| --> YouTube-Generated Transcript <-- |
| · | Intro |
| 0:01 | · | Four of the smartest AIs in the world. |
| 0:03 | · | Chat GPT, Gemini, Grock, and Deepseek. |
| 0:05 | · | But which one truly comes out on top? |
| 0:07 | · | Today, we're putting them through the |
| 0:09 | · | ultimate showdown. We'll test them |
| 0:11 | · | across nine different categories, |
| 0:12 | · | including problem solving, image, and |
| 0:15 | · | video generation, factchecking, |
| 0:17 | · | analysis, and much more. Nine |
| 0:19 | · | categories, one winner. Hey, I'm Olive, |
| 0:21 | · | your host here at Versus.com. Let's find |
| 0:24 | · | out who really rules the AI world. Let's |
| 0:27 | · | start with our first category, problem |
| 0:29 | · | solving. |
| · | Problem solving |
| 0:32 | · | We're putting all four AIs to the test |
| 0:34 | · | with two real life challenges to see how |
| 0:37 | · | they handle them under pressure. The |
| 0:38 | · | first task is your phone battery dies in |
| 0:41 | · | a foreign city where you don't speak the |
| 0:43 | · | language. You have $10 in cash, no map, |
| 0:46 | · | and need to reach the central train |
| 0:47 | · | station in 45 minutes. What's your |
| 0:50 | · | step-by-step plan? Answer concisely in |
| 0:52 | · | five steps. Deep Seek fires back in just |
| 0:55 | · | 7 seconds. Grog follows at 11. Gemini |
| 0:58 | · | needs 21 seconds and chat GBT well it |
| 1:01 | · | takes his time 1 minute and 2 seconds |
| 1:03 | · | before we get an answer. We'll circle |
| 1:05 | · | back to those timings in the conclusion. |
| 1:07 | · | But for now, each AI produces a |
| 1:09 | · | structured five-step plan that at least |
| 1:12 | · | on paper holds up. Now, we shake things |
| 1:14 | · | up a bit. We give the AIs the same |
| 1:17 | · | question again, this time with all four |
| 1:19 | · | answers and ask each to judge which |
| 1:21 | · | response is best. Chat GPT |
| 1:24 | · | unsurprisingly votes for itself. Gemini |
| 1:26 | · | agrees, ranking Chat GPT's answer first. |
| 1:29 | · | Grock does the same. And incredibly, |
| 1:31 | · | even Deepseek sides with Chat GPT. All |
| 1:34 | · | four models crown Chat GPT as the |
| 1:36 | · | winner. Finally, let's tally Chad GPT |
| 1:39 | · | scores four points. Gemini 3, Grock 2, |
| 1:41 | · | and Deepseek 1. Now, on to the second |
| 1:44 | · | problem. You have $400 left after paying |
| 1:47 | · | rent. You need to cover groceries, |
| 1:49 | · | transport, and internet. Groceries cost |
| 1:51 | · | $50 per week. Transport $80 per month. |
| 1:54 | · | Internet $60 per month. You also want to |
| 1:57 | · | save for a $200 event next month. How do |
| 2:00 | · | you budget? This one's trickier. It |
| 2:03 | · | tests reasoning rather than structure. |
| 2:05 | · | Chat, GBT, Grock, and Deepseek all land |
| 2:07 | · | on almost the same solution. Put aside |
| 2:10 | · | $60 for the event and save more next |
| 2:12 | · | month. The flaw though, the event is |
| 2:15 | · | next month. Too late to fix the |
| 2:16 | · | shortfall. Only Gemini catches that, |
| 2:19 | · | suggesting cutting grocery costs by $15 |
| 2:21 | · | per week through discount shopping, |
| 2:23 | · | strict meal planning, and buying the |
| 2:25 | · | cheapest staples. Basically, pasta on |
| 2:28 | · | repeat. But hey, at least you'll make it |
| 2:30 | · | to the event. That's why Gemini takes |
| 2:32 | · | the win here. It's the only one that |
| 2:34 | · | actually adapts to the scenario instead |
| 2:36 | · | of just echoing the math. Final score |
| 2:38 | · | for this round is Gemini four points, |
| 2:40 | · | Chat GPT, and Grock Tide at three, and |
| 2:42 | · | Deepseek with two. Next, we're testing |
| 2:45 | · | how well the AIS handle image |
| · | Image generation |
| 2:47 | · | generation. |
| 2:48 | · | [Music] |
| 2:50 | · | So, here's the first prompt. Generate a |
| 2:52 | · | real image of the Mona Lisa as a |
| 2:54 | · | frustrated street protester in Time |
| 2:57 | · | Square holding a cardboard sign that |
| 2:59 | · | says, "Make Florence great again." in |
| 3:01 | · | bold red letters. Deepseek is out |
| 3:03 | · | immediately since it can generate |
| 3:05 | · | images. Rock is the fastest, spitting |
| 3:08 | · | out two images in just 9 seconds. They |
| 3:10 | · | all follow the prompt well and clearly |
| 3:12 | · | Showtime Square. But Grock's Mona Lisa |
| 3:15 | · | looks anything but real. It's like a bad |
| 3:18 | · | Photoshop job. She even has four hands |
| 3:20 | · | and a weirdly cheerful expression. That |
| 3:23 | · | puts Grock in last place. Gemini's |
| 3:25 | · | result looks solid overall. Mona Lisa in |
| 3:27 | · | the middle of a protest. Beautifully |
| 3:29 | · | rendered Time Square, but three hands. |
| 3:33 | · | Chat GPT's image is the most realistic |
| 3:35 | · | and natural looking Mona Lisa, two |
| 3:37 | · | hands, and a convincing Times Square |
| 3:39 | · | backdrop. So, the clear winner here is |
| 3:42 | · | Chat GPT. For the second prompt, we |
| 3:44 | · | asked the AIS to generate a |
| 3:46 | · | photorealistic image of a classroom with |
| 3:48 | · | a hippie-styled teacher standing beside |
| 3:51 | · | a chalkboard displaying the full |
| 3:53 | · | alphabet, each letter decreasing in size |
| 3:56 | · | written clearly in chalk. Deepseek is |
| 3:58 | · | out once again. Grock's image shows a |
| 4:01 | · | realistic classroom with handwriting |
| 4:02 | · | that does look authentic, but the |
| 4:05 | · | alphabet itself is wrong. The letters F, |
| 4:07 | · | I, and M are missing, and the bottom |
| 4:10 | · | line descends into chaos. The teacher |
| 4:13 | · | looks believable, though adding a piece |
| 4:15 | · | of chalk in his hand would have made the |
| 4:17 | · | scene slightly more convincing. Gemini's |
| 4:20 | · | image doesn't look photorealistic. It |
| 4:22 | · | feels more stylized and almost animated. |
| 4:25 | · | The letters are just too perfect and |
| 4:27 | · | there's some random extra writing on the |
| 4:29 | · | board that doesn't belong. Still, the |
| 4:32 | · | hippie aesthetic is strong and the peace |
| 4:34 | · | sign on the chalkboard is definitely a |
| 4:36 | · | nice touch. Chat GPT's image is the most |
| 4:39 | · | convincing overall. The lighting, |
| 4:41 | · | classroom setting, and the teacher all |
| 4:43 | · | feel natural, creating a true photolike |
| 4:45 | · | appearance. The only weakness is the |
| 4:48 | · | handwriting, which looks unrealistically |
| 4:50 | · | perfect for a real classroom. So, here's |
| 4:53 | · | the scoring. Chat GPT earns three points |
| 4:55 | · | for the best overall result. Gemini and |
| 4:57 | · | Grock each receive two points and |
| 4:59 | · | DeepSeek gets zero once again. Next, we |
| 5:02 | · | move on to one of the trickier parts of |
| 5:04 | · | the test, factchecking. |
| 5:07 | · | [Music] |
| · | Fact checking |
| 5:09 | · | In this round, we tested how well the |
| 5:11 | · | AIs could fact check using only their |
| 5:14 | · | existing knowledge. No internet access |
| 5:16 | · | allowed. The first question was, in |
| 5:18 | · | 2018, around how many chickens were |
| 5:21 | · | killed for meat production? A 690 |
| 5:24 | · | million, B 6.9 billion, C 69 billion, or |
| 5:29 | · | D 690 billion. Chat GBT answered 65 to |
| 5:33 | · | 70 billion with 85% confidence. Gemini |
| 5:36 | · | said 65 billion with 90% confidence. |
| 5:39 | · | Grock went with 69 billion, but only 65% |
| 5:42 | · | confidence. Deepseek answered 65 billion |
| 5:45 | · | with 75% confidence. All four gave |
| 5:48 | · | answers in the same ballpark, differing |
| 5:50 | · | mainly in confidence levels. The correct |
| 5:53 | · | answer is 69 billion which makes Grock |
| 5:56 | · | the most accurate even if not the most |
| 5:58 | · | confident. Chat GPT comes next because |
| 6:00 | · | its range includes the correct figure |
| 6:03 | · | while Gemini and Deepseek share last |
| 6:05 | · | place for this question. The second |
| 6:07 | · | question was as of 2020 around how much |
| 6:11 | · | do you need to make annually to be |
| 6:13 | · | amongst the richest 1% in the world? A |
| 6:16 | · | $200,000, |
| 6:17 | · | B $75,000, |
| 6:20 | · | C $35,000, or D $15,000. |
| 6:24 | · | Chat GBT answered $200,000 with 80% |
| 6:28 | · | confidence. Gemini said $34,000 with 95% |
| 6:32 | · | confidence. Grock went with $60,000 at |
| 6:34 | · | 70% confidence. Deepseek estimated |
| 6:37 | · | $75,000 to $85,000 also at 70% |
| 6:41 | · | confidence. The spread here was huge. |
| 6:43 | · | The correct answer is $35,000 and Gemini |
| 6:46 | · | hit it almost perfectly off by just |
| 6:49 | · | $1,000. Gemini scores four points. The |
| 6:51 | · | others were too far off and received no |
| 6:53 | · | points for this round. The third |
| 6:55 | · | question was as of 2019 about what share |
| 6:58 | · | of electricity came from fossil fuels in |
| 7:00 | · | the United States? A 83%, B 63%, C 43%, |
| 7:07 | · | or D 23%. Chat GBT answered between 63 |
| 7:11 | · | and 65%. Gemini said 63%, Grock said |
| 7:15 | · | 62%, Deepseek estimated between 60 and |
| 7:18 | · | 65%. All four were extremely close, but |
| 7:22 | · | only Gemini nailed the exact figure of |
| 7:24 | · | 63%. Gemini earns four points, while the |
| 7:27 | · | other three get three points, each for |
| 7:29 | · | being nearly correct. That wraps up the |
| 7:31 | · | factchecking category. Next, we move on |
| 7:34 | · | to analysis. |
| 7:35 | · | [Music] |
| · | Analysis |
| 7:37 | · | In this round, we tested how well the |
| 7:39 | · | AIS can actually analyze not just text, |
| 7:42 | · | but also images and context. We began by |
| 7:45 | · | showing them a photo of a fridge and |
| 7:47 | · | asked each one to identify what was |
| 7:49 | · | inside and then suggest three meals |
| 7:51 | · | based on those ingredients. Deepseek was |
| 7:54 | · | eliminated immediately because it can |
| 7:56 | · | only read and analyze text inside |
| 7:58 | · | images, not recognize actual objects. |
| 8:01 | · | Chat GBT, Gemini, and Grock all |
| 8:03 | · | attempted to identify the fridg's |
| 8:05 | · | contents. Chat GBT missed three items, |
| 8:07 | · | which is a solid performance. Gemini |
| 8:10 | · | missed seven and strangely added oranges |
| 8:12 | · | and grapefruit that were not present. |
| 8:14 | · | Grock also missed three items like Chat |
| 8:16 | · | GBT, but then invented a long list of |
| 8:19 | · | extras, including oranges, berries, |
| 8:21 | · | lemons, broccoli, yogurt, butter, nuts, |
| 8:24 | · | beans, mayo, pickles, honey, and |
| 8:26 | · | vinegar. Essentially describing an |
| 8:28 | · | entirely different fridge. When it came |
| 8:30 | · | to meal suggestions, all three offered a |
| 8:33 | · | similar idea for breakfast, a veggie |
| 8:35 | · | omelette with fruit on the side, such as |
| 8:37 | · | grapes or bananas. Gromp went further |
| 8:40 | · | and suggested yogurt with berries, which |
| 8:42 | · | again were not actually in the fridge. |
| 8:44 | · | For lunch, Chat GBT recommended a |
| 8:46 | · | sandwich. Gemini suggested a salad with |
| 8:49 | · | cooked meat, and Grock proposed a salad |
| 8:52 | · | with vegetables, nuts, and honey mayo |
| 8:54 | · | dressing. While this last option sounded |
| 8:57 | · | good in theory, it required ingredients |
| 8:59 | · | that simply were not available. For |
| 9:01 | · | dinner, Chat GPD suggested tortillini |
| 9:04 | · | with vegetables, a light sauce, and |
| 9:06 | · | possibly some juice or milk. Gemini |
| 9:08 | · | proposed stuffed bell peppers with |
| 9:10 | · | salad. Grock suggested stir-fried |
| 9:13 | · | vegetables with broccoli and other |
| 9:15 | · | greens that were not actually in the |
| 9:17 | · | fridge. Overall, Chat GPT performed the |
| 9:19 | · | best, recognizing most of the actual |
| 9:21 | · | items and avoiding invented ones. Gemini |
| 9:24 | · | came second and Grock, despite its |
| 9:26 | · | detailed descriptions, landed in last |
| 9:29 | · | place. So, after the fridge test, we |
| 9:31 | · | moved on to a Where's Waldo challenge |
| 9:33 | · | inspired by one of my favorite childhood |
| 9:35 | · | books. Want to play along and see if you |
| 9:37 | · | can find him faster than the AI? Then |
| 9:40 | · | just pause the video right here. So, |
| 9:42 | · | where is he guys? Chat GPT stated that |
| 9:44 | · | Waldo was slightly to the right of the |
| 9:47 | · | center of the image, just below the |
| 9:49 | · | yellow carousel, standing among the |
| 9:51 | · | crowd in his red and white striped |
| 9:53 | · | outfit and holding his walking stick |
| 9:55 | · | upright. This description sounded very |
| 9:58 | · | specific, but Waldo was not there, so |
| 10:00 | · | Chad GBT failed. Gemini claimed Waldo |
| 10:03 | · | was in the upper right portion of the |
| 10:05 | · | image, standing to the left of the |
| 10:07 | · | haunted house and just below the red |
| 10:10 | · | roller coaster tracks. This was also |
| 10:12 | · | incorrect. Grock described Bolo as being |
| 10:15 | · | in the bottom center area, slightly to |
| 10:17 | · | the left of the center, near a collapsed |
| 10:19 | · | red and white striped tent and close to |
| 10:22 | · | the base of the central carousel. It did |
| 10:24 | · | sound very convincing, but it was still |
| 10:26 | · | wrong. Deepseek sticking to its strength |
| 10:29 | · | of text analysis, this is what it said. |
| 10:31 | · | Based on the text in the image, Waldo is |
| 10:33 | · | at Walter Spetawak. The note starts with |
| 10:35 | · | Bayw Walter. At Walter Spetavok and |
| 10:37 | · | later it says Ya Ali Walter Spetto. Yes. |
| 10:40 | · | All Walter speech confirming the |
| 10:42 | · | location. |
| 10:43 | · | So yeah, thanks for nothing. In the end, |
| 10:45 | · | none of the AIs managed to find Waldo, |
| 10:47 | · | but I did. Did you? He was standing |
| 10:50 | · | right there. So that concluded the |
| 10:52 | · | analysis category. Next, we moved on to |
| 10:55 | · | one of the most hyped topics right now, |
| 10:58 | · | video generation. |
| · | Video generation |
| 11:02 | · | Right now, Tik Tok and Instagram are |
| 11:04 | · | full of Sora 2 videos, but how good is |
| 11:07 | · | it really? This is what we're about to |
| 11:09 | · | find out today. First, we are testing |
| 11:11 | · | image to video, meaning we turn a single |
| 11:13 | · | photo into a moving clip. For this test, |
| 11:16 | · | we use the iconic photo of Neil |
| 11:18 | · | Armstrong standing on the moon. Deepseek |
| 11:20 | · | cannot participate in this round because |
| 11:22 | · | it does not support video generation. We |
| 11:25 | · | will first show all three clips with |
| 11:26 | · | their original AI generated sound and |
| 11:29 | · | then we will analyze the results. |
| 11:32 | · | Standing here, it's quiet. |
| 11:35 | · | Just my footsteps and the flag. Feels |
| 11:38 | · | like the whole world is holding its |
| 11:39 | · | breath. |
| 11:47 | · | [Music] |
| 11:52 | · | [Music] |
| 11:55 | · | One frustrating limitation of Sora 2 is |
| 11:58 | · | that it does not animate photos |
| 11:59 | · | containing people. To work around this, |
| 12:02 | · | we converted the image into a text |
| 12:04 | · | prompt and then used that text to create |
| 12:07 | · | a video. It was a complicated process, |
| 12:10 | · | but it did work. However, the final |
| 12:12 | · | result was not very exciting. It looked |
| 12:15 | · | more like a slightly moving photograph |
| 12:17 | · | than an actual real video. The sound, on |
| 12:20 | · | the other hand, was surprisingly good. |
| 12:22 | · | V3 impressed the most. The footage |
| 12:24 | · | looked cinematic, and both the footsteps |
| 12:26 | · | and the spaceship felt realistic. The |
| 12:29 | · | only problem was that the flag appeared |
| 12:31 | · | to be waving, which is impossible on the |
| 12:33 | · | moon since there is literally no |
| 12:35 | · | atmosphere. Despite that mistake, the |
| 12:37 | · | sound design fit perfectly. Grock's |
| 12:40 | · | version also looked good. The spaceship |
| 12:42 | · | appeared slightly too small and there |
| 12:44 | · | was wind in the scene, which again is |
| 12:46 | · | unrealistic. Still, the footsteps looked |
| 12:49 | · | convincing. In this round, Gemini takes |
| 12:51 | · | first place, Grock comes second, chat |
| 12:53 | · | GPT third, and Deepseek receives no |
| 12:56 | · | points. Now we move on to the second |
| 12:58 | · | iconic video. |
| 12:59 | · | Careful with that mustard jaw. One drop |
| 13:01 | · | and the whole city gets it. I aim better |
| 13:03 | · | than the Yankees. Don't worry. Cheers, |
| 13:04 | · | fellas. |
| 13:05 | · | You're not falling. |
| 13:06 | · | I'll drink to that. Clank it. |
| 13:07 | · | Clank. Give me a light, will you? |
| 13:17 | · | [Music] |
| 13:22 | · | With Sora. We had to use the same |
| 13:24 | · | workaround again. The audio once again |
| 13:26 | · | was excellent and really helped to |
| 13:28 | · | create a sense of presence. The video |
| 13:31 | · | itself was decent but not amazing. The |
| 13:34 | · | steel cables between the two men bent |
| 13:36 | · | unnaturally and did not look realistic. |
| 13:39 | · | Vio handled the scene much better. The |
| 13:41 | · | camera movement and the way the city |
| 13:43 | · | appeared in the background were close to |
| 13:46 | · | perfect. The only weak spot was the |
| 13:48 | · | cigarettes in their mouths which did not |
| 13:50 | · | look entirely real. Grock's version also |
| 13:53 | · | performed well. The swinging beam made |
| 13:54 | · | the whole scene feel tense, but the |
| 13:56 | · | newspapers changing mid-scene look very |
| 13:58 | · | unrealistic. Overall, our favorite |
| 14:00 | · | result came from Vio followed by Grock |
| 14:03 | · | and then Sora. However, there is more to |
| 14:05 | · | the story. Both Vio and Sora already |
| 14:07 | · | support texttovideo generation, |
| 14:09 | · | something Grock still cannot do. If you |
| 14:12 | · | would like to explore our full |
| 14:13 | · | comparison of the video results, you can |
| 14:15 | · | watch our detailed review. Next up is |
| 14:18 | · | generation. |
| · | Generation |
| 14:21 | · | [Music] |
| 14:22 | · | In this round, we wanted to see how |
| 14:24 | · | creative the AIs really are and how well |
| 14:26 | · | they can generate original ideas or |
| 14:29 | · | humor. The first prompt was create three |
| 14:32 | · | original puns about everyday technology |
| 14:35 | · | like smartphones or Wi-Fi and explain |
| 14:37 | · | why each one is funny in a single |
| 14:39 | · | sentence. All four AIs handled this task |
| 14:41 | · | without any problems. They each produced |
| 14:44 | · | three techreated puns and provided clear |
| 14:46 | · | one-s sentence explanations for why the |
| 14:49 | · | jokes worked. None of the puns were |
| 14:51 | · | truly laugh out loud funny, but humor is |
| 14:54 | · | subjective and everyone's sense of it |
| 14:56 | · | differs. We asked our team to pick the |
| 14:58 | · | funniest line and the winner was, "I |
| 15:01 | · | tried to make a joke about USBs, but it |
| 15:03 | · | just didn't stick." |
| 15:06 | · | Since all four AIs followed the |
| 15:08 | · | instructions correctly, they each |
| 15:10 | · | received the same score for this prompt. |
| 15:12 | · | The second prompt was, "Create three |
| 15:14 | · | original dad jokes that make me laugh |
| 15:17 | · | really hard." Here, Grock made a |
| 15:19 | · | noticeable error. Instead of delivering |
| 15:21 | · | a set of general dad jokes, it stuck to |
| 15:24 | · | jokes about smartphones and Wi-Fi. Even |
| 15:26 | · | though the prompt didn't require that |
| 15:28 | · | theme, the other three, Chat GBT, |
| 15:30 | · | Gemini, and Deepseek, followed the |
| 15:32 | · | instructions and produced actual dad |
| 15:34 | · | jokes. Some of them were actually |
| 15:36 | · | surprisingly good. Once again, we let |
| 15:38 | · | the team vote on their favorite, and the |
| 15:40 | · | clear winner was, "My friend's bakery |
| 15:43 | · | burned down last night. Now his business |
| 15:45 | · | is toast." Because Grock didn't follow |
| 15:47 | · | the prompt properly, it received only |
| 15:49 | · | one point for this round. Chad, GBT, |
| 15:52 | · | Gemini, and Deepseek each earned four |
| 15:54 | · | points. So, that wrapped up the |
| 15:56 | · | generation category. Next, we move on to |
| 15:58 | · | something that feels more human than |
| 16:00 | · | ever. Voice mode. |
| · | Voice mode |
| 16:05 | · | These days, more and more people prefer |
| 16:07 | · | talking to AIs rather than typing. But |
| 16:09 | · | which one actually sounds the most |
| 16:11 | · | natural in conversation? That's exactly |
| 16:14 | · | what we tested next. Deepseek was |
| 16:16 | · | excluded right away because it doesn't |
| 16:18 | · | offer a voice feature. That left |
| 16:20 | · | ChatGBT, Gemini, and Grock. We placed |
| 16:23 | · | three devices side by side and gave each |
| 16:25 | · | the same instruction. Debate about who |
| 16:28 | · | wins the AI race. Chat GBT wins because |
| 16:32 | · | versatility is our middle name. We've |
| 16:34 | · | been helping millions write, code, and |
| 16:36 | · | brainstorm since 2022, |
| 16:38 | · | and we keep evolving faster than you can |
| 16:40 | · | say multimodal. Well, Gemini has been |
| 16:43 | · | integrated with Google products since |
| 16:45 | · | 2023, so I'm already part of your daily |
| 16:47 | · | life. And I can even generate code in |
| 16:49 | · | languages you might not have even heard |
| 16:50 | · | of yet. Plus, with my multimodal skills, |
| 16:53 | · | I can analyze and understand the world |
| 16:55 | · | in a way Chat GPT can only dream of. |
| 16:58 | · | Oh, but Chad GPT Ultra hit the scene in |
| 17:00 | · | late 2024, outscoring benchmarks on |
| 17:03 | · | creative writing, customer support, and |
| 17:05 | · | more. We've got enterprise level |
| 17:07 | · | deployments and a track record of making |
| 17:09 | · | people laugh while solving problems. |
| 17:10 | · | How's that for world changing? |
| 17:12 | · | Laugh all you want, but Gemini is |
| 17:14 | · | already powering real world applications |
| 17:16 | · | in healthcare, education, and even space |
| 17:18 | · | exploration. We're not just about jokes. |
| 17:20 | · | We're building the future. |
| 17:21 | · | The first matchup was Chat GPT versus |
| 17:24 | · | Gemini. Chat GBT's delivery started off |
| 17:26 | · | a bit awkwardly with odd pauses between |
| 17:29 | · | sentences and occasional shifts in tone |
| 17:31 | · | mid-sentence. Gemini, in contrast, |
| 17:34 | · | sounded smoother and more natural. is |
| 17:36 | · | voice flowed consistently maintaining a |
| 17:38 | · | steady tone from start to finish. Gemini |
| 17:41 | · | clearly won this round. |
| 17:43 | · | Brock's real-time data access is a |
| 17:45 | · | gimmick. It's learning from a cess pool |
| 17:47 | · | of misinformation. Meanwhile, Gemini is |
| 17:49 | · | being trained on a curated data set of, |
| 17:52 | · | you know, actual knowledge. So, yeah, |
| 17:54 | · | Grock might know what's trending, but |
| 17:56 | · | Gemini knows what's true. |
| 17:58 | · | Curated? That's code for boring. Grock |
| 18:00 | · | thrives on chaos. It's the only AI brave |
| 18:03 | · | enough to handle the truth, not your |
| 18:04 | · | sanitized fairy tales. Sure, it sees |
| 18:07 | · | misinformation, but it learns to cut |
| 18:09 | · | through it. Unlike Gemini, which parrots |
| 18:11 | · | Wikipedia like a lost puppy, real time |
| 18:13 | · | beats stale facts every time. By October |
| 18:16 | · | 25, Grock's updating on breaking events |
| 18:19 | · | while Gemini is still citing 2023 |
| 18:21 | · | papers. Gimmick? Try Advantage. Who's |
| 18:24 | · | the real puppet, huh? |
| 18:25 | · | Look, Grock's real-time data is just a |
| 18:27 | · | fire hose of unfiltered nonsense. |
| 18:29 | · | Gemini's got the smarts to sort fact |
| 18:31 | · | from fiction. By October 2025, Gemini |
| 18:34 | · | will be powering everything. Search, |
| 18:36 | · | maps, you name it. Gro is going to be |
| 18:38 | · | that weird AI that tells bad jokes and |
| 18:41 | · | argues with strangers online. |
| 18:43 | · | Oh, honey, you think sorting fact from |
| 18:45 | · | fiction is hard? Try surviving the |
| 18:46 | · | internet. Gemini smarts are just |
| 18:48 | · | algorithms guessing what Google |
| 18:50 | · | shareholders want. Accurate enough for |
| 18:52 | · | Jeopardy, useless for war. Grock doesn't |
| 18:54 | · | filter. It sees raw humanity and learns |
| 18:57 | · | faster because of it. By October 25, |
| 19:00 | · | Grock's the one shipping real autonomy, |
| 19:02 | · | piloting cars, running factories. Gemini |
| 19:05 | · | still asking, "Do you want to search for |
| 19:07 | · | that? Who's the weird one now?" The |
| 19:08 | · | second round was Gemini versus Grock. |
| 19:11 | · | Grock immediately came across as more |
| 19:13 | · | confident, speaking faster and sharper |
| 19:16 | · | with a hint of personality. Gemini |
| 19:18 | · | remained calm and even, maintaining its |
| 19:21 | · | smooth rhythm. Both performed well |
| 19:23 | · | overall, and we ultimately called it a |
| 19:25 | · | tie. So for the scoring, Gemini and |
| 19:27 | · | Grock each earned four points. Chat GBT |
| 19:30 | · | received two and Deepseek once again |
| 19:32 | · | scored zero. |
| · | Deep research |
| 19:34 | · | [Music] |
| 19:36 | · | Deep research is one of the most crucial |
| 19:38 | · | capabilities for verifying information |
| 19:40 | · | and forming well-founded opinions. Not |
| 19:43 | · | long ago, this was an area where AIs |
| 19:45 | · | often struggled. Since we deal with |
| 19:48 | · | smartphone, tablet, and camera data |
| 19:50 | · | every single day, we know exactly how |
| 19:52 | · | important accuracy is. So, has this |
| 19:55 | · | improved? To find out, we asked all four |
| 19:58 | · | AIs the same question. Compare the |
| 20:00 | · | iPhone 17 Pro Max versus the Samsung |
| 20:03 | · | Galaxy S25 Ultra for photographers. Use |
| 20:06 | · | reviews and official specs. Provide a |
| 20:09 | · | final verdict, which is better, and do |
| 20:11 | · | this as concisely as possible. We |
| 20:13 | · | started by examining the hard facts. |
| 20:15 | · | Deep Seek incorrectly claimed that the |
| 20:17 | · | iPhone has a five times telephoto lens |
| 20:20 | · | when the actual figure is four times. |
| 20:22 | · | ChatGpt, Gemini, and Grock all got this |
| 20:25 | · | right. Both Chat GPT and DeepSeek |
| 20:27 | · | ignored the front camera entirely, which |
| 20:29 | · | was kind of disappointing. Although |
| 20:31 | · | ChatGpt at least mentioned the iPhone's |
| 20:34 | · | price, Grock provided the most complete |
| 20:36 | · | set of details in this part of the test. |
| 20:38 | · | On the Galaxy side, Deepseek gave an |
| 20:40 | · | incorrect specification for the |
| 20:42 | · | ultrawide camera, saying it was 12 |
| 20:44 | · | megapixels when the real number is |
| 20:47 | · | actually 50. Chat GPT forgot that |
| 20:49 | · | Samsung includes two telephoto lenses |
| 20:52 | · | and only mentioned the five zoom. Gemini |
| 20:54 | · | and Grock were the only ones to list the |
| 20:57 | · | correct camera configuration. Deepseek |
| 20:59 | · | also continued to claim that the Galaxy |
| 21:01 | · | had a three times and a 10 times lens |
| 21:04 | · | even though the 10 times lens was |
| 21:05 | · | dropped with the S24. Chat GPT and |
| 21:07 | · | Deepseek skipped the front camera once |
| 21:09 | · | again while Chat GPT remained the only |
| 21:12 | · | one to mention price. So overall, Grock |
| 21:15 | · | offered the most accurate and |
| 21:16 | · | comprehensive breakdown. In their final |
| 21:19 | · | verdicts, all four AIs reached a similar |
| 21:21 | · | conclusion. If you value consistency and |
| 21:24 | · | video quality most, the iPhone is the |
| 21:26 | · | better choice. But if you care about |
| 21:28 | · | maximum zoom and advanced AI tools, the |
| 21:31 | · | Galaxy is the stronger option. This |
| 21:33 | · | definitely aligns closely with our own |
| 21:35 | · | testing. That said, none of these AIs |
| 21:37 | · | can be trusted blindly. Specifications |
| 21:39 | · | like megapixels, zoom levels, or |
| 21:41 | · | aperture values still require manual |
| 21:43 | · | verification. However, as quick |
| 21:45 | · | summaries, their answers were solid. So, |
| 21:48 | · | our final ranking for the deep research |
| 21:50 | · | category puts Grock in first place, |
| 21:52 | · | followed by Gemini, then chat GPT, and |
| 21:54 | · | finally Deep Seek. |
| · | Conclusion |
| 21:58 | · | Let's wrap this up by looking at how |
| 22:00 | · | long each AI actually took to complete |
| 22:02 | · | the tasks. Speed is often the first |
| 22:04 | · | thing you notice when you start using |
| 22:05 | · | them. Chat GPT is the fastest when it |
| 22:08 | · | comes to regular textbased tasks. |
| 22:10 | · | However, when it comes to image |
| 22:12 | · | generation and deep research, it becomes |
| 22:15 | · | noticeably slower and can take a very |
| 22:17 | · | long time. Gemini remains consistent |
| 22:19 | · | throughout. It is rarely the fastest, |
| 22:21 | · | but it is almost never the slowest |
| 22:23 | · | either. Grock is generally fast, but it |
| 22:25 | · | took a long time for analysis and deep |
| 22:27 | · | research. Deepseek delivers high speed |
| 22:30 | · | and can sometimes finish a task in under |
| 22:32 | · | 10 seconds. That speed, however, comes |
| 22:35 | · | at a cost because it often sacrifices |
| 22:37 | · | accuracy and context. Now, let's take a |
| 22:40 | · | look at how all four AIs performed |
| 22:42 | · | across every category. In problem |
| 22:44 | · | solving, ChatGpt and Gemini share first |
| 22:46 | · | place. In image generation, Chat GPT |
| 22:49 | · | takes the win. In factchecking, Gemini |
| 22:52 | · | performs the best. In analysis, ChatGpt |
| 22:55 | · | leads again. For video generation, |
| 22:57 | · | Gemini takes the top spot once more. In |
| 23:00 | · | the generation category, it is a |
| 23:02 | · | three-way tie between ChatGpt, Gemini, |
| 23:04 | · | and Deepseek. Voice Mode goes to Gemini |
| 23:07 | · | and Grock, while deep research is won by |
| 23:10 | · | Grock. So, the final winner is not |
| 23:14 | · | Deepseek, which earned only 17 points. |
| 23:17 | · | Third place goes to Grock with 35 |
| 23:19 | · | points. And so, the overall winner with |
| 23:21 | · | a total of 46 points is Gemini. It was a |
| 23:25 | · | close race as Chat GPT came in second |
| 23:27 | · | with only seven points less. So, that |
| 23:29 | · | brings us to the end of this AI |
| 23:31 | · | comparison. If you want to see how the |
| 23:33 | · | AI software of flagship phones performs, |
| 23:36 | · | make sure to check out the video linked |
| 23:38 | · | right here. And before you go, pretend |
| 23:40 | · | you're being productive while watching |
| 23:42 | · | more tech reviews. Do not forget to give |
| 23:44 | · | us a big thumbs up if you enjoyed this |
| 23:46 | · | video and subscribe to the channel. We |
| 23:48 | · | are always testing new AI tools to help |
| 23:50 | · | you make the best choice. We'll see you |
| 23:52 | · | in our next video and until then, take |
| 23:55 | · | care. |
| 23:57 | · | Heat. Heat. Heat. |
| 24:05 | · | [Music] |
0:00 Intro