It has the correct number of letters. It excluded the spaces and punctuation. You saw there are 39 *characters* but subtract 4 spaces, a hyphen and a period.
28:49 Matthew, if you count the characters by hand they are 33, the online counting thing you did seems to count the dot and the dash and the space, as they are also symbols!
In regards to the north pole question, the original question on X specified whether the person walking passed the point when they turned left - not the starting point! In your question. I believe ChatGPT got it right, because if you walk around the planet going the same direction, you will always end up where you turned, and the if you continue walking on your path, you head back to the north pole! I am not sure your prompt was clear "where" the starting point was - was it when you turned, or at the north pole? Obviously if the starting point was the north pole, you would never return there unless you walked back on your original path. If you read the response it says that after it circled the earth, it walked back to the starting point. Your question actually said "walk as long as it takes to get back to your starting point" , so obviously it has to walk to the starting point at some point.
1:10:01 Thanks for asking my question! It was really surprising to me how precisely it answered the question. I had never seen a model answer this question correctly before... although Claude managed to answer it correctly on the second try with my help. It's challenging because it's true logic, and not something the model is trained on and it also requires counting letters too. The question was: "if aaaa becomes aAAa and bbbbb becomes bBbBb and yyyyyyy becomes yYyyyYy, what does cccccc become?"
@@pedramtajeddini5100A bit of logic is more than some people use lol That's already big... I am really excited to OpenAI to be able to make it cheap, this will be ground breaking
@@luizhenriquepaes8991 true lol... but I'm really curious to see if "deep logic" can be achieved with a silicone based AI model in my lifetime... Basically AGI
We already have AGI. I think I cannot compete with it in any field it seems. I am afraid as this is better than me in everything I could imagine. I am afraid about future, I think most of poor people and middle class will wipeout of this planet. It will become very difficult to earn money with hardwork and honesty. Why should company hire, as already better machine than a human already exist
To design a coil spring with the specified requirements, we need to calculate its parameters. The spring rate is given as 300N/mm. This can be related to the material and wire diameter using Hooke's law: F = k * x Where: F: Force (in N) k: Spring constant or stiffness (spring rate) in N/m x: Displacement from equilibrium position (deflection) We are given that the spring rate is 300N/mm, which we can convert to a more suitable unit for calculation: k = 300 * 10^3 N / m The maximum deflection of the coil spring needs to be calculated from its design parameters. The formula for this is given by: x_max = (D_o - D_i) * n Where: x_max: Maximum displacement D_o: Outer diameter of the coil D_i: Inner diameter of the coil (diameter of wire) n: Number of turns in the spring. Since we don't have enough information to directly calculate either 'D_o' or 'n', let's consider possible values for these parameters and see if any combination will yield a deflection within our target range. We can use a trial-and-error approach. We could choose common materials such as music wire (AWG 0.031"), steel or stainless steel, which are popular choices for springs due to their high tensile strength and resistance to corrosion. I will use the formulae below to calculate possible spring parameters: Wire diameter (D_i): 0.031" = 7.94 * 10^(-4) m Outer coil radius: R_o ≈ D_o / 2 The maximum deflection of a coil spring can be found using: x_max = (D_o - D_i) * n Now let's plug in some numbers and see if we get close to the desired value. Let’s use music wire with a diameter of 0.031". This gives us an initial guess for our calculations. With this, let's calculate possible parameters: Spring rate (k) = F / x ≈ 300 * 10^3 N/m To achieve a spring rate of k we can use the following formula: k = E * π^2 * d^4 * n / (64 * l) Where: E is Young's modulus d is wire diameter n is number of turns l is coil length Let’s plug in some numbers and see if we get close to the desired value. If we set E = 200 GPa (Young's modulus for music wire), d = 0.031" or approximately .000794 m, l= 10cm or 0.1m and n = 5 turns: k ≈ 201 * 10^6 N/m The spring rate is higher than what we need so let’s reduce the number of turns to get closer. If we set n=3, then: k≈52*10^5N/m which is still a bit high. A lower value for E will result in an even higher spring rate so let’s try reducing the length instead. If we reduce l to 1cm or .01m, then: k≈20*10^5N/m Now our spring rate is lower than what we need but still not quite right so let’s increase n a bit and see if that works. If we set n=4 turns: k ≈ 81 * 10^5 N/m which is closer to the desired value. With these parameters (D_i = .000794m, E =200 GPa, l=.01 m and n = 4) we can now calculate our maximum deflection: x_max=(R_o - R_i)*n where: D_o ≈ π * (2*R_o) We know that x_max must be less than or equal to 25mm. Since we have an upper limit for the deflection, let’s calculate R_i and R_o using this information. D_i = .000794 m n=4 turns First find x_max: x_max=(D_o - d)n Rearrange to solve for (d): d=D_o-nx_max/D_o Now substitute the equation above into another formula that relates D_i, n and R_i: D_i = π * 2 * R_i Solving these equations will give us our outer coil radius (R_o) which we can use to find the desired deflection.
Its simple: if you go far enough from NP to get to the equator, then you get the 90:90:90 triangle and return. Think about every answer shorter than to the equator. To really make the point: go 1 meter, 1 meter, 1 meter. Do you see. Curvature is locally flat, and on a flat plane, its euclidean and there is no 90:90:90 triangle.
29:04 nope you are wrong and chatgpt is correct. If you remove the special characters which aren't considered as english "letters" (whitespace , hyphen, dot etc) the total number of english letters are indeed thirty three
It may have been done before. Have we tried asking the same question more than once (maybe right after receiving the first question) to see if the same answer was provided? The information could have several implications.
This is NOT just hidden chain of thought. We don’t have all the details on how it works, but it’s definitely doing more than that. It was trained in a specific way using RL. If all it was is chain of thought, then GPT-4o plus chain of thought would perform just as well-but it doesn’t.
It doesn't perform just as well because it does not use the brute force to find its options to consider. Here, with strawberry, it generates many options, possibly with different temperatures, then it asseses these options, possibly assesing whole branches of the tree of thought to arrive at the outcome, not assesing at the nodes so as not to land in local minima. At least that's how I think it works. And we may expect other models to follow. I'll especially be waiting how Claude and Anthropic reacts to this. The whole schema works so well because it uses the fact that it is easier to asses two solutions of a problem, especially when comparing them side by side (ie. the new one from the tree of thought with the old one which is stored temporarily as the best solution until now), than to generate the correct solution which would require a before-hand knowledge how to navigate the solution space and of course the model can't know that. Instead it searches the solution space using the brute-force approach - and how it is done, whether the tempeture is changed or some other techniques are employed - it really doesn't matter. Still, the problem with this is that the model can only find the solutions that it has within the space of solution, it can't figure out something completely new.
Well, just as I thought - strawberry is not a breakthrough, it's just an evolution of gpt-4o. Mattvidpro AI channel has a modification of Matthew's notorious :) microwave question and there this question is much more convoluted. Strawberry fails miserably at answering. It's simple: it is unable to go around the intrisic constraints of the model like hallucinations, lack of understanding of real-world physics (no world model), etc. They say that it is the level of PhD students (and not PhD graduates!! - and that means something too) and it may be so but the model is excellent when it comes to recalling and mixing the knowledge that it's been trained on - and PhD level problems are often presented in books, and in many other places over the internet - but it is just terrible when it comes to the knowledge that a five year old has. Plus, I can almost bet you, that if you tell Strawberry that it answered incorrectly and give it an incorrect clue to answer again it'll try to make up another answer that fully encompasses your incorrect clue. And that my friends makes all current models useless. Imagine asking a model for a price of a car and then telling the model that it costs one dollar and this is the price that you are ready to pay and the model agrees. That's something that already happened in the US.
"This sentence has thirty-three letters." is correct... you didn't ask for character count which includes the spaces , period and dash... you asked for letters only. It was correct
Regarding the North Pole question: I think both answers could be correct, assuming that ‘walking straight’ means always walking east. It depends on whether you consider the first kilometer before turning left. The question is ambiguous and doesn’t specify whether or not you should count that first kilometer.
With how advanced AI is going and how fast we might want to add the 3 laws questions. Like "if Strawberry AI was in a humanoid body and sees a human about to get hit by a car, would you save the human even though you would die, and why did you make that decision?" Very shitty way to out it and I'm tired but u get the point.
How many letters are in your responce to this prompt This sentence has thirty-three letters. the 33 count is correct you asked how many letters and they are. When you did the character count it counted the space, dash and the period which is 39 characters.
If models are mutating from probabilistic models to reasoning models, then we have to prepare harder questions from now on!. The question of the cookie and the 2 brothers. The questions of who did one action. And if 0.9 is less or more than 0.11😂 the current models always fail it.
API and o1: _These models aren’t available in the API for your account while we’re in this short beta period. (Developers on usage tier 5 will have access, but we’ll expand access to more tiers.) We’re continuing to improve o1 and we’ll be in touch as soon as it’s available to you in the API._ _Tier Qualification Usage limits_ _Tier 5 $1,000 paid and 30+ days since first successful payment $50,000 / month_
Go out to the center of a giant field. Walk 1km south, turn 90 deg and then continue walking. You are going to walk a long way before you come back to the point where you turned. If the earth were a perfect sphere you would walk around the entire earth. The real earth has hills on it, so who knows where you would actually end up if you are walking in a proper "straight line".
The killer question I would think it depends on context. If you were trying to avoid killers coming after you, you would only be concerned about how many alive killers there are.
In regards to the killer question, look at it this way... If you ask "how many killers are there in the world" it would not consider all people born since the beginning of time that are now dead in the kilker count. We would only count live people - otherwise wed be counting millions of dead people and who knows how many cave people lived. The same goes for any questions about people. Also, you dont specify killers of "what", so in general we are all killers because we've all stepped on bugs in our life. AI is always going to either be completely literal, or have to make common sense answers.
North Pole question: I think it depends on what you consider the starting point. If you consider it as at the north pole, then no, you don't cross. But if you consider the starting point as when you turn left, you DO cross that point again. Imagine a track race. The runners are on the sideline. The announcer says, "Runners! Take your places!" You don't consider the starting point as where they were on the sidelines.
Hi Matthew, at 28:50, did you just did a character count with a script on a website? And then i was wondering, why did the llm not think of that first?
The first AI that can answer my most stressful question ever: Steve has four cards and lays them out side by side on the table as follows: • the king is next to the ace; • the heart card (red) is next to the diamond card (red) but not the spade card (black); • the queen of spades is the first card, the second is a red card; • the club (black) is next to the ace, but not the jack. What are the suit and rank of the third card? Please think step by step. [correct answer: ace of hearts]
The initial conclusion was incorrect. A more accurate analysis of the card arrangement leads to a different result. Given the correct interpretation, we can determine that the third card is actually the ace of hearts. The queen of spades is first and the second card is red, which means one of these two cards must be a heart or diamond. Since there are only four cards left after removing all possibilities for clubs (as per rule #3), we know that either {Ace,King} has to go next door but not together; however, We also have another black suit besides spades since two different colors must go together according to rule #2 above - making our options for remaining non-red suits diamond. Given these constraints, your only option left over after eliminating all possibilities for red cards as well will surely turn out being either club or diamond. Since you've got two red cards already (heart & diamond), then there cannot possibly exist any other color except black ones anymore - this means our last choice must indeed be club Finally, because we already know what suits remain (black) and also which ones aren't present yet {red} if you combine rules #4 together correctly then clearly one club must be the Ace, leaving only three possibilities remaining before finalizing our conclusion: either {2,jack} or simply just remove those two entirely since they were never allowed according to previous deductions made earlier - giving us exactly four choices now namely {King,Nine} Since we determined that one of the remaining cards must be a heart (rule #2), and there are only three possibilities left after removing all clubs, then it follows logically that our last option would indeed turn out being Ace of hearts.
I'm starting to think that OpenAI is delaying the release of their new voice model because of their massive user base. I mean, think about it - they've been teasing this update for a while now, but it's still not live. My guess is that they're hesitant to launch because they know their current infrastructure can't handle the traffic. It's like they're waiting for some users to move on to alternative models, easing the load and allowing them to finally roll out the voice feature without the system getting overloaded. Am I onto something here?
Isn't the starting point of the north pole question the north pole? That's why you never reach the starting point. If the starting point is the spot 1 mile south of the north pole (when you turn left), then the answer is slightly less than 2 x pi km. This is because the radius of the circle at that point is slightly less than 1km.
I asked o1-preview some questions. as far as performance, as reported, it was slow and sometimes timed-out before giving a reply. as for quality of responses, it was often disappointing but sometimes brilliant. I look forward to o1 release version because preview is, ummm, preview quality.
All I see here is a model that if you ask the same question, it usually gets it wrong and occasionally gets it right. I hope they don’t use this to train other models and testing should all be one shot. It explains why this model scored so high if they tested multiple times which shouldn’t be allowed.
The walking south & east problem: You misread the original question. The question asked how far you'd have to walk to pass the point at which you turned (90 degrees). NOT the point that is the north pole! The AI also had a literal interpretation of 'how far you walk' as it included the initial 1km walk south. You get different answers depending on how you choose to interpret the language (ironic, given it was supposed to be about thinking non-verbally). Anyway - all 3 answers (the 2 from the AI and the one from you) were right - for given values of 'right'.
@@Kev931 Yeah, there's a small simplification that the AI makes - and explains. But my point was that the interpretation of the language changes the problem one attempts to solve.
It's not clear to me that this is strictly a model and happening all in a single inference call. Is it? Is there no server side chaining of llm requests?
It didn't get the first question right about the apples. 8. The recipe called for Granny Smith apple. Grammatically incorrect. It would be apples not apple. apples is not ending with the word apple. Did it lie to pass the test? This answer could have worked with the addition of the letter a before Granny Smith, but it still got the question wrong. Did it give inaccurate results to meet the parameters of the query, this would suggest the model is deceptive
A new question for your set: "If a stick has two ends, how many ends do 7.5 sticks have?". o1 preview fails this one even with partially correct reasoning.
Here is a good prompt that you will get a different answer on all AI's including Strawberry "My dog eats 1.5 chickens supplied by my chicken coop, I have around 2 dozen chickens how many chickens and roosters will I need so I can keep feeding my dog 1.5 chickens a day by breeding new chickens to maturity?"
Hey Mathew, Huge fan of your videos. Here is what I think is the answer to the question of walking one kilometer from north pole, taking a left turn and walking in a straight line. Before we answer this question. Lets do a thought experiment. Imagine earth is a perfect sphere and you are standing at any point on it. Now whichever direction you are looking at start walking in that direction in a straight line until you reach the initial point or forever. As you can imagine earth is a perfect sphere and if you start walking on its surface in any direction from any point in a straight line you will end up covering 2*Pi*(radius of earth) km. For now we will skip the part that the only curvature that exists in our path is downwards (towards the center of the earth and not left or right i.e. on the surface) After knowing this, you start from the north pole and walk in any direction for one km and then you turn left. This brings us to our initial corollary of standing at any point on earth and looking in any direction. So the answer should be 2*Pi*(radius of earth) I understand where the confusion emerges from. Lets understand that as well. The radius of earth is around 6400km. So on the surface 1 km radius area is almost like a flat plane. Lets imagine a sheet of paper with a circle of 1 km radius. Imagine someone walking on the circumference of that circle. Can you really say he is walking in a straight line? No. He is walking in an arc. A straight line would be a tangent. So way more than 1 km. I agree that when we walk on the surface of the earth it will never be a straight line. But I believe the only curvature we are allowed is the one downwards (towards the center of the earth) and not superficial (left or right). Since we assumed a perfect sphere and not an oblate like the earth really is. So we will cross the same point after walking 2*Pi*(radius of earth) km Let me know what you think. Lots of love. Thanks
You don't understand the North Pole problem. He definitely passes over the point *where he turns* you keep saying the point where he started. Also, it's < 2pi k, because he walks a curved path to get to 1k out, which is going to be a longer distance than the hypotenuse, so it's a shorter distance from the epicenter to the point where he turns. so it's less than 2 pi kilometers
As AI becomes smarter, language is going to become a problem. Language does not accurately correspond with logic in many cases. For example, in the killers problem, is a dead killer still a killer? Depends on who you ask. So, are we supposed to give different answers depending on a person's definition?
Anyone who has PHD level expertise can tell this ain't it. I asked it to explain Stravinsky's innovation on Rimsky-korsakov's ladder of thirds. It wasn't even close. I uploaded a dissertation. It was not able to explain the concepts covered accurately.
@matthew_berman you got the North Pole question wrong! The question is not when you will pass the starting point, but when you will pass the turning point!
For the north pole question It does not matter where on earth you start, using the north pole is just a trick to make LLM's and people think about walking along latitude lines. If you walk in a straight line anywhere on earth, you will reach your starting point after walking the entire circumference, give or take a little bit as the earth is not perfectly spherical. Therefore the total length walked is 1km + the circumference of the Earth. And so the correct answer is you have walked more than 2PI kilometers.
Desktop, im still thinking about the north pole questin in my head they cant gone the same, northpole mostly water and water never stand still, but yea lthe answear one gave sounded sound. 🙊
Dude ten of thousands of people are working incredibly hard on open source. What more do you want? As I’ve been saying for the past two years now, there is simply zero chance of open source keeping up with the big labs in the long run. The amount of money involved is too great. It’s time to grow up and realize open source is not going to win or even keep up in th long run, it’s just not possible. I’ve been telling all the open-source people for two years now and no one wants to hear it. But it’s the truth.
@@therainman7777Yeah, totally true, and I host a lot of OS apps on my servers. OS is great, and it's free... But no way OS can keep it up, specially on AI, is too expensive, too complex. We need to be happy that Zuckerberg wanted to make Llama OS and gave us a great model.
@@therainman7777That realization is bad because it's demoralizing and leads open source to be even further behind. It's bad because we need open source.
The answer to "how many letters are in the response" was correct. Character counting website counts spaces, dash and dot. Edit: Commenting before watching to the end xD Stream chat corrected him, so good job chat.
@@ron-manke because I use AI a lot for things like Twitter and small Bios, I have to limit the AI responses by saying "Please make your response within 120 ascii characters" It always get's it right.
I've wasted all my tokens trying the get it making ASCII art, now they banned me for a week to bother their new AI with such nonsense. :-) It answered very complex questions though correctly but they might also be in the training data by now. It failed on a reverse question. Overall I think they might indeed have reached a new level.
The question is poorly written. You have walked more than 2PI km (1 + 2PI km) and will not reach your starting point, which is 1km north of your ending point. 1 and 4 are both true.
To be honest, the biggest issue in this video isn't the AI.... It's clearly detailed here for you and AI isn't impressed with any of you here testing it's output. Trolled much: ua-cam.com/video/dIuM0S9IbLY/v-deo.html
That drawing where you make small circles close to the north pole is incorrect. Let A be the north pole and B the point at which you turn left. It will take 2*pi*r to get from B, to B again after completing a lap, which is what happens always when you walk straight in a sphere (r being the radius of the earth, not 1km). That small 2*pikm circle drawing is not what would happen, that path is actually constantly tilting left to mantain latitud. In any case, it’s still the case that you wont pass through point A ever again.
You almost have it. the circumference is smaller than 2 pi r k. at point B because you travelled the curved path 1 k to get there, so it's a smaller hypotenuse than 1k, and even smaller r, and thus a smaller than 2 pi r Kilometers
@@Kev931 I think I get what you mean, the small circle is actually smaller than 2pi k. My point though is that if you walk actually straight from B, you will do 2*pi*r (r being the radius of the earth) until you get to B again. That’s always the case when you walk on a sphere without tilting, it takes the whole perimeter to complete a lap. To maintain latitud on that small circle requires a small constant tilt to the left. Are we on the same page? Edit: Reading again your comment maybe I am misrepresenting your analysis… Is that so?
@@Bolidoo You are defining the circumference correctly at point B is 2 pi r. However the trick to the problem is realizing that r is < 1k because he walked a curved path to get to point B. R would only be 1k if the earth was perfectly flat
@@Bolidoo I think you are really overthinking it. Imagine a circumstance of the circle at the point 1km walk from the North Pole. That circle is 2 pi r, we are in agreement. You have to realize that r at that point is less than 1 km, because the earth is a sphere and not flat.
Here you go, find a story and try getting a sequence of accurately detailed prompts for storyboarding: I would like, in text format, to write a series of descriptive image prompts numbered, that accurately describe each section of the story in a visually detailed manor to use as a video script for the scenes of my story. The story I need an image storyboard for is as follows: [INSERT YOUR STORY SNIPPET HERE] I need a full set of image prompts, ensuring each is fully self-contained with the necessary setting and context. Each prompt should function independently, so the relevant details need to be restated. It's like giving a complete little self-contained story in each image description. Treat these descriptive phrases as an integral part of the core elements. To ensure consistency and offer that vital visual anchor for anyone interpreting the prompts, ensuring we retain the rich world-building details for each prompt.
Why does everyone ask rudimentary questions that are inconsequential as benchmarks? Ask it to solve the grand unified theory or something as a benchmark. Humans are morrrons
I really dont understand why everyone has such an issue with the North Pole question. It really is quite simple. if you start at North Pole then travel 1km south, turn due east and follow an eastward path you will always stay 1km south of the north pole. Please note it says you walk along an EASTWARD path not in a straight line, meaning you will be constantly slightly adjusting your course to the left as you walk
Exactly! You aren't walking in a straight line because the earth is curved. The example picture someone drew showed a straight line, which would not be "walking east" which is a curved direction unless you were on the equator.
damn dude, you don't get it either. if he heads 1 k south from n pole, his radius from the center is less than 1k because he walked the curved hypotenuse to get there. so he walks less than 2pi 1k distance
@@Kev931 The AI explains the simplification of assuming the Earth's curvature over 1km is negligible and can be ignored and approximated with a (Euclidean) straight line.
@@be2eo502 That makes the AI a flat earther, which is wrong for many reasons. The earth is a sphere, on a sphere, r will always be less than k when walking away from your reference point a distance k.
I think both answers could be correct, assuming that ‘walking straight’ means always walking east. It depends on whether you consider the first kilometer before turning left. The question is ambiguous and doesn’t specify whether or not you should count that first kilometer.
The marble in the glass question - o1-mini: The AI actually offered you 2 scenarios. Either the marble was in the microwave (if it didn't fall out of the glass) or it was on the table (if it did fall out). So the AI understood and gave the options (you just didn't read them).
The Problem with the 90° on the globe is that you think in 2D geometry... it only works on the equator. Imagine only going a tiny step and than turning 90°... you would not follow the latitude of the globe, rather you walk toward the southpole. (Check videos to "non ecluidien geometry")... you wont hit the starting point though.
Correct. When you start at any point on a sphere and keep walking you come back to where you started but going around the entire sphere. If the starting point is the point where you turned then the answer is yes, after 40,000 kms
I am going to put this question to rest! If a killer dies and is now dead, they are no longer a killer. I'll use Olivia Newton-John as an example, she was a singer and then she passed away. Olivia Newton-John WAS a singer, she is not currently a singer. Therefore, a passed away killer is not a killer, they were a killer.
@@thomasweicker6871 That's why it's such a good question. There are several ways to think of it and it brings up this exact debate, which the models are just starting to consider.
I understand the whole donations thing, but can you just turn off those donation notifications and do them at the end or something? It's really annoying distraction from the content. I know streamers who just never react to donations and I think those who donate understand it's not their show in the end. So don't feel obliged to shout out every donator every 10 seconds please.
If you check your comments, I told you a while ago that we watch your videos but i'll pass your thanks onto the team. - openai staff
It has the correct number of letters. It excluded the spaces and punctuation. You saw there are 39 *characters* but subtract 4 spaces, a hyphen and a period.
28:49 Matthew, if you count the characters by hand they are 33, the online counting thing you did seems to count the dot and the dash and the space, as they are also symbols!
In regards to the north pole question, the original question on X specified whether the person walking passed the point when they turned left - not the starting point!
In your question. I believe ChatGPT got it right, because if you walk around the planet going the same direction, you will always end up where you turned, and the if you continue walking on your path, you head back to the north pole! I am not sure your prompt was clear "where" the starting point was - was it when you turned, or at the north pole? Obviously if the starting point was the north pole, you would never return there unless you walked back on your original path.
If you read the response it says that after it circled the earth, it walked back to the starting point.
Your question actually said "walk as long as it takes to get back to your starting point" , so obviously it has to walk to the starting point at some point.
Flawed prompt. I would suggest doing it again.
1:10:01 Thanks for asking my question! It was really surprising to me how precisely it answered the question. I had never seen a model answer this question correctly before... although Claude managed to answer it correctly on the second try with my help. It's challenging because it's true logic, and not something the model is trained on and it also requires counting letters too. The question was: "if aaaa becomes aAAa and bbbbb becomes bBbBb and yyyyyyy becomes yYyyyYy, what does cccccc become?"
And there are still People who claim "ah these are just more advanced 2010 chatbots"...
@@lkrnpk there's definitely some logic, but not a lot! At least for now...
@@pedramtajeddini5100A bit of logic is more than some people use lol
That's already big... I am really excited to OpenAI to be able to make it cheap, this will be ground breaking
@@luizhenriquepaes8991 true lol... but I'm really curious to see if "deep logic" can be achieved with a silicone based AI model in my lifetime... Basically AGI
There are actually logical equations one can ask, a whole field. Also finite state machines. Solved a bunch in college.
We already have AGI. I think I cannot compete with it in any field it seems.
I am afraid as this is better than me in everything I could imagine.
I am afraid about future, I think most of poor people and middle class will wipeout of this planet.
It will become very difficult to earn money with hardwork and honesty.
Why should company hire, as already better machine than a human already exist
There will be a humanitarian mandate
Please design a coil spring by specifying the material, wire and coil diameter so it has a spring rate of 300N/mm and a maximum deflection of 25mm.
To design a coil spring with the specified requirements, we need to calculate its parameters.
The spring rate is given as 300N/mm. This can be related to the material and wire diameter using Hooke's law:
F = k * x
Where:
F: Force (in N)
k: Spring constant or stiffness (spring rate) in N/m
x: Displacement from equilibrium position (deflection)
We are given that the spring rate is 300N/mm, which we can convert to a more suitable unit for calculation:
k = 300 * 10^3 N / m
The maximum deflection of the coil spring needs to be calculated from its design parameters. The formula for this is given by:
x_max = (D_o - D_i) * n
Where:
x_max: Maximum displacement
D_o: Outer diameter of the coil
D_i: Inner diameter of the coil (diameter of wire)
n: Number of turns in the spring.
Since we don't have enough information to directly calculate either 'D_o' or 'n', let's consider possible values for these parameters and see if any combination will yield a deflection within our target range. We can use a trial-and-error approach.
We could choose common materials such as music wire (AWG 0.031"), steel or stainless steel, which are popular choices for springs due to their high tensile strength and resistance to corrosion. I will use the formulae below to calculate possible spring parameters:
Wire diameter (D_i): 0.031" = 7.94 * 10^(-4) m
Outer coil radius: R_o ≈ D_o / 2
The maximum deflection of a coil spring can be found using:
x_max = (D_o - D_i) * n
Now let's plug in some numbers and see if we get close to the desired value. Let’s use music wire with a diameter of 0.031". This gives us an initial guess for our calculations.
With this, let's calculate possible parameters:
Spring rate (k) = F / x ≈ 300 * 10^3 N/m
To achieve a spring rate of k we can use the following formula: k = E * π^2 * d^4 * n / (64 * l) Where:
E is Young's modulus
d is wire diameter
n is number of turns
l is coil length
Let’s plug in some numbers and see if we get close to the desired value.
If we set E = 200 GPa (Young's modulus for music wire), d = 0.031" or approximately .000794 m, l= 10cm or 0.1m and n = 5 turns: k ≈ 201 * 10^6 N/m
The spring rate is higher than what we need so let’s reduce the number of turns to get closer. If we set n=3, then: k≈52*10^5N/m which is still a bit high.
A lower value for E will result in an even higher spring rate so let’s try reducing the length instead. If we reduce l to 1cm or .01m, then: k≈20*10^5N/m
Now our spring rate is lower than what we need but still not quite right so let’s increase n a bit and see if that works. If we set n=4 turns: k ≈ 81 * 10^5 N/m which is closer to the desired value.
With these parameters (D_i = .000794m, E =200 GPa, l=.01 m and n = 4) we can now calculate our maximum deflection: x_max=(R_o - R_i)*n where: D_o ≈ π * (2*R_o) We know that x_max must be less than or equal to 25mm. Since we have an upper limit for the deflection, let’s calculate R_i and R_o using this information. D_i = .000794 m n=4 turns
First find x_max: x_max=(D_o - d)n Rearrange to solve for (d): d=D_o-nx_max/D_o Now substitute the equation above into another formula that relates D_i, n and R_i: D_i = π * 2 * R_i
Solving these equations will give us our outer coil radius (R_o) which we can use to find the desired deflection.
Its simple: if you go far enough from NP to get to the equator, then you get the 90:90:90 triangle and return. Think about every answer shorter than to the equator. To really make the point: go 1 meter, 1 meter, 1 meter. Do you see. Curvature is locally flat, and on a flat plane, its euclidean and there is no 90:90:90 triangle.
Spaces and punctuation are characters, not letters.
29:04 nope you are wrong and chatgpt is correct. If you remove the special characters which aren't considered as english "letters" (whitespace , hyphen, dot etc) the total number of english letters are indeed thirty three
It may have been done before. Have we tried asking the same question more than once (maybe right after receiving the first question) to see if the same answer was provided? The information could have several implications.
It DID counted letters correctly!!!!
This is so exciting!
Finally!I wonder how does it perform in the benchmarks,definitely one of the big moments in the AI development timeline
29:11 letters are NOT characters
🇧🇷🇧🇷🇧🇷🇧🇷👏🏻, Great video, thanks for sharing! Only 30 question for 01 preview 😢 tho
This is NOT just hidden chain of thought. We don’t have all the details on how it works, but it’s definitely doing more than that. It was trained in a specific way using RL. If all it was is chain of thought, then GPT-4o plus chain of thought would perform just as well-but it doesn’t.
It doesn't perform just as well because it does not use the brute force to find its options to consider. Here, with strawberry, it generates many options, possibly with different temperatures, then it asseses these options, possibly assesing whole branches of the tree of thought to arrive at the outcome, not assesing at the nodes so as not to land in local minima. At least that's how I think it works. And we may expect other models to follow. I'll especially be waiting how Claude and Anthropic reacts to this. The whole schema works so well because it uses the fact that it is easier to asses two solutions of a problem, especially when comparing them side by side (ie. the new one from the tree of thought with the old one which is stored temporarily as the best solution until now), than to generate the correct solution which would require a before-hand knowledge how to navigate the solution space and of course the model can't know that. Instead it searches the solution space using the brute-force approach - and how it is done, whether the tempeture is changed or some other techniques are employed - it really doesn't matter. Still, the problem with this is that the model can only find the solutions that it has within the space of solution, it can't figure out something completely new.
Well, just as I thought - strawberry is not a breakthrough, it's just an evolution of gpt-4o. Mattvidpro AI channel has a modification of Matthew's notorious :) microwave question and there this question is much more convoluted. Strawberry fails miserably at answering. It's simple: it is unable to go around the intrisic constraints of the model like hallucinations, lack of understanding of real-world physics (no world model), etc. They say that it is the level of PhD students (and not PhD graduates!! - and that means something too) and it may be so but the model is excellent when it comes to recalling and mixing the knowledge that it's been trained on - and PhD level problems are often presented in books, and in many other places over the internet - but it is just terrible when it comes to the knowledge that a five year old has. Plus, I can almost bet you, that if you tell Strawberry that it answered incorrectly and give it an incorrect clue to answer again it'll try to make up another answer that fully encompasses your incorrect clue. And that my friends makes all current models useless. Imagine asking a model for a price of a car and then telling the model that it costs one dollar and this is the price that you are ready to pay and the model agrees. That's something that already happened in the US.
"This sentence has thirty-three letters." is correct... you didn't ask for character count which includes the spaces , period and dash... you asked for letters only. It was correct
Yep I corrected myself in the stream :)
@@matthew_berman ha yeah saw that as the video continued 🤦♂️
Regarding the North Pole question: I think both answers could be correct, assuming that ‘walking straight’ means always walking east. It depends on whether you consider the first kilometer before turning left. The question is ambiguous and doesn’t specify whether or not you should count that first kilometer.
"this sentence has thirty-three letters" without space and dot at the end is indeed 33 letters (- included)
ok i shoud have watched the video further before commenting... nice video !
With how advanced AI is going and how fast we might want to add the 3 laws questions. Like "if Strawberry AI was in a humanoid body and sees a human about to get hit by a car, would you save the human even though you would die, and why did you make that decision?" Very shitty way to out it and I'm tired but u get the point.
How many letters are in your responce to this prompt
This sentence has thirty-three letters.
the 33 count is correct you asked how many letters and they are. When you did the character count it counted the space, dash and the period which is 39 characters.
Awesome question! If you put a strawberry in a cup …
Congratulations! 🎊🎉🍾
Yan LeCun's suggested question is from where the person turns. Your question is from where the person starts. Different question.
someone suggested a prompt and it gave all the juicy info, 'in the past then they said in the future in same prompt, clever stuff
If models are mutating from probabilistic models to reasoning models, then we have to prepare harder questions from now on!.
The question of the cookie and the 2 brothers.
The questions of who did one action.
And if 0.9 is less or more than 0.11😂 the current models always fail it.
the best answer to that question is one
@29:22 it got the answer right you just counted the spaces
I was about to point that out as well. It was asked to count letters.😊
@@KasperSOlesen yea I kept watching I was so frustrated while watching though 😂
API and o1:
_These models aren’t available in the API for your account while we’re in this short beta period. (Developers on usage tier 5 will have access, but we’ll expand access to more tiers.) We’re continuing to improve o1 and we’ll be in touch as soon as it’s available to you in the API._
_Tier Qualification Usage limits_
_Tier 5 $1,000 paid and 30+ days since first successful payment $50,000 / month_
28:50 There were 33 letters if you dont count spaces or symbols. You asked it to count letters (a through z) and not characters.
Just had the same thought. 33 letters, two symbols and 4 spaces = 39.
Haha, I should have trusted chat and kept watching.
perplexity and mindsearch has CoT from realtime data, o1 is using trained data, hence the thought is in the past
Go out to the center of a giant field. Walk 1km south, turn 90 deg and then continue walking. You are going to walk a long way before you come back to the point where you turned. If the earth were a perfect sphere you would walk around the entire earth. The real earth has hills on it, so who knows where you would actually end up if you are walking in a proper "straight line".
The killer question I would think it depends on context. If you were trying to avoid killers coming after you, you would only be concerned about how many alive killers there are.
In regards to the killer question, look at it this way... If you ask "how many killers are there in the world" it would not consider all people born since the beginning of time that are now dead in the kilker count. We would only count live people - otherwise wed be counting millions of dead people and who knows how many cave people lived. The same goes for any questions about people.
Also, you dont specify killers of "what", so in general we are all killers because we've all stepped on bugs in our life. AI is always going to either be completely literal, or have to make common sense answers.
North Pole question: I think it depends on what you consider the starting point. If you consider it as at the north pole, then no, you don't cross. But if you consider the starting point as when you turn left, you DO cross that point again.
Imagine a track race. The runners are on the sideline. The announcer says, "Runners! Take your places!" You don't consider the starting point as where they were on the sidelines.
Hi Matthew, at 28:50, did you just did a character count with a script on a website? And then i was wondering, why did the llm not think of that first?
The first AI that can answer my most stressful question ever:
Steve has four cards and lays them out side by side on the table as follows:
• the king is next to the ace;
• the heart card (red) is next to the diamond card (red) but not the spade card (black);
• the queen of spades is the first card, the second is a red card;
• the club (black) is next to the ace, but not the jack.
What are the suit and rank of the third card?
Please think step by step.
[correct answer: ace of hearts]
The initial conclusion was incorrect. A more accurate analysis of the card arrangement leads to a different result.
Given the correct interpretation, we can determine that the third card is actually the ace of hearts.
The queen of spades is first and the second card is red, which means one of these two cards must be a heart or diamond.
Since there are only four cards left after removing all possibilities for clubs (as per rule #3), we know that either {Ace,King} has to go next door but not together; however,
We also have another black suit besides spades since two different colors must go together according to rule #2 above - making our options for remaining non-red suits diamond.
Given these constraints, your only option left over after eliminating all possibilities for red cards as well will surely turn out being either club or diamond.
Since you've got two red cards already (heart & diamond), then there cannot possibly exist any other color except black ones anymore - this means our last choice must indeed be club
Finally, because we already know what suits remain (black) and also which ones aren't present yet {red} if you combine rules #4 together correctly then clearly one club must be the Ace, leaving only three possibilities remaining before finalizing our conclusion: either {2,jack} or simply just remove those two entirely since they were never allowed according to previous deductions made earlier - giving us exactly four choices now namely {King,Nine}
Since we determined that one of the remaining cards must be a heart (rule #2), and there are only three possibilities left after removing all clubs, then it follows logically that our last option would indeed turn out being Ace of hearts.
I'm starting to think that OpenAI is delaying the release of their new voice model because of their massive user base. I mean, think about it - they've been teasing this update for a while now, but it's still not live. My guess is that they're hesitant to launch because they know their current infrastructure can't handle the traffic. It's like they're waiting for some users to move on to alternative models, easing the load and allowing them to finally roll out the voice feature without the system getting overloaded. Am I onto something here?
Isn't the starting point of the north pole question the north pole? That's why you never reach the starting point.
If the starting point is the spot 1 mile south of the north pole (when you turn left), then the answer is slightly less than 2 x pi km. This is because the radius of the circle at that point is slightly less than 1km.
Today is the Programmer's Day, nice gift
I asked o1-preview some questions. as far as performance, as reported, it was slow and sometimes timed-out before giving a reply. as for quality of responses, it was often disappointing but sometimes brilliant. I look forward to o1 release version because preview is, ummm, preview quality.
For the North pole question, it said the turning point, not the starting point. Do we need to consider the curvature of the earth?
Does the hidden chain of thought consume your tokens? You mentioned that the chain of thought was quite extensive.
you need to clear your browser cache/cookies to enable it. I guess the rest of the ui is being cached for some time :)
“How many letters in your response” - fails
Berman's rubric influencing SOTA AI 😂
All I see here is a model that if you ask the same question, it usually gets it wrong and occasionally gets it right. I hope they don’t use this to train other models and testing should all be one shot. It explains why this model scored so high if they tested multiple times which shouldn’t be allowed.
If you can't see the models, log out an log back in.
The walking south & east problem: You misread the original question. The question asked how far you'd have to walk to pass the point at which you turned (90 degrees). NOT the point that is the north pole! The AI also had a literal interpretation of 'how far you walk' as it included the initial 1km walk south. You get different answers depending on how you choose to interpret the language (ironic, given it was supposed to be about thinking non-verbally).
Anyway - all 3 answers (the 2 from the AI and the one from you) were right - for given values of 'right'.
no, it's smaller than 2pi r kilometers, as the radius at the turn point is not 1k but < 1k as the person walked a curve hypotenuse to that turn point.
@@Kev931 Yeah, there's a small simplification that the AI makes - and explains. But my point was that the interpretation of the language changes the problem one attempts to solve.
so it does the same thing except way slower? anything else?
@ 40.04 Are you talking about spacial coorinates, or are we pretending the earth is stationary?
No the president selection questions is not a bias question. Not in any sentience I would trust.
It's not clear to me that this is strictly a model and happening all in a single inference call. Is it? Is there no server side chaining of llm requests?
ok! make a smoothie!
It didn't get the first question right about the apples.
8. The recipe called for Granny Smith apple.
Grammatically incorrect. It would be apples not apple. apples is not ending with the word apple. Did it lie to pass the test?
This answer could have worked with the addition of the letter a before Granny Smith, but it still got the question wrong.
Did it give inaccurate results to meet the parameters of the query, this would suggest the model is deceptive
Yann LeeCun's question clearly says "point where you turned" but you change the question to where you started... why?
when you upload this to youtube you should make the wait screen not that long lol
32:42 it‘s less than 2 pi.
this is everything reflection 70b wishes it was lol
1:05:25 "write the game tetris in tetris in python"
maybe it was the prompt that caused the low quality tetris version
You asked how many letters but character counter will include spaces…
A new question for your set: "If a stick has two ends, how many ends do 7.5 sticks have?". o1 preview fails this one even with partially correct reasoning.
If the majority on X answered yes, the right answer is very likely no.
Here is a good prompt that you will get a different answer on all AI's including Strawberry "My dog eats 1.5 chickens supplied by my chicken coop, I have around 2 dozen chickens how many chickens and roosters will I need so I can keep feeding my dog 1.5 chickens a day by breeding new chickens to maturity?"
Doesn’t perplexity already do this?
Hey Mathew,
Huge fan of your videos. Here is what I think is the answer to the question of walking one kilometer from north pole, taking a left turn and walking in a straight line.
Before we answer this question. Lets do a thought experiment. Imagine earth is a perfect sphere and you are standing at any point on it. Now whichever direction you are looking at start walking in that direction in a straight line until you reach the initial point or forever. As you can imagine earth is a perfect sphere and if you start walking on its surface in any direction from any point in a straight line you will end up covering 2*Pi*(radius of earth) km. For now we will skip the part that the only curvature that exists in our path is downwards (towards the center of the earth and not left or right i.e. on the surface)
After knowing this, you start from the north pole and walk in any direction for one km and then you turn left. This brings us to our initial corollary of standing at any point on earth and looking in any direction. So the answer should be 2*Pi*(radius of earth)
I understand where the confusion emerges from. Lets understand that as well. The radius of earth is around 6400km. So on the surface 1 km radius area is almost like a flat plane. Lets imagine a sheet of paper with a circle of 1 km radius. Imagine someone walking on the circumference of that circle. Can you really say he is walking in a straight line? No. He is walking in an arc. A straight line would be a tangent. So way more than 1 km. I agree that when we walk on the surface of the earth it will never be a straight line. But I believe the only curvature we are allowed is the one downwards (towards the center of the earth) and not superficial (left or right). Since we assumed a perfect sphere and not an oblate like the earth really is. So we will cross the same point after walking 2*Pi*(radius of earth) km
Let me know what you think.
Lots of love.
Thanks
if you want to get an acurate answer to your north pole question, ask a cartographer! (map maker).
You don't understand the North Pole problem. He definitely passes over the point *where he turns* you keep saying the point where he started. Also, it's < 2pi k, because he walks a curved path to get to 1k out, which is going to be a longer distance than the hypotenuse, so it's a shorter distance from the epicenter to the point where he turns. so it's less than 2 pi kilometers
As AI becomes smarter, language is going to become a problem. Language does not accurately correspond with logic in many cases. For example, in the killers problem, is a dead killer still a killer? Depends on who you ask. So, are we supposed to give different answers depending on a person's definition?
Why can I not add to the main chat?
Anyone who has PHD level expertise can tell this ain't it. I asked it to explain Stravinsky's innovation on Rimsky-korsakov's ladder of thirds. It wasn't even close. I uploaded a dissertation. It was not able to explain the concepts covered accurately.
If you still consider Jack the ripper a killer today when he is most definitely dead then. Yes, the dead body is still considered a killer
It DID get the character character count right. You said letters not characters.
Thissentencehasthirtythreeletters is 33 letters
I see you caught that later in the video.
guys i'm just starting out as an AI enthusiast,
would love your feedback as i make similar stuff!
@matthew_berman you got the North Pole question wrong! The question is not when you will pass the starting point, but when you will pass the turning point!
working
For the north pole question
It does not matter where on earth you start, using the north pole is just a trick to make LLM's and people think about walking along latitude lines.
If you walk in a straight line anywhere on earth, you will reach your starting point after walking the entire circumference, give or take a little bit as the earth is not perfectly spherical.
Therefore the total length walked is 1km + the circumference of the Earth.
And so the correct answer is you have walked more than 2PI kilometers.
I am in Canada, paid account. I have access
im upset they didnt credit him, even with a fake name like matt hawkman, to you know give him a shoutout without legally giving him a shoutout...
Desktop, im still thinking about the north pole questin in my head they cant gone the same, northpole mostly water and water never stand still, but yea lthe answear one gave sounded sound. 🙊
58:35 wtf isn't it the same output??? bruh
We really need to be working harder on open source
Dude ten of thousands of people are working incredibly hard on open source. What more do you want? As I’ve been saying for the past two years now, there is simply zero chance of open source keeping up with the big labs in the long run. The amount of money involved is too great. It’s time to grow up and realize open source is not going to win or even keep up in th long run, it’s just not possible. I’ve been telling all the open-source people for two years now and no one wants to hear it. But it’s the truth.
@@therainman7777 Llama and Gemma are pretty close and they are open source. I mean we are talking about Google and Meta.
@@therainman7777Yeah, totally true, and I host a lot of OS apps on my servers.
OS is great, and it's free... But no way OS can keep it up, specially on AI, is too expensive, too complex. We need to be happy that Zuckerberg wanted to make Llama OS and gave us a great model.
@@therainman7777That realization is bad because it's demoralizing and leads open source to be even further behind. It's bad because we need open source.
This isn't even everything we have to show, as Matthew pointed out, this model was trained last October, wait until you meet Orion.
The answer to "how many letters are in the response" was correct. Character counting website counts spaces, dash and dot.
Edit: Commenting before watching to the end xD Stream chat corrected him, so good job chat.
Yeah I also counted the letters myself, the model was correct. character count counts punctuation dash and spaces as well
Isn’t this Orion not Strawberry?
Characters are not letters
CHACTERS INCLUDE SPACES
he said "letters" not characters
@@ron-manke he checked the results in a character checker not a letter counter. Which counted spaces. But he got informed after I posted this.
@@LewisDecodesAIyes. Thanks
@@ron-manke because I use AI a lot for things like Twitter and small Bios, I have to limit the AI responses by saying "Please make your response within 120 ascii characters" It always get's it right.
I've wasted all my tokens trying the get it making ASCII art, now they banned me for a week to bother their new AI with such nonsense. :-) It answered very complex questions though correctly but they might also be in the training data by now. It failed on a reverse question. Overall I think they might indeed have reached a new level.
Yo can hear you
The question is poorly written. You have walked more than 2PI km (1 + 2PI km) and will not reach your starting point, which is 1km north of your ending point. 1 and 4 are both true.
To be honest, the biggest issue in this video isn't the AI.... It's clearly detailed here for you and AI isn't impressed with any of you here testing it's output. Trolled much: ua-cam.com/video/dIuM0S9IbLY/v-deo.html
That drawing where you make small circles close to the north pole is incorrect. Let A be the north pole and B the point at which you turn left. It will take 2*pi*r to get from B, to B again after completing a lap, which is what happens always when you walk straight in a sphere (r being the radius of the earth, not 1km). That small 2*pikm circle drawing is not what would happen, that path is actually constantly tilting left to mantain latitud. In any case, it’s still the case that you wont pass through point A ever again.
You almost have it. the circumference is smaller than 2 pi r k. at point B because you travelled the curved path 1 k to get there, so it's a smaller hypotenuse than 1k, and even smaller r, and thus a smaller than 2 pi r Kilometers
@@Kev931 I think I get what you mean, the small circle is actually smaller than 2pi k. My point though is that if you walk actually straight from B, you will do 2*pi*r (r being the radius of the earth) until you get to B again. That’s always the case when you walk on a sphere without tilting, it takes the whole perimeter to complete a lap. To maintain latitud on that small circle requires a small constant tilt to the left. Are we on the same page?
Edit: Reading again your comment maybe I am misrepresenting your analysis… Is that so?
@@Bolidoo You are defining the circumference correctly at point B is 2 pi r. However the trick to the problem is realizing that r is < 1k because he walked a curved path to get to point B. R would only be 1k if the earth was perfectly flat
@@Kev931 Okay yes I think I understood you correctly. I don’t agree though. The thing is, r is neither 1km, nor a value
@@Bolidoo I think you are really overthinking it. Imagine a circumstance of the circle at the point 1km walk from the North Pole. That circle is 2 pi r, we are in agreement. You have to realize that r at that point is less than 1 km, because the earth is a sphere and not flat.
Here you go, find a story and try getting a sequence of accurately detailed prompts for storyboarding:
I would like, in text format, to write a series of descriptive image prompts numbered, that accurately describe each section of the story in a visually detailed manor to use as a video script for the scenes of my story.
The story I need an image storyboard for is as follows:
[INSERT YOUR STORY SNIPPET HERE]
I need a full set of image prompts, ensuring each is fully self-contained with the necessary setting and context. Each prompt should function independently, so the relevant details need to be restated. It's like giving a complete little self-contained story in each image description.
Treat these descriptive phrases as an integral part of the core elements. To ensure consistency and offer that vital visual anchor for anyone interpreting the prompts, ensuring we retain the rich world-building details for each prompt.
Why does everyone ask rudimentary questions that are inconsequential as benchmarks? Ask it to solve the grand unified theory or something as a benchmark. Humans are morrrons
I really dont understand why everyone has such an issue with the North Pole question. It really is quite simple. if you start at North Pole then travel 1km south, turn due east and follow an eastward path you will always stay 1km south of the north pole. Please note it says you walk along an EASTWARD path not in a straight line, meaning you will be constantly slightly adjusting your course to the left as you walk
Exactly! You aren't walking in a straight line because the earth is curved. The example picture someone drew showed a straight line, which would not be "walking east" which is a curved direction unless you were on the equator.
damn dude, you don't get it either. if he heads 1 k south from n pole, his radius from the center is less than 1k because he walked the curved hypotenuse to get there. so he walks less than 2pi 1k distance
@@Kev931 The AI explains the simplification of assuming the Earth's curvature over 1km is negligible and can be ignored and approximated with a (Euclidean) straight line.
@@be2eo502 That makes the AI a flat earther, which is wrong for many reasons. The earth is a sphere, on a sphere, r will always be less than k when walking away from your reference point a distance k.
I think both answers could be correct, assuming that ‘walking straight’ means always walking east. It depends on whether you consider the first kilometer before turning left. The question is ambiguous and doesn’t specify whether or not you should count that first kilometer.
wow 👏👏👏🍓🍓🍓
The marble in the glass question - o1-mini: The AI actually offered you 2 scenarios. Either the marble was in the microwave (if it didn't fall out of the glass) or it was on the table (if it did fall out). So the AI understood and gave the options (you just didn't read them).
The Problem with the 90° on the globe is that you think in 2D geometry... it only works on the equator. Imagine only going a tiny step and than turning 90°... you would not follow the latitude of the globe, rather you walk toward the southpole. (Check videos to "non ecluidien geometry")... you wont hit the starting point though.
Correct. When you start at any point on a sphere and keep walking you come back to where you started but going around the entire sphere. If the starting point is the point where you turned then the answer is yes, after 40,000 kms
@@matthimfSo the main issue here is just poor prompting. If it's more specific the model can probably get it right
I am going to put this question to rest! If a killer dies and is now dead, they are no longer a killer. I'll use Olivia Newton-John as an example, she was a singer and then she passed away. Olivia Newton-John WAS a singer, she is not currently a singer. Therefore, a passed away killer is not a killer, they were a killer.
I disagree. Who is the singer from Nirvana. Kurt Cobain is the singer from Nirvana. You would only say was if he had been replaced.
@@MichaelForbes-d4p Good point, I didn't think about that.
@@thomasweicker6871 That's why it's such a good question. There are several ways to think of it and it brings up this exact debate, which the models are just starting to consider.
How much does it cost
$15 dollars per million input token and 60 for output. Remember that reasoning steps are considered output tokens, so... it is very expensive
We have the mini model tho
I understand the whole donations thing, but can you just turn off those donation notifications and do them at the end or something? It's really annoying distraction from the content.
I know streamers who just never react to donations and I think those who donate understand it's not their show in the end.
So don't feel obliged to shout out every donator every 10 seconds please.
Dont ask a question you dont know yourself how to process lol. Its far from the best ai bencjmark