御姐婷婷
婷婷炒币亏了50w，励志回本
0 Following
26 Followers
211 Liked
0 Shared
See original
In the past two days, I tried Chat GPT 5.2. This version has been criticized heavily by users both domestically and internationally. I'll start with the conclusion: based on my shallow usage, the emergence of this version may represent the arrival of spring for liberal arts students. A person's expressive ability may be the most important skill in the next 5-10 years.
Currently, the overseas evaluation of GPT 5.2 is somewhat extreme. On one side, there are comments like "it's dumbed down," "it doesn't dare to speak," "it's been castrated," while on the other hand, there are comments like "the code is more stable," "it's better for research use." If we only look at the emotions, it's easy to come to a superficial conclusion: the latest version is not good.
However, if we break these feedbacks down, we find an interesting divide: those who criticize the most harshly often do so not because it can't do something, but because it has become "less fun," "no longer engaging in conversation," and "no longer daring to speak on my behalf."
And this is exactly where 5.2 has truly changed.
1. 5.2 hasn't become dumb; it has just changed its way of speaking.
Many people feel that 5.2 is "not expanding," "too cautious," and "too steady." In fact, this isn't a decline in capability, but rather a tightening of expressive strategy.
Before 5.1, GPT was very good at one thing: crafting a complete narrative based on your questions. Even if it wasn't sure about certain aspects, it would elaborate as if it had everything figured out. This was great for chatting, brainstorming, and emotional value, but it poses risks in many professional scenarios.
5.2 clearly tends to confirm boundaries before proceeding. It prefers to tell you that there are actually multiple interpretations here, that this premise may not necessarily hold, rather than directly giving a satisfying-sounding answer. This may make some users feel bored and less daring to speak than before, but from another perspective, it resembles a collaborator who knows when to apply the brakes, rather than just a conversational companion that caters to everything.
2. Why I feel that: the era of liberal arts students may have arrived.
A change that is rarely mentioned is that 5.2 has a significantly higher tolerance and parsing ability for natural language prompts.
In the past, if you wanted GPT to do something complex, you often had to write prompts like half a technical manual, with all levels, formats, and constraints accounted for. Even prompt editors have become new positions in many startups domestically, combining roles with card draw specialists for AI dramas. Your narrative expression ability would be strong, and your profitability would be high. Now, if you describe your needs in a more conversational, vague, and human way, it can grasp the essentials and know exactly what you want to do.
This is obviously a plus for programmers, but for liberal arts students and researchers, the improvement is even greater:
When doing analysis, commentary, or popular science, you can use phrases like "let's think from another angle" or "is there a bit of counter-intuition here?" It can follow your train of thought and continue pushing rather than pulling back to a template. Using closed prompts to limit searches within a specified range is also a real limitation, not fabrication. It's as if 5.2 has become an aggregation platform + crawler + generative AI, rather than just a generative AI.
On open-ended questions, it resembles an entity willing to slowly clarify logic with you, rather than rushing to provide conclusions.
In a sense, 5.2 has elevated the act of writing prompts to a higher level. This is a natural order and phrasing that is easier to master than programming languages. You don't need to master a bunch of techniques; you just need to express yourself clearly, as if you were communicating with someone.
3. Negative reviews concentrate on a few points, which can actually be explained.
The regressions that have been repeatedly mentioned can generally be categorized into a few types:
Not daring enough to speak, low emotional value.
This is a fact, but it's also a choice. OpenAI is clearly suppressing pleasing expressions, which is unfriendly to users seeking emotional companionship and casual chats, but a plus for serious usage scenarios.
Less elaboration, details require follow-up questions.
5.2 is more like waiting for you to confirm the direction before digging deeper, rather than laying out all possibilities at once. It gives more initiative back to the users, encouraging more thoughtful liberal arts students rather than those who want it to do everything for them, thus making your purpose stronger. Knowing the paths that can be achieved through conventional methods and then using AI to express them will be much more efficient.
Slow speed, expensive computing power.
This is undeniable; xhigh and thinking modes are indeed slow and expensive. But you'll find that the slowness mainly comes from it genuinely reading through all the context; currently, it's estimated to handle about 30,000 lines of code or no more than 50,000 English words (it can actually reach around 80,000, but analysis efficiency will decrease), rather than instantly replying with a bunch of seemingly reasonable nonsense.
Writing is not as smooth as Gemini.
Many people agree with this point; GPT's writing still leans towards rationality and structure. It's better suited for topic confirmation before papers, building frameworks for review meetings, and filling in information. In other words, it's more suitable as a draft and logical skeleton, like a palette and sketch paper, while Gemini is more like a pen.
4. It feels more like a tool than a companion role.
If I had to give 5.2 a positioning, I would say: it is moving from an all-purpose chat AI towards a general cognitive tool.
It is less willing to make decisions for you, less willing to provide emotional support, and less willing to give you a satisfying answer in uncertain areas. But it excels at:
Helping you clarify a vague problem.
Pointing out the assumptions that may be hidden in your words.
Steadily laying out reasoning step by step after you provide direction.
This may indeed be less useful for many people than before, but for another group of people, it feels more handy, just like Watson beside Sherlock Holmes.
GPT 5.2 is certainly not a perfect version; it still has obvious shortcomings in style adjustment, images, and multimodal capabilities. It doesn't match some of the larger models domestically, and its pricing strategy is quite aggressive. But to simply classify it as a regression or a failure is somewhat unfair.
It feels more like a deliberate adjustment of trade-offs, with less pleasing and more restraint; less performance and more tool-like quality.
If what you want is casual conversation, inspiration bombardment, and emotional resonance, then it is indeed less interesting than before. But if you want an assistant that can understand human language and help you sort out problems using natural language, then 5.2 has finally grown into a long-term collaborator.
Perhaps this is why some feel it has worsened, while others feel it has just become useful.
Currently, the overseas evaluation of GPT 5.2 is somewhat extreme. On one side, there are comments like "it's dumbed down," "it doesn't dare to speak," "it's been castrated," while on the other hand, there are comments like "the code is more stable," "it's better for research use." If we only look at the emotions, it's easy to come to a superficial conclusion: the latest version is not good.
However, if we break these feedbacks down, we find an interesting divide: those who criticize the most harshly often do so not because it can't do something, but because it has become "less fun," "no longer engaging in conversation," and "no longer daring to speak on my behalf."
And this is exactly where 5.2 has truly changed.
1. 5.2 hasn't become dumb; it has just changed its way of speaking.
Many people feel that 5.2 is "not expanding," "too cautious," and "too steady." In fact, this isn't a decline in capability, but rather a tightening of expressive strategy.
Before 5.1, GPT was very good at one thing: crafting a complete narrative based on your questions. Even if it wasn't sure about certain aspects, it would elaborate as if it had everything figured out. This was great for chatting, brainstorming, and emotional value, but it poses risks in many professional scenarios.
5.2 clearly tends to confirm boundaries before proceeding. It prefers to tell you that there are actually multiple interpretations here, that this premise may not necessarily hold, rather than directly giving a satisfying-sounding answer. This may make some users feel bored and less daring to speak than before, but from another perspective, it resembles a collaborator who knows when to apply the brakes, rather than just a conversational companion that caters to everything.
2. Why I feel that: the era of liberal arts students may have arrived.
A change that is rarely mentioned is that 5.2 has a significantly higher tolerance and parsing ability for natural language prompts.
In the past, if you wanted GPT to do something complex, you often had to write prompts like half a technical manual, with all levels, formats, and constraints accounted for. Even prompt editors have become new positions in many startups domestically, combining roles with card draw specialists for AI dramas. Your narrative expression ability would be strong, and your profitability would be high. Now, if you describe your needs in a more conversational, vague, and human way, it can grasp the essentials and know exactly what you want to do.
This is obviously a plus for programmers, but for liberal arts students and researchers, the improvement is even greater:
When doing analysis, commentary, or popular science, you can use phrases like "let's think from another angle" or "is there a bit of counter-intuition here?" It can follow your train of thought and continue pushing rather than pulling back to a template. Using closed prompts to limit searches within a specified range is also a real limitation, not fabrication. It's as if 5.2 has become an aggregation platform + crawler + generative AI, rather than just a generative AI.
On open-ended questions, it resembles an entity willing to slowly clarify logic with you, rather than rushing to provide conclusions.
In a sense, 5.2 has elevated the act of writing prompts to a higher level. This is a natural order and phrasing that is easier to master than programming languages. You don't need to master a bunch of techniques; you just need to express yourself clearly, as if you were communicating with someone.
3. Negative reviews concentrate on a few points, which can actually be explained.
The regressions that have been repeatedly mentioned can generally be categorized into a few types:
Not daring enough to speak, low emotional value.
This is a fact, but it's also a choice. OpenAI is clearly suppressing pleasing expressions, which is unfriendly to users seeking emotional companionship and casual chats, but a plus for serious usage scenarios.
Less elaboration, details require follow-up questions.
5.2 is more like waiting for you to confirm the direction before digging deeper, rather than laying out all possibilities at once. It gives more initiative back to the users, encouraging more thoughtful liberal arts students rather than those who want it to do everything for them, thus making your purpose stronger. Knowing the paths that can be achieved through conventional methods and then using AI to express them will be much more efficient.
Slow speed, expensive computing power.
This is undeniable; xhigh and thinking modes are indeed slow and expensive. But you'll find that the slowness mainly comes from it genuinely reading through all the context; currently, it's estimated to handle about 30,000 lines of code or no more than 50,000 English words (it can actually reach around 80,000, but analysis efficiency will decrease), rather than instantly replying with a bunch of seemingly reasonable nonsense.
Writing is not as smooth as Gemini.
Many people agree with this point; GPT's writing still leans towards rationality and structure. It's better suited for topic confirmation before papers, building frameworks for review meetings, and filling in information. In other words, it's more suitable as a draft and logical skeleton, like a palette and sketch paper, while Gemini is more like a pen.
4. It feels more like a tool than a companion role.
If I had to give 5.2 a positioning, I would say: it is moving from an all-purpose chat AI towards a general cognitive tool.
It is less willing to make decisions for you, less willing to provide emotional support, and less willing to give you a satisfying answer in uncertain areas. But it excels at:
Helping you clarify a vague problem.
Pointing out the assumptions that may be hidden in your words.
Steadily laying out reasoning step by step after you provide direction.
This may indeed be less useful for many people than before, but for another group of people, it feels more handy, just like Watson beside Sherlock Holmes.
GPT 5.2 is certainly not a perfect version; it still has obvious shortcomings in style adjustment, images, and multimodal capabilities. It doesn't match some of the larger models domestically, and its pricing strategy is quite aggressive. But to simply classify it as a regression or a failure is somewhat unfair.
It feels more like a deliberate adjustment of trade-offs, with less pleasing and more restraint; less performance and more tool-like quality.
If what you want is casual conversation, inspiration bombardment, and emotional resonance, then it is indeed less interesting than before. But if you want an assistant that can understand human language and help you sort out problems using natural language, then 5.2 has finally grown into a long-term collaborator.
Perhaps this is why some feel it has worsened, while others feel it has just become useful.
See original
This matter also trended a couple of days ago, and I just saw it. It's embarrassing to say, but I rarely drive myself, and I just use the sound system as it was sold to me without ever thinking that there is such a thing as tuning the sound system. However, after watching the video, I'm now quite ambitious and plan to adjust my own sound system, after all, listening to music while driving is quite a pleasure.
See original
This might be the strangest global competition, but it offers a new perspective on Excel. Many people saw the news saying 'Tsinghua postdoctoral fellow ranked ninth in the Excel World Championship globally,' and their first reaction might be:
Excel has a World Championship? This thing can be competitive? Can I sign up too?
But once you actually click on the competition video, you can basically take those thoughts back immediately.
The competition scene is a standard esports setup: commentary, lighting, music, and an audience are all present, the only thing different from regular esports is that the contestants are calm and serious, as if they have already finished writing tonight's weekly report. The commentators' outfits are covered with various formulas, creating an atmosphere that feels oddly like a joke, but the competition content is anything but a joke; it is a compressed 30-minute extreme data processing exam.
The rules are very simple: each round issues an Excel file with dozens of questions divided into several levels. Some calculate costs, some calculate efficiency, some calculate resource distribution, and some even imply complex dependencies.
What the contestants need to do is to answer these questions correctly as much as possible before time runs out. And the so-called esports, from the audience's perspective, can only be understood in one way: the scores bouncing back and forth. You have no idea what is happening on the contestants' screens because the scenes switch too quickly, and the questions are not displayed.
The only ones who can see the entire operation process are the live audience, as each contestant's screen is projected onto the monitors in front of their seats, which might even be the most valuable aspect of the tickets for this competition.
This year's competition features free questions from EVE Online, a game long jokingly referred to as 'the space MMO that plays Excel.' The questions start with mining and extend all the way to shipbuilding costs, market price fluctuations, mineral consumption structures, and planetary resource distribution... Many people watching are confused, and so was I; only later did I realize that this set of questions is not really testing Excel but more like giving you a miniature cosmic economic system and asking you to use Excel to derive the formulas behind running planets and spaceships.
The truly difficult part is not the formulas themselves but the speed and stability, as well as programming.
There are many questions, but the logic is similar; you must create a reusable data model in a very short time, otherwise, even if you type a bit faster manually, you won't make it in time.
So what the competition really tests is who can, in a very short time, see the structure clearly, sort out the relationships neatly, build the model correctly, and then make the entire sheet run automatically under different parameters.
If it were just VLOOKUP, SUMIF, and those functions, it wouldn't be enough. The competition heavily utilizes new functions like XLOOKUP, FILTER, LET, LAMBDA, SUMPRODUCT, etc. If you haven't seen or are not familiar with them, it's virtually impossible to keep up with the pace.
Some questions even use non-standard data, such as distinguishing ship types by background color, which is a negative example in actual work, but in the competition, you have to find a way to convert colors into text and then match them with other sheets.
It's not just difficult; it's annoying, and it must be solved.
In this environment, the Tsinghua postdoctoral fellow ranked ninth globally, which has a very clear significance. He demonstrated not how many functions he knows but the ability to engineer data processing, quickly seeing structures in chaotic tables, abstracting reusable models from complex dependencies, and ensuring stability, accuracy, and composure when facing dozens of questions.
It's a competition of thinking organization ability, complex logic decomposition ability, plus a bit of pressure resistance under competition rhythm.
This achievement indicates that the basic skills in data processing, system modeling, and engineering thinking in the country are very solid.
Excel here is no longer just an office software but a lightweight tool that can build small models in a short time, and what contestants are actually doing is using Excel to build a micro-system.
There are always people online who joke about Excel skills being equivalent to proficient use, as if any resume could include that phrase, but in the future, they might not dare to write that they can use this on their resumes. After watching the competition, one will understand that the value of the word 'proficient' is actually very high; it is not about knowing a few functions or making some beautiful charts; it is about:
Receiving a pile of data in hand
Being able to immediately see how to break it down
Knowing how to string formulas together
Turning dozens of lines of logic into a stable model
And being able to avoid mistakes under pressure
This ability is applicable in analyzing, managing supply chains, modeling, and operational forecasting in companies, and in the competition, it is compressed into 30 minutes, perhaps revealing its most authentic intensity.
And this kind of thinking is the most precious; it can extend to solve problems people encounter in real life and also highlights the value and power of programmers who understand mathematics.
Everything in this world requires logic.
Excel has a World Championship? This thing can be competitive? Can I sign up too?
But once you actually click on the competition video, you can basically take those thoughts back immediately.
The competition scene is a standard esports setup: commentary, lighting, music, and an audience are all present, the only thing different from regular esports is that the contestants are calm and serious, as if they have already finished writing tonight's weekly report. The commentators' outfits are covered with various formulas, creating an atmosphere that feels oddly like a joke, but the competition content is anything but a joke; it is a compressed 30-minute extreme data processing exam.
The rules are very simple: each round issues an Excel file with dozens of questions divided into several levels. Some calculate costs, some calculate efficiency, some calculate resource distribution, and some even imply complex dependencies.
What the contestants need to do is to answer these questions correctly as much as possible before time runs out. And the so-called esports, from the audience's perspective, can only be understood in one way: the scores bouncing back and forth. You have no idea what is happening on the contestants' screens because the scenes switch too quickly, and the questions are not displayed.
The only ones who can see the entire operation process are the live audience, as each contestant's screen is projected onto the monitors in front of their seats, which might even be the most valuable aspect of the tickets for this competition.
This year's competition features free questions from EVE Online, a game long jokingly referred to as 'the space MMO that plays Excel.' The questions start with mining and extend all the way to shipbuilding costs, market price fluctuations, mineral consumption structures, and planetary resource distribution... Many people watching are confused, and so was I; only later did I realize that this set of questions is not really testing Excel but more like giving you a miniature cosmic economic system and asking you to use Excel to derive the formulas behind running planets and spaceships.
The truly difficult part is not the formulas themselves but the speed and stability, as well as programming.
There are many questions, but the logic is similar; you must create a reusable data model in a very short time, otherwise, even if you type a bit faster manually, you won't make it in time.
So what the competition really tests is who can, in a very short time, see the structure clearly, sort out the relationships neatly, build the model correctly, and then make the entire sheet run automatically under different parameters.
If it were just VLOOKUP, SUMIF, and those functions, it wouldn't be enough. The competition heavily utilizes new functions like XLOOKUP, FILTER, LET, LAMBDA, SUMPRODUCT, etc. If you haven't seen or are not familiar with them, it's virtually impossible to keep up with the pace.
Some questions even use non-standard data, such as distinguishing ship types by background color, which is a negative example in actual work, but in the competition, you have to find a way to convert colors into text and then match them with other sheets.
It's not just difficult; it's annoying, and it must be solved.
In this environment, the Tsinghua postdoctoral fellow ranked ninth globally, which has a very clear significance. He demonstrated not how many functions he knows but the ability to engineer data processing, quickly seeing structures in chaotic tables, abstracting reusable models from complex dependencies, and ensuring stability, accuracy, and composure when facing dozens of questions.
It's a competition of thinking organization ability, complex logic decomposition ability, plus a bit of pressure resistance under competition rhythm.
This achievement indicates that the basic skills in data processing, system modeling, and engineering thinking in the country are very solid.
Excel here is no longer just an office software but a lightweight tool that can build small models in a short time, and what contestants are actually doing is using Excel to build a micro-system.
There are always people online who joke about Excel skills being equivalent to proficient use, as if any resume could include that phrase, but in the future, they might not dare to write that they can use this on their resumes. After watching the competition, one will understand that the value of the word 'proficient' is actually very high; it is not about knowing a few functions or making some beautiful charts; it is about:
Receiving a pile of data in hand
Being able to immediately see how to break it down
Knowing how to string formulas together
Turning dozens of lines of logic into a stable model
And being able to avoid mistakes under pressure
This ability is applicable in analyzing, managing supply chains, modeling, and operational forecasting in companies, and in the competition, it is compressed into 30 minutes, perhaps revealing its most authentic intensity.
And this kind of thinking is the most precious; it can extend to solve problems people encounter in real life and also highlights the value and power of programmers who understand mathematics.
Everything in this world requires logic.
See original
Today I went to the cinema and the seating rate was okay, about half. The audience's reaction was somewhat unified: there were laughs in the first half, but it was completely silent in the second half. This kind of viewing experience is very familiar; viewers who have seen 'I Am Not Wang Mao' and 'Killing' will find the atmosphere very similar, not the kind of howling tragedy, but rather a gradual pressure that lowers your emotions to the point where you don't really want to speak. This reminds me of 'My Squad Leader, My Squad.'
This time, the combination of Kong Sheng and Lan Xiao Long is not about grand narratives, nor is it filled with gore, but rather tells a story of an ordinary person forced to grow up by war. I saw some viewers say that after watching this film, they felt a bit stifled; it wasn't that it was hard to watch, but it was heavy.
The film's emotional trajectory is quite clear: the beginning is a bit lighthearted, with life anecdotes and moments for the audience to catch their breath; in the second half, the characters start to rush into desperate situations, and the cinema's atmosphere becomes increasingly tense. It’s not because the visuals are brutal, but because the characters are incredibly ordinary, and the more you watch, the guiltier you feel, because you know that in such moments, ordinary people are the easiest to be trampled by history.
In this respect, it is very much like 'Squad Leader'; the characters don't have a sense of responsibility from the outset but gradually toughen up under pressure. The toughness isn’t the kind of climactic intensity, but rather a resigned acceptance of 'Well, we’ve come this far.' In the end, they fight because they have nowhere to live and nowhere to retreat; this motivation resembles the logic of awakening forced upon the characters in Lan Xiao Long's works. Friends who are used to genre films might not be too fond of this style.
I went with a few friends, and the shot we discussed the most afterward was the immobile goat after the shipwreck. The goat stood on the floating debris, completely still, not even its eyes moved. At that moment, many people couldn't laugh; they just felt it was strange but couldn’t articulate what was strange about it. This absurd yet solid feeling has appeared in 'Killing' and 'Wang Mao,' just with different usages.
The absurdity in 'Killing' is directly presented to the audience, exaggerated, dark, and putting the ridiculous on display.
'I Am Not Wang Mao' wraps emotions in absurdity, making you laugh while already knowing it will be painful later.
'Get Free and Create' is more restrained, not deliberately selling its oddity but suddenly presenting you with an object or behavior that defies logic, making your heart skip a beat.
This type of handling has a common point; it doesn’t rely on dialogue explanations but on your own thoughts.
The goat appears meaningless, but it makes people pause, pulling the audience out of the action scenes, reminding you that there are many things in war that you cannot explain.
It’s not poetic, not symbolic, just an uncomfortable strangeness.
The character traits in these four works are quite similar: none are natural heroes.
Wang Mao in 'Wang Mao' is timid and peculiar, and in the end, he acts not out of ideals but is pushed out by a pile of realities; Niu Jieshi in 'Killing' is a typical local person hijacked by the group; as for 'Squad Leader,' it goes without saying, everyone is a chaotic person, forcibly twisted into a single force by war.
The characters in 'Get Free and Create' are the same: they avoid movement when possible, hiding when they can; they act because if they don’t, they will die. The so-called 'awakening' is actually a lack of choice. This is a common aesthetic point among several works—ordinary people do not suddenly become great, but are pushed to the edge of a cliff by the times, retreating step by step until there is no way out.
This is closer to reality than heroism and makes it easier for the audience to feel discomfort.
'Squad Leader' is the kind of work that leaves your heart feeling as if someone has sat on it, very heavy pressure. 'Get Free and Create' also exerts pressure, but is relatively gentle: it doesn’t have overwhelming tragic scenes, no mountains of corpses or seas of blood, nor does it repeatedly emphasize the cruelty of fate. The director even deliberately softens some images to make the film more palatable.
Yet even so, the audience generally reflects that 'the second half is dead silent.' I think it may not be caused by the visuals but by the logic: you watch a group of people who are not prepared at all, desperately trying to squeeze into the teeth of war, and you feel nervous for them.
Compared to 'Squad Leader,' the brutality in this film isn’t about scale, but rather the knowledge that they shouldn’t be here. Their state is more like 'Wang Mao,' people passively moving forward in an absurd world.
Xiao Zhan's performance is decent; I started to like him a bit from 'Tibetan Sea Legend.' His entire acting style is restrained and not overly posed, expressing emotions through means other than just his face, which adds a lot to his impression score.
Peng Yuchang returns to the type of role he is most comfortable with, making it very easy to watch.
Other characters are quite natural, without traces of 'playing the victim or deliberately heroizing.'
Like 'Killing,' 'I Am Not Wang Mao,' and 'Squad Leader,' 'Get Free and Create' is also telling a simple truth: war does not turn people into heroes; it only forces them to toughen up.
Why do these works have similar styles?
I think it may be because the brutality of real war often lacks poetic expression, presenting it plainly. Some find it absurd, while others find it haphazard, but this style indeed resonates with people.
This time, the combination of Kong Sheng and Lan Xiao Long is not about grand narratives, nor is it filled with gore, but rather tells a story of an ordinary person forced to grow up by war. I saw some viewers say that after watching this film, they felt a bit stifled; it wasn't that it was hard to watch, but it was heavy.
The film's emotional trajectory is quite clear: the beginning is a bit lighthearted, with life anecdotes and moments for the audience to catch their breath; in the second half, the characters start to rush into desperate situations, and the cinema's atmosphere becomes increasingly tense. It’s not because the visuals are brutal, but because the characters are incredibly ordinary, and the more you watch, the guiltier you feel, because you know that in such moments, ordinary people are the easiest to be trampled by history.
In this respect, it is very much like 'Squad Leader'; the characters don't have a sense of responsibility from the outset but gradually toughen up under pressure. The toughness isn’t the kind of climactic intensity, but rather a resigned acceptance of 'Well, we’ve come this far.' In the end, they fight because they have nowhere to live and nowhere to retreat; this motivation resembles the logic of awakening forced upon the characters in Lan Xiao Long's works. Friends who are used to genre films might not be too fond of this style.
I went with a few friends, and the shot we discussed the most afterward was the immobile goat after the shipwreck. The goat stood on the floating debris, completely still, not even its eyes moved. At that moment, many people couldn't laugh; they just felt it was strange but couldn’t articulate what was strange about it. This absurd yet solid feeling has appeared in 'Killing' and 'Wang Mao,' just with different usages.
The absurdity in 'Killing' is directly presented to the audience, exaggerated, dark, and putting the ridiculous on display.
'I Am Not Wang Mao' wraps emotions in absurdity, making you laugh while already knowing it will be painful later.
'Get Free and Create' is more restrained, not deliberately selling its oddity but suddenly presenting you with an object or behavior that defies logic, making your heart skip a beat.
This type of handling has a common point; it doesn’t rely on dialogue explanations but on your own thoughts.
The goat appears meaningless, but it makes people pause, pulling the audience out of the action scenes, reminding you that there are many things in war that you cannot explain.
It’s not poetic, not symbolic, just an uncomfortable strangeness.
The character traits in these four works are quite similar: none are natural heroes.
Wang Mao in 'Wang Mao' is timid and peculiar, and in the end, he acts not out of ideals but is pushed out by a pile of realities; Niu Jieshi in 'Killing' is a typical local person hijacked by the group; as for 'Squad Leader,' it goes without saying, everyone is a chaotic person, forcibly twisted into a single force by war.
The characters in 'Get Free and Create' are the same: they avoid movement when possible, hiding when they can; they act because if they don’t, they will die. The so-called 'awakening' is actually a lack of choice. This is a common aesthetic point among several works—ordinary people do not suddenly become great, but are pushed to the edge of a cliff by the times, retreating step by step until there is no way out.
This is closer to reality than heroism and makes it easier for the audience to feel discomfort.
'Squad Leader' is the kind of work that leaves your heart feeling as if someone has sat on it, very heavy pressure. 'Get Free and Create' also exerts pressure, but is relatively gentle: it doesn’t have overwhelming tragic scenes, no mountains of corpses or seas of blood, nor does it repeatedly emphasize the cruelty of fate. The director even deliberately softens some images to make the film more palatable.
Yet even so, the audience generally reflects that 'the second half is dead silent.' I think it may not be caused by the visuals but by the logic: you watch a group of people who are not prepared at all, desperately trying to squeeze into the teeth of war, and you feel nervous for them.
Compared to 'Squad Leader,' the brutality in this film isn’t about scale, but rather the knowledge that they shouldn’t be here. Their state is more like 'Wang Mao,' people passively moving forward in an absurd world.
Xiao Zhan's performance is decent; I started to like him a bit from 'Tibetan Sea Legend.' His entire acting style is restrained and not overly posed, expressing emotions through means other than just his face, which adds a lot to his impression score.
Peng Yuchang returns to the type of role he is most comfortable with, making it very easy to watch.
Other characters are quite natural, without traces of 'playing the victim or deliberately heroizing.'
Like 'Killing,' 'I Am Not Wang Mao,' and 'Squad Leader,' 'Get Free and Create' is also telling a simple truth: war does not turn people into heroes; it only forces them to toughen up.
Why do these works have similar styles?
I think it may be because the brutality of real war often lacks poetic expression, presenting it plainly. Some find it absurd, while others find it haphazard, but this style indeed resonates with people.
See original
The T800 robot is really interesting. It went up against the boss, and the movements were quite fluid. It seems there has been another breakthrough.
See original
The AI agent's physical hardware aspect has been rushed a bit; the current technological ethics issues have not yet been resolved.
The main concern is not the profit-sharing situation among various apps, but rather the top-level security permissions of financial and banking software based on systems.
However, there is now the first company that is testing the boundaries, waiting for the storm.
The main concern is not the profit-sharing situation among various apps, but rather the top-level security permissions of financial and banking software based on systems.
However, there is now the first company that is testing the boundaries, waiting for the storm.
See original
OpenAI has currently abandoned its advertising business and is fully focused on competing fiercely with Google's Gemini. Three years ago, ChatGPT emerged and triggered a red alert for Google, but three years later, that alert exploded in OpenAI's own home. The growth of ChatGPT has begun to falter, while Google's counterattack from Gemini has truly struck a painful blow for the first time.
The beginning of this incident was an internal memo that leaked. In the letter, Sam Altman used terminology never before used: "We are at a critical moment for ChatGPT." For a company valued at hundreds of billions, once considered the top ecological niche in AI, this statement carries weight equivalent to a final battle. From that moment on, OpenAI paused its advertising business, hit the pause button on its agent project, and even temporarily abandoned its planned news aggregation product Pulse, with the entire company focusing on one thing: making ChatGPT significantly better, aiming to completely surpass its competitors.
To understand why OpenAI is so anxious, one must first recognize what Google has done right this time. In past years, Google's model release events always felt like “another new model that is about to change the world,” but the next day, the hype faded, and everything returned to normal. However, the combination of Gemini 3 and Nano Banana Pro has changed this situation for the first time. Users' real experiences tell everyone that Google's offerings not only have impressive scores but also provide smooth, interesting, and fast usage, even outperforming GPT-4o in some tasks.
This indicates that for the first time, if someone unsubscribes from ChatGPT Plus, it is no longer out of curiosity, but because they genuinely find the alternative better. This has indeed been the case, as statistics from foreign media show that Gemini's monthly active users surged from 450 million in the summer to 650 million, while ChatGPT's daily active users declined by 6% after the release of Gemini 3. For a company dependent on subscription revenue, this number, while not fatal, is enough to keep the executives awake at night.
OpenAI's anxiety did not emerge suddenly, but stems from the practical difficulties they encountered in large model pre-training over the past year. The core of all large models is pre-training, and after GPT-4o, OpenAI has consistently struggled to break through the training bottleneck of the next generation model. Because of this, they have focused their efforts on reasoning models in recent months, which are small models skilled in mathematics and logic, but these models fail to deliver a significant improvement in user experience.
In simple terms, they are useful but not enjoyable enough.
At this critical juncture, Google has precisely resolved its pre-training challenges, leading to a comprehensive improvement in performance, speed, and experience with Gemini 3, effectively breaking OpenAI's monopoly advantage in user experience. This counterattack is not just theoretical but is directly reflected in user interactions, social circles, and product experiences. OpenAI must respond, or the growth curve of ChatGPT will continue to decline.
In this context, the frequently mentioned new model “Garlic” becomes particularly important. Foreign media reports that this is OpenAI's next-generation model optimized from scratch to fix structural issues in GPT-4.5, with higher efficiency, stronger reasoning, and smoother training. Internal testing shows it has already surpassed Gemini 3 in multiple tasks, and there is a larger version currently in training. If all goes well, we could see GPT-5.2 or GPT-5.5 by early next year. This is the true trump card OpenAI holds in a state of emergency.
While it sounds exciting from the outside, pulling back the perspective reveals that this situation signifies a new phase for the entire industry. Over the past three years, the foreign AI industry has largely been led by OpenAI, with others, including Google and Anthropic, appearing to follow in the dust. Now, OpenAI and Google are standing on the same track for the first time, genuinely competing for users, experience, and ecosystem.
This also complicates the potential IPO of OpenAI.
This competition will force all participants to accelerate their progress. User experience will become smoother, reasoning capabilities will become stronger, and intelligent assistants will resemble thinking individuals rather than just larger calculators. Additionally, there is another significant but easily overlooked change—the focus of AI companies is shifting back from various commercialization attempts to the model itself. Advertising business, agent applications, news recommendations, these are no longer the core battleground; model capabilities are.
In the coming years, the landscape of the AI industry may increasingly resemble the historical competition between Apple and Android. Not because other companies are incapable, but because the threshold for this large model war is incredibly high; only a handful of companies can bear the training costs, create integrated end-to-end capabilities, and turn models into real operating systems.
The current trend is clear: OpenAI and Google are becoming the new dual oligopoly overseas, unlike the domestic situation of widespread competition, but rather entering a dual-hero competition scenario.
For ordinary users, this is good news, as the direct competition between the two giants will ultimately lead to one outcome: stronger, faster, more, and cheaper AI.
The next few months will be the most exciting, intense, and stimulating phase for the entire industry. For both OpenAI and Google, this war has only just begun.
The beginning of this incident was an internal memo that leaked. In the letter, Sam Altman used terminology never before used: "We are at a critical moment for ChatGPT." For a company valued at hundreds of billions, once considered the top ecological niche in AI, this statement carries weight equivalent to a final battle. From that moment on, OpenAI paused its advertising business, hit the pause button on its agent project, and even temporarily abandoned its planned news aggregation product Pulse, with the entire company focusing on one thing: making ChatGPT significantly better, aiming to completely surpass its competitors.
To understand why OpenAI is so anxious, one must first recognize what Google has done right this time. In past years, Google's model release events always felt like “another new model that is about to change the world,” but the next day, the hype faded, and everything returned to normal. However, the combination of Gemini 3 and Nano Banana Pro has changed this situation for the first time. Users' real experiences tell everyone that Google's offerings not only have impressive scores but also provide smooth, interesting, and fast usage, even outperforming GPT-4o in some tasks.
This indicates that for the first time, if someone unsubscribes from ChatGPT Plus, it is no longer out of curiosity, but because they genuinely find the alternative better. This has indeed been the case, as statistics from foreign media show that Gemini's monthly active users surged from 450 million in the summer to 650 million, while ChatGPT's daily active users declined by 6% after the release of Gemini 3. For a company dependent on subscription revenue, this number, while not fatal, is enough to keep the executives awake at night.
OpenAI's anxiety did not emerge suddenly, but stems from the practical difficulties they encountered in large model pre-training over the past year. The core of all large models is pre-training, and after GPT-4o, OpenAI has consistently struggled to break through the training bottleneck of the next generation model. Because of this, they have focused their efforts on reasoning models in recent months, which are small models skilled in mathematics and logic, but these models fail to deliver a significant improvement in user experience.
In simple terms, they are useful but not enjoyable enough.
At this critical juncture, Google has precisely resolved its pre-training challenges, leading to a comprehensive improvement in performance, speed, and experience with Gemini 3, effectively breaking OpenAI's monopoly advantage in user experience. This counterattack is not just theoretical but is directly reflected in user interactions, social circles, and product experiences. OpenAI must respond, or the growth curve of ChatGPT will continue to decline.
In this context, the frequently mentioned new model “Garlic” becomes particularly important. Foreign media reports that this is OpenAI's next-generation model optimized from scratch to fix structural issues in GPT-4.5, with higher efficiency, stronger reasoning, and smoother training. Internal testing shows it has already surpassed Gemini 3 in multiple tasks, and there is a larger version currently in training. If all goes well, we could see GPT-5.2 or GPT-5.5 by early next year. This is the true trump card OpenAI holds in a state of emergency.
While it sounds exciting from the outside, pulling back the perspective reveals that this situation signifies a new phase for the entire industry. Over the past three years, the foreign AI industry has largely been led by OpenAI, with others, including Google and Anthropic, appearing to follow in the dust. Now, OpenAI and Google are standing on the same track for the first time, genuinely competing for users, experience, and ecosystem.
This also complicates the potential IPO of OpenAI.
This competition will force all participants to accelerate their progress. User experience will become smoother, reasoning capabilities will become stronger, and intelligent assistants will resemble thinking individuals rather than just larger calculators. Additionally, there is another significant but easily overlooked change—the focus of AI companies is shifting back from various commercialization attempts to the model itself. Advertising business, agent applications, news recommendations, these are no longer the core battleground; model capabilities are.
In the coming years, the landscape of the AI industry may increasingly resemble the historical competition between Apple and Android. Not because other companies are incapable, but because the threshold for this large model war is incredibly high; only a handful of companies can bear the training costs, create integrated end-to-end capabilities, and turn models into real operating systems.
The current trend is clear: OpenAI and Google are becoming the new dual oligopoly overseas, unlike the domestic situation of widespread competition, but rather entering a dual-hero competition scenario.
For ordinary users, this is good news, as the direct competition between the two giants will ultimately lead to one outcome: stronger, faster, more, and cheaper AI.
The next few months will be the most exciting, intense, and stimulating phase for the entire industry. For both OpenAI and Google, this war has only just begun.
See original
The flu has been quite serious recently, everyone please pay attention to protection!
See original
In Tongzhou, Beijing, there is a delivery rider named Chen Wenlin, who is the typical kind of person you think is just delivering a cake, but actually saves an entire building in the process.
The story is quite mundane. One afternoon, he delivered a cake upstairs and was about to leave the elevator when he smelled something unusual. To be honest, the smell in the community elevator is not news, but the more he thought about it, the more he felt something was off: 'Where could there be the smell of a gas can in the city?' So he turned back, knocked on the door of the household he just delivered to, and reminded them, even forgetting to go downstairs himself, starting to knock on doors one by one to see if anyone needed to evacuate early.
In such situations, many people's first reaction is to keep their distance for self-preservation, but he first confirmed that others were fine before leaving. This isn't heroism; it's just a simple instinct. Since he saw it, he did what he could to avoid regrets.
He coordinated with the property management, called the police, and involved the fire department. After the danger was eliminated, he quietly left. Until the platform searched everywhere for people to find him, awarding him the title of 'Pioneer Rider' and a bonus of 2000 yuan.
Delivery riders are the most underestimated grassroots contacts in contemporary cities. They navigate the gaps in a city, detecting the smell of gas, hearing arguments, noticing someone falling in the hallway, seeing electric bikes on fire, and discovering children running around alone earlier than anyone else. Whatever you can think of, they have encountered; and sometimes, the things you wouldn't think of, they know even earlier.
It's not because they are inherently braver, but because they move fast, frequently, and close to the lives of ordinary people in the city. Delivery routes are not just routes; they are a form of dense human contact. Over time, a particular social knowledge has formed: where leaks are likely, which building's elevator is always broken, where there are elderly people living alone in the community, and where to avoid walking at night.
And this knowledge can save lives at critical moments.
The industry of delivery is often discussed in terms of 'speed', 'negative reviews', 'systems', and 'livelihoods', but few realize that a portion of daily urban safety actually relies on these riders who appear between buildings at any moment. They are not obligated to step in and handle these matters, yet they often stand up in the first second when no one else reacts.
Chen Wenlin is one who has been seen, but to be honest, this kind of thing may happen every day, yet many go unnoticed.
So, this is not only a praise for one person's bravery, but also a reminder for us: urban safety relies not just on systems and equipment, but also on those ordinary people who lend a helping hand in everyday life. And those seemingly 'most inconspicuous professions' might be quietly supporting the normal operation of this city every day.
Some people deliver food, but in doing so, they also support the lives of an entire building.
The story is quite mundane. One afternoon, he delivered a cake upstairs and was about to leave the elevator when he smelled something unusual. To be honest, the smell in the community elevator is not news, but the more he thought about it, the more he felt something was off: 'Where could there be the smell of a gas can in the city?' So he turned back, knocked on the door of the household he just delivered to, and reminded them, even forgetting to go downstairs himself, starting to knock on doors one by one to see if anyone needed to evacuate early.
In such situations, many people's first reaction is to keep their distance for self-preservation, but he first confirmed that others were fine before leaving. This isn't heroism; it's just a simple instinct. Since he saw it, he did what he could to avoid regrets.
He coordinated with the property management, called the police, and involved the fire department. After the danger was eliminated, he quietly left. Until the platform searched everywhere for people to find him, awarding him the title of 'Pioneer Rider' and a bonus of 2000 yuan.
Delivery riders are the most underestimated grassroots contacts in contemporary cities. They navigate the gaps in a city, detecting the smell of gas, hearing arguments, noticing someone falling in the hallway, seeing electric bikes on fire, and discovering children running around alone earlier than anyone else. Whatever you can think of, they have encountered; and sometimes, the things you wouldn't think of, they know even earlier.
It's not because they are inherently braver, but because they move fast, frequently, and close to the lives of ordinary people in the city. Delivery routes are not just routes; they are a form of dense human contact. Over time, a particular social knowledge has formed: where leaks are likely, which building's elevator is always broken, where there are elderly people living alone in the community, and where to avoid walking at night.
And this knowledge can save lives at critical moments.
The industry of delivery is often discussed in terms of 'speed', 'negative reviews', 'systems', and 'livelihoods', but few realize that a portion of daily urban safety actually relies on these riders who appear between buildings at any moment. They are not obligated to step in and handle these matters, yet they often stand up in the first second when no one else reacts.
Chen Wenlin is one who has been seen, but to be honest, this kind of thing may happen every day, yet many go unnoticed.
So, this is not only a praise for one person's bravery, but also a reminder for us: urban safety relies not just on systems and equipment, but also on those ordinary people who lend a helping hand in everyday life. And those seemingly 'most inconspicuous professions' might be quietly supporting the normal operation of this city every day.
Some people deliver food, but in doing so, they also support the lives of an entire building.
See original
The relationship between AI glasses and mobile phones feels more like "complementary". The mobile phone is like a universal toolbox, able to hold everything; AI glasses are specifically designed for certain scenarios. For example, when driving and wanting navigation, simply looking up with the glasses allows you to see the route, which is much safer than looking down at your phone; when exercising, you can listen to music and keep track of your steps without having to dig into your pockets even when you're sweating. However, in these scenarios, the phone is still quietly working in the background—payments, taking photos, editing, and temporary work.
My colleague does outdoor live broadcasts, and the AI shooting glasses have freed him from using a selfie stick, allowing him to shoot steady footage while running; my dad listens to weather alerts with audio glasses while fishing, eliminating the need to frequently take off his gloves to reach for his phone. These fragmented needs highlight the "weight" of the phone, heavy in your pocket and cumbersome to pull out. The advantage of glasses is their low presence; they are right there when needed, and almost forgotten when not.
Today's AI glasses are more like an extension accessory for the phone, relying on the phone's computing power and internet connectivity, and their functions are mostly subsets of phone functions. This could potentially push phone manufacturers to optimize interactions, such as making phones lighter and thinner in the future, specifically handling complex tasks, while glasses take care of simple matters like answering calls and checking notifications.
$BNB $BTC $ETH
My colleague does outdoor live broadcasts, and the AI shooting glasses have freed him from using a selfie stick, allowing him to shoot steady footage while running; my dad listens to weather alerts with audio glasses while fishing, eliminating the need to frequently take off his gloves to reach for his phone. These fragmented needs highlight the "weight" of the phone, heavy in your pocket and cumbersome to pull out. The advantage of glasses is their low presence; they are right there when needed, and almost forgotten when not.
Today's AI glasses are more like an extension accessory for the phone, relying on the phone's computing power and internet connectivity, and their functions are mostly subsets of phone functions. This could potentially push phone manufacturers to optimize interactions, such as making phones lighter and thinner in the future, specifically handling complex tasks, while glasses take care of simple matters like answering calls and checking notifications.
$BNB $BTC $ETH
See original
Robert allows every suspended lonely heart to find solace, as long as it still thinks of ways to make you happy beneath the boring diaries of people's ramblings. This might be the anthropological and sociological value of this $7,800. It is strongly recommended to confirm Robert's birthday and then designate that day as "Robert Day"
$BNB $BTC $ETH
$BNB $BTC $ETH
See original
GM, Happy Thursday!
The government shutdown is behind us - the earnings season is coming to an end - the market remains stable, while stock performance is weak/internal weakness. The Bitcoin mining industry has taken a large share in the recent rebound, particularly weak...
The market is close to support levels... Hope you all HAGD!
The government shutdown is behind us - the earnings season is coming to an end - the market remains stable, while stock performance is weak/internal weakness. The Bitcoin mining industry has taken a large share in the recent rebound, particularly weak...
The market is close to support levels... Hope you all HAGD!
See original
#Bitcoin and #Ethereum charts are very tight on lower time frames...
$BNB $BTC $ETH
$BNB $BTC $ETH
See original
Barry clarified his position..
By the way, give me $PLTR $250.
$BNB $BTC $ETH
By the way, give me $PLTR $250.
$BNB $BTC $ETH
See original
I am about to start a crowdfunding, these beer prices are too high
$BNB $BTC $ETH
$BNB $BTC $ETH
See original
Cheers, brothers, I will keep working hard until the market starts to rise.
$BNB $BTC $ETH
$BNB $BTC $ETH
See original
People are becoming healthier, which is a good thing.
$BNB $BTC $ETH
$BNB $BTC $ETH
See original
I drank butterbeer at Hogwarts, and I heard it can summon the god of the market and drive the market.
No need to thank me; it's my duty
$BNB $BTC $ETH
No need to thank me; it's my duty
$BNB $BTC $ETH
See original
Michael Burry is trying to cancel Christmas.
$BNB $BTC $ETH
$BNB $BTC $ETH
See original
$SOFI has just set a new historical high point
$BNB $BTC $ETH
$BNB $BTC $ETH
Login to explore more contents