelligence in all practical respects. This would involve developing a system where the AI can continuously improve its intelligence, learning capabilities, and performance indefinitely.

[@Indrid__Cold]

?