DEV Community: Jensen King

DeepSeekMath-V2: How far are we from true AGI when AI learns to self-negate?

Jensen King — Fri, 28 Nov 2025 11:54:06 +0000

Just today, DeepSeek has just released this 19-page paper titled DeepSeekMath-V2. At this particular juncture, reading this paper evokes a sense of solemnity and composure.

Foreign media have reported that in order to bypass the ban and obtain Nvidia's high-end chips, several Chinese tech giants are having to go abroad and transfer their AI model training computing power centers to overseas data centers. This is a "computing power exodus" for survival.
However, under such a suffocating siege of computing power, DeepSeek demonstrated another possibility of "looking inward" in its paper: If external computing power is restricted, then we will evolve the inference efficiency to the extreme.

That score that surpassed the limit of human capabilities

Let's look at the result, because this result is the best reward for long-termism.
In the recently concluded 2024 Putnam Mathematics Competition, DeepSeekMath-V2 scored 118 out of 120 points.
This is the most prestigious undergraduate mathematics competition in North America, with an extremely high level of difficulty.
And the highest human score in this competition was only 90 points.
In both IMO 2025 (International Mathematical Olympiad) and CMO 2024, it achieved top-level gold medal results.

What is even more ironic is that in the comparative test, it completely outperformed Google Gemini 2.5 Pro, which has top-level computing power support, as well as OpenAI's GPT-5-Thinking-High.

How did it acquire the skill of being able to fight both left and right simultaneously?

Why is DeepSeekMath-V2 so powerful?

Refuse to provide the answer. Please show me the process instead.

Traditional AI training is like training a puppy: you give it an arithmetic problem, and as long as the final answer is correct, you give it a bone.
But there is a major flaw here: Just because the answer is correct doesn't mean one has truly understood.
AI might simply guess randomly, or use incorrect logic to arrive at the correct numbers.
DeepSeekMath-V2, even when dealing with open-ended questions without a standard answer, no longer gets obsessed with the result.
It introduced an extremely strict marking teacher who used a magnifying glass to scrutinize each derivation step and give marks accordingly: 1 point for logical rigor, 0.5 points for minor flaws, and 0 points for completely fabricated content.

Who will supervise the exam graders?

But here is an even more interesting question: If that marking teacher, that is, the AI itself, would it pretend to understand when it doesn't, or create unnecessary problems on its own?
To solve this problem, DeepSeek developed a Meta-Verification (meta-validation), which is essentially a check on the checker itself.
This is like giving the examiners an additional supervisor. The supervisor's job is not to do the questions, but to check whether the examiners' grading is reasonable:
"You pointed out that this step is wrong. Is it really wrong?"
"You gave a perfect score, but there were obvious jumps in your performance!"

Self-reflection in a split-personality manner

What's even more interesting is that DeepSeek has adjusted the reward mechanism: even if you make a mistake in the question, as long as you accurately point out where you went wrong, you can still get a high score!
This forces the model to conduct numerous internal deductions, self-denials and corrections before it can output the final answer.
It no longer acts with blind confidence, but has learned to be skeptical.

"Problem Solvers" and "Mathematicians" on the Road to AGI

Why did DeepSeekMath-V2 manage to outperform in the context of limited computing power? Because it completely revolutionized the way AI thinks.
Let me give an example. Currently, the evolution of AI has two paths:

The top-performing problem solvers

The previous large-scale models (including many of the current state-of-the-art ones) were essentially top-notch problem solvers.
It is like the student with the best memory in that class, having an extensive database of questions in his mind. When faced with a problem, it quickly retrieves the memory and answers it instantly through pattern matching.
Its limitation is that it can only solve problems that have been encountered or are similar to those encountered before. It is consuming the existing knowledge. Once it encounters completely new and unfamiliar problems outside the question bank, it will start to fabricate randomly.

A True Mathematician

DeepSeekMath-V2 forced itself to evolve into a serious mathematician.
Mathematicians do not rely on rote memorization. When confronted with an unknown conjecture, mathematicians possess the "meta-capability of learning and reasoning".
It will pause to think carefully, make step-by-step deductions, and during this process, create new knowledge that does not exist in the training data.
The paper reveals the source of this ability - Self-Verification.
DeepSeek trained an extremely rigorous validator that not only scored the results but also meticulously scrutinized the derivation process for any flaws.

Proof Generator
Responsible for proposing bold solutions to problems.

Proof Verifier
Taking on the role of a stern mentor, he would coldly point out: "This step involves a logical leap. Please start over."

Meta-Verification
Even the model would reflect on the thought, "Was the mistake I made just a mistake?" 。

Great things often emerge from scarcity.

Great things often emerge from scarcity. I have always believed that resource scarcity is the catalyst for innovation, while abundant resources tend to breed mediocrity.
While the outside world was discussing how to transport GPUs and how to circumvent the ban, DeepSeek chose to train the AI in the uncharted territory of algorithms, enabling it to learn self-denial and slow thinking.
DeepSeekMath-V2 is not only a mathematical model; it is also a metaphor.
It tells us that there is more than one path to AGI. If that wide road, paved with NVIDIA graphics cards, is blocked, then like mathematicians, using extreme logic and rationality, and carving out a small path in the wilderness, perhaps one can go further.
In this era of rapid change, learning to think slowly like a mathematician is not only the evolutionary direction of AI, but also the proper attitude that every technology professional should adopt.

Would you like to learn more about the practical implementation methods of DeepSeek and cutting-edge AI tools? Click here to enter the first-class cabin of the AI future: https://www.deployai365.com/

This might be the most comprehensive review of productivity uses for the Nano Banana Pro that you can find.

Jensen King — Tue, 25 Nov 2025 03:58:22 +0000

Introduction

I once thought that AI drawing had reached its golden age. It wasn't until I deeply experienced the three core upgrades of Nanobanana Pro that I realized this might be the critical point for AI drawing. It's no longer just a paintbrush; it has gained logic, knowledge, and literacy.

(If you have used up the official limit, there is a quick experience link at the end!)

High-Fidelity & Consistency

Finally, AI can be integrated into the workflow.
Previously, we used AI for graphic design, similar to buying lottery tickets.
Now, we use it for graphic design, just like invoking a disciplined and memory-based brand visual asset.
This has liberated all industries that require a unified visual identity, from fashion, architecture to brand IP operations. The efficiency has increased by more than ten times.
What it gradually replaced was traditional photography and 3D rendering content.

Official example

Example 1: A family portrait poster for products with multiple SKUs

Directly replaces the high-cost photography and post-production synthesis for multi-SKU products.
Achieves physical fidelity for complex materials such as glass refraction and plastic matte finish.
Improves the efficiency of commercial rendering to minutes, with production starting immediately upon delivery.

Professional studio photography of a cosmetic set including a tall glass toner bottle, a white plastic tube, and a round cream jar arranged on a wet stone surface. Soft daylight, tropical plants in background. Ensure accurate light refraction on glass vs. matte finish on plastic. High resolution, 4k.

Example 2: Interior Design

It achieves dual control of structural logic and object fidelity, accurately reproducing the floor plan and integrating specified furniture.
Designers can quickly switch styles and furniture, achieving "what you see is what you get" customized solution delivery.

Generate a photorealistic 3D interior design rendering based on the layout structure of the provided Floor Plan (Image A). Perspective: Eye-level view from the entrance looking into the living room. Furniture Integration: Strictly populate the room using the specific furniture items from the reference images. Style & Atmosphere: Classic Bauhaus style. Emphasize geometric harmony, functional aesthetics, and industrial materials like tubular steel and glass. Use a neutral background palette (white/grey) with bold geometric accents in primary colors (red, yellow, blue) for the rugs or wall art. Clean lines, open space. High-end architectural visualization.

Example 3: Three-view Diagram of Character

The identity locking and structure control of the characters have been achieved, with zero errors in multiple-angle details.
The generated results can directly serve as a reference blueprint for 3D modeling, significantly accelerating the concept art process.

Character sheet of a Cyberpunk female warrior with neon blue hair and tactical armor. Three views: Front view, Side view, and Back view. Consistent costume details, consistent facial features, neutral lighting, gray background. Concept art style.

Example 4: Rapid Construction of Conceptual Scenarios

Achieved a creative integration of multiple inconsistent elements such as background, main subject and special effects.
Increased the output efficiency of complex visual storyboards by a hundred times, suitable for rapid proposal in the early stage of film and television production.

Based on the given image, the scene is synthesized to show a corridor setting. The top light shines on the girl squatting in the middle, and directly above her behind is a huge soul. The picture has a grainy style, with an overall color tone of blue and green, low saturation, a dim scene, surrealistic effect, and a natural integration effect.

World Knowledge & Real-Time Integration

Previously, no one had ever thought that AI-generated images could also utilize reasoning models and internet search.
Now, AI truly behaves like a knowledgeable person when observing the world.
It can see the internal structure, water and electricity pipelines, and decorative finishes from the facade of a building.
It can discern the process from planting, harvesting, roasting, to hand-brewing from a cup of coffee.
It can search for relevant information on the internet and generate images.
Essentially, it saves the time you would spend disassembling the principle, conducting internet search, and constructing prompts.
It addresses the issue of the disconnection between information search, output, and visualization.

Official case

Example 5: Educational Science Popularization

It has completely solved the problem in teaching content creation where it is difficult to balance professional knowledge and visual design.
AI demonstrates a strong cognitive logic and rendering accuracy regarding historical eras, physical formulas, and mathematical symbols.
The generated results can be directly used for textbook or courseware production, significantly improving the efficiency of knowledge dissemination and academic publishing.

A horizontal timeline infographic illustration showing the major dynasties of China in chronological order (Han, Tang, Song, Ming, Qing). Feature iconic architectural elements for each era (e.g., Great Wall for Ming). Accurate text labels in Chinese. Hand-drawn educational style.

Generate a clean, high-resolution educational infographic in vector art style. Use a distinct color palette to differentiate force and acceleration vectors. The image must illustrate Newton's Second Law of Motion. Clearly depict the object with a constant mass (m), showing the relationship between Net Force ($\Sigma F$) and Acceleration (a). The acceleration vector must be exactly in the same direction as the Net Force vector. A block with mass (m) sliding down a frictionless incline plane at a 30-degree angle. The diagram must show the force of gravity ($F_g$), the normal force ($F_N$), and the resulting Net Force and Acceleration vectors pointing down the slope. Label all forces with clear vector arrows (e.g., $F_{Push}$, $F_{Friction}$, $F_g$). Ensure the formula $$\Sigma F = ma$$ is prominently displayed and perfectly legible. The acceleration vector should be easily distinguishable (e.g., a green arrow).

Example 6: Scientific Illustration

The huge challenge of converting multi-dimensional and complex scientific theory texts into integrated 3D mechanism diagrams has been overcome.
AI has demonstrated strong spatial logic and cognitive abilities.
The generated results have a publishing-level vector graphic style and accuracy, significantly accelerating the process of scientific mapping and publication.

Analyze the attached text (or document) detailing the discovery of a 'Airspace structure establishment.'
Task: Generate a high-resolution core schematic diagram. The spatial structure of the reference diagram is integrated into a 3D perspective view, placed in an illustrative urban environment. From this single image, one can clearly see the six different division methods at different positions.
Style & Content: Clean, minimalist vector diagram style (suitable for publication). Use clear, color-coded components . Crucially, all arrows and labels must be sharp and scientifically accurate. 4K resolution.

Native Text Rendering & Localization

He solved the last problem of AI image captioning, obtaining the ultimate control over the text.
Solving the problem of text display has always been a major challenge for AI image generation.
Now it can well support multi-language generation.
He gradually replaces the work of PS and translation, strengthening the ability of individual combat.

Official case

Example 7: Translation of Cross-border E-commerce Products

It has completely solved the problem of seamless localization of product marketing graphics and text in cross-border e-commerce.
The AI automatically translates and perfectly replaces the text, while accurately retaining the design, layout and font style of the original image.
It has increased the production efficiency of visual assets for global e-commerce by a hundred times, accelerating the speed of products entering the international market.

Translate all visible text in the attached e-commerce image from Chinese to English.
Crucial Requirement: The output image must maintain the exact original perspective, lighting, and background of the original image. The new translated text must perfectly adhere to the surface texture, curvature, and placement of the object

Example 8: Poster Design

The problem of integrating the artistic illustration style with the large amount of structured Chinese text information in the event poster has been solved.
AI has achieved native rendering and precise layout of long texts in complex backgrounds, ensuring that all key information is clearly readable.
The efficiency of graphic design and information communication has been combined into one, directly producing high-visibility event promotional materials.

Create a poster for an event with a tech-inspired theme. The main color scheme should be dark green and black, with a touch of light green elements. The style should be simple, modern, professional and clear.
Core content of the poster:

Top label: Offline Workshop | Open Only to Beginners

Main visual: An axonometric drawing of a cartoon character developing code using ClaudeCode in a room

Main title (bold and prominent): How to Install ClaudeCode in China

Subtitle: Experience Sharing Session on Installation

Key information modules:

Time: November 8th (Saturday) 15:00 - 17:00

Location: 2nd floor, 218 Room, Meihao Li Zhi Hotel, Shenzhen South District Science Park (Exit from the High-tech Central subway station A or B)

Event description: Specifically designed for users who have not used ClaudeCode before, to help quickly master installation and usage skills

Event flow: 10-step process (introduction at the beginning → self-introduction → introduction of ClaudeCode → installation configuration → on-site practical operation → case presentation → prompt word organization → work release → presentation → group photo)

Participation benefits: Zero-base installation guidance, precise prompt word skills, practical project operation, one-on-one problem-solving

Bottom: Scan the QR code to register (with a WeChat QR code style, indicating "Scan to add friend and note 'Event' to register") Visual elements requirements: Use simple icons (calendar, location, information, process, gift) to correspond to each module. The text hierarchy should be clear, and the overall combination of technological feel and practicality should be achieved. The key information should be arranged either on the left or right side of the page, and should not occupy a large area of the graphic space. Most of the graphic areas should contain the main visual content.

Use Case 9: Front-end Design

The problem of quickly generating UI concepts and high-fidelity sketches in product design has been solved.
The AI has perfectly followed the design language standards of iOS and other platforms, achieving pixel-level precise interface rendering.
The early working efficiency of UI designers has been increased by tens of times, accelerating the product iteration cycle.

Mobile app UI design for a travel booking application. Clean, modern interface. Screen shows a card with a photo of Paris, and a clear button labeled 'Book Flight' and price '$599'. iOS design language.

Example 10: Store Material Design

The problem of quickly implementing the brand logo onto various physical materials and scenarios has been solved visually.
AI has achieved perfect matching of the logo with different materials (such as fabric and paper) in terms of perspective, light and texture.
Designers can instantly generate high-quality brand application display images for proposals and material sample production.

A branding mockup set on a wooden table. Includes a canvas tote bag with a minimalist coffee logo, a business card, and a trifold brochure. Consistent visual identity. Warm, inviting coffee shop vibe.

Generate a scenario display diagram for the generation of this package, with four scenarios per group.

At last

We are at a critical point in history. The development of AI drawing has undergone a great leap from novelty (being able to draw something) to artistry (being beautifully drawn) and finally to practicality (being able to solve problems).
The upgrade of Nanobanana Pro focuses not on making it draw more beautifully, but on finally making AI a powerful tool that can be trusted, delivered, and integrated into business workflows.
If you are still using the traditional workflow, then you are leaving the opportunities to those innovators who are brave enough to combine logic, knowledge, and art in a package. We are at a critical point in history. The development of AI drawing has undergone a great leap from novelty (being able to draw something) to artistry (being beautifully drawn) and finally to practicality (being able to solve problems).
The upgrade of Nanobanana Pro focuses not on making it draw more beautifully, but on finally making AI a powerful tool that can be trusted, delivered, and integrated into business workflows.
If you are still using the traditional workflow, then you are leaving the opportunities to those innovators who are brave enough to combine logic, knowledge, and art in a package.

Towards AI-driven new workflows: Immediately go and obtain the low-cost API KEY for Nanobanana Pro, and together move the toolbox to https://www.deployai365.com/zh/dalle?model=AI3/nano-banana-2! (Our price is one fifth of the official price. Come and experience it now!)

Merchants were driven crazy by AI-generated image refunds, but this time Nano Banana Pro really came to the rescue.

Jensen King — Fri, 21 Nov 2025 11:26:14 +0000

To be honest, this year's Double Eleven made me experience a kind of "physiological fear" towards AI for the first time. It wasn't because AI took someone's job or because Skynet woke up. It was due to the phenomenon of AI image editing for no-reason returns that emerged in the e-commerce circle. After receiving the normal goods, buyers used the AI redrawing function to add fake traces of damage or defects to the goods in just a few seconds and then applied for a no-reason return. The merchants were speechless in the face of those incredibly realistic photos, and even when the platform intervened, it could only rule in favor of the buyers.

Technology was supposed to be a ladder, but some people have turned it into a scythe.
While we were still marveling at the generative capabilities of AI, Pandora's box had actually been opened long ago.
In a world where what we see is no longer the truth, trust has become the most expensive luxury.
Just when everyone thought this cat-and-mouse game was unsolvable, Google suddenly played its trump card.

Just now, Nano Banana Pro was officially launched.
Here's a summary of the updates:
The official name of Nano Banana Pro is Gemini 3 Pro Image Preview.
It supports generating 1K, 2K and 4K resolutions, and supports various common aspect ratios.
It supports multi-language long text image generation, multiple image editing rounds, and interleaved image-text generation.
It supports real-time acquisition of the latest knowledge through search, and can merge up to 14 input images into 1 output image.
Each generated image comes with a SynthID digital watermark.
After seeing this news, I suddenly thought that this is definitely the current AI security ethical understanding solution.

What is SynthID digital watermarking?
It means that the image content generated by Google AI in the future is already marked with a watermark at the underlying level.
Although the watermarks that we can see with our naked eyes can be removed by upgrading to a premium membership,
the SynthID that exists between the pixels will not disappear. You can ask Gemini about it, and he can still accurately tell you that this is an image generated by AI.

Whether it's changing the color, applying lossy compression, adjusting the size, or even performing local cropping on the image, as long as the core structure of the image remains intact, it can be detected.
There is no need for manual intervention. The code teaches itself. At this moment, technology is no longer an accomplice of the evil-doers, but has become the goalkeeper.

Why is this considered a significant update in AI ethical security?
The release of Nanobanana Pro is not merely a functional update; it sends a signal to the entire industry that the era of unbridled AI has come to an end.
First, this is the "good currency" responding to the "bad currency"
Over the past year, we have witnessed too many instances of AI-generated fakes: fake photos of Trump's arrest, fabricated war ruins, and those "one-click undress" features that make ordinary people look like they're naked. Content creators are running naked, and businesses are running naked.
The emergence of Nano Banana Pro is like issuing identity cards to genuine content in the vast ocean of the internet.
Secondly, it redefines "seeing"
Previously, when we looked at pictures, we focused on composition and lighting. In the future, when we look at pictures, we might have to go through a verification process first. This may sound a bit sad, but it is a necessary pain that we must endure.
The popularization of models like Pro will make traceability a standard feature of AI content. AI content without watermarks may find it difficult to move forward in the future, just like people without identification documents.
Third, the concretization of technical ethics
People always talk about AI ethics, thinking it's just a philosophical argument happening in a meeting room. But Nanobanana Pro tells you that ethics is code. This time, Google directly incorporated security mechanisms into the underlying model - that's exactly right.
Making a gun better doesn't make you a capable person. Designing an insurance bolt when making a gun is what a top-notch manufacturer should do.

We are at the eve of a new era of trust reconstruction.
Of course, as the power of the magician increases, the power of the fraudster also grows. The Nanobanana Pro cannot completely eliminate all fraud. There will definitely be hackers who will study ways to remove these digital watermarks.
This is the fate of the technological world - an eternal battle of offense and defense.
But at least, we can see the light at the end of the tunnel.
For developers and entrepreneurs, this is even more of a huge opportunity.
When authenticity becomes scarce, business models centered around content verification, copyright protection, and the deployment of trustworthy AI will experience an explosion.
We are standing at a critical point: on the left hand side is the boundless creativity brought by AI, and on the right hand side is the chaos and nothingness brought by AI.
The Nanobanana Pro is like a nail, trying to pin the running AI to the "controllable" wall.
As for whether this nail can be firmly fixed, it not only depends on Google, but also on each of us users - how to choose and deploy these technologies.
Technology has no good or bad, but the people who use it do. Fortunately, this time, technology itself is on the side of light.

Do you want to know how to quickly deploy AI models with security features like Nanobanana Pro in your business? Or do you want to learn more about practical solutions for implementing enterprise-level AI security?
Don't leave your AI to fend for itself. Click here to take action now: https://www.deployai365.com/

Two AI philosophies: Gemini 3 Pro, born to understand the complexity of the real world

Jensen King — Thu, 20 Nov 2025 11:31:24 +0000

Gemini 3 Pro

I glanced at that evaluation form. I rubbed my eyes. Took a sip of coffee. Then looked at it again. I really thought it was because I stayed up too late and my eyes were blurry, causing me to have hallucinations.
The truth is harsh and cold: The era of squeezing toothpaste is over. While we were still foolishly waiting for OpenAI's next update, Google dropped a nuclear bomb directly. Gemini 3 Pro is not just an upgrade; it is a complete betrayal of our known timeline of AI evolution.

Superior specifications: Not just an upgrade, but a complete replacement.

On November 18th, the Gemini application, search AI mode, AI Studio, Vertex AI, and the new Google Antigravity development platform were launched. The essence of Gemini 3 Pro is not the addition of a logical structure to GPT-5, but rather a general engine that attempts to simulate the entire world in a brute-force manner.
Its underlying philosophy has undergone a qualitative change. Unlike OpenAI's obsession with long chains of logic and artificial rules, Google chose the "simulation + rules" approach. The model is designed to capture the unspoken patterns in the real world, not just understanding words, but like AlphaFold predicting proteins and WeatherNext predicting the weather, it uses neural networks to fit this chaotic, irrational, but real physical world. This enables it to have absolute dominance in fields that require contextual understanding, such as humanities, society, and history.
Google has restructured this powerful model capability into three core scenarios, each of which hits the current AI's pain points:

Learning —— The All-Modal Learning Devourer

It comes with a standard 1 million token context window, which means you can simply throw an entire textbook or a video tutorial that lasts for several hours directly at it. Its visual and auditory comprehension capabilities have reached an extreme level. It not only understands videos but can also translate handwritten recipes and even directly analyze complex video tutorials and generate interactive learning materials.

Building —— The front end is more like magic rather than code

In the WebDev Arena arena, it achieved a high score of 1487 Elo. This is where Gemini 3 excels. It is skilled at generating complex and interactive web interfaces with zero samples, such as rendering more rich interactive web interfaces, rather than simply spitting out a bunch of HTML code.

Planning —— The Core of the Agent

It performed exceptionally well in long-term planning benchmark tests, and it is no longer the chatbot that just takes one step at a time. It can navigate complex multi-step workflows, such as booking services or organizing the inbox, which involve tasks that require memory, reasoning, and decision-making simultaneously. These are now its strong suits.

Furthermore, in response to the doubts raised by the outside world regarding logical reasoning, Google has also launched an enhanced reasoning mode called "Gemini 3 Deep Think", which will be made available to Ultra subscribers within a few weeks. This compensates for the shortcomings of the simulation-based models in pure logical reasoning, ensuring that the model, while maintaining a good taste in Vibe, does not fail in hard logic.

Performance Analysis: Layered Leadership

Let's talk about those exaggerated figures. Almost all the indicators show a discontinuous lead. Let's just mention a few interesting ones. Let's start with the final exam for humanity.

Humanity's Last Exam

This is a well-known and widely recognized benchmark set in the AI industry, used to evaluate the general knowledge acquisition, interdisciplinary reasoning, and complex context understanding capabilities of large language models. The questions are often open-ended, interdisciplinary, highly context-dependent, and contain common-sense traps. It is called "the last exam for humans" because its goal is to test whether the model can handle the most difficult and least formalizable parts of the human knowledge system, which require wisdom and experience.

In this test, GPT-5.1 achieved 26.5%, Claude Sonnet got 4.5% which was 13.7%, while Gemini 3 Pro, after enabling the tool usage, directly reached 45.8%. This is not a victory, it's a massacre. This proves the victory of the world model approach at this stage. It's not just about solving the questions; it's about using the massive parameter quantity to understand the underlying real logic of the questions.

MathArena Apex

This is a highly specialized and extremely challenging benchmark test set for mathematical reasoning. It is specifically designed to evaluate the ability of large language models to handle complex, multi-step, and highly abstract mathematical problems. The questions usually require more than ten steps of complex reasoning, and they are the ultimate test of long-chain logic.

GPT-5.1 only received a score of 1.0 (yes, exactly 1 point), while Claude Sonnet scored 4.5 and got 1.6 points, and Gemini 3 Pro managed to score 23.4 points. It's like watching a primary school student still counting on their fingers to do addition and subtraction, while the college student Gemini 3 has already started building a rocket by hand using calculus right beside them.

ScreenSpot-Pro

This is an emerging and extremely challenging benchmark test, specifically designed to evaluate the understanding and practical operation capabilities of large language models regarding graphical user interfaces. It is also known as the GUI Grounding (interface landing) test. The model needs to be able to recognize and understand the meanings of various buttons just like a human, and also needs to operate accurately in a high-noise environment of various advertisements. For LLMs that are proficient in text and code, understanding the graphical interface is a huge challenge.

The GPT-5.1 scored 3.5%, while the Gemini 3 Pro reached as high as 72.7%. This is the most alarming data. The Gemini 3 Pro is an extremely powerful AI Agent when it comes to operating computers. It is the sole true god in the field of AI Agents. It can almost accurately identify every button, every icon, and every line of text on the screen.

Weakness

Although Gemini 3 Pro has an extremely powerful front-end coding capability, in terms of general software engineering skills, it is still slightly inferior to GPT-5.1 and Claude Sonnet 4.5 at present. It is a top-level designer, but not yet the strongest backend architect.

Application Evaluation

To verify whether it is just empty talk or a genuine god, we conducted some more challenging practical tests in GOOGLE AI STUDIO.

Scene 1: Analyzing the paper and generating an interactive learning page

I became a bit more cautious. So I decided to first hand over the paper to the AI, interacted with it for a round, and after it helped me construct a sufficiently complete prompt, I then went to the AI STUDIO to generate the front-end page.
The result was astonishing...

The UI layout of Aesthetic Online is completely on par with the UI of major brand product pages (as shown in the picture above). Compared with the report interpretation I did using minimax-M2 before (as shown in the picture below), there is a significant qualitative improvement.

This interactive 3D model page is also a feature that I specifically requested. After two rounds of adjustments, the completion rate is quite high. Basically, all the information mentioned in the paper has been structured and organized layer by layer.

The information recovery rate of the paper is also very high, and it has presented almost all the important conclusions in a graphical form.

Scene 2: Replicating the social media homepage and creating a web page similar to a Reddit community

I made this part quite simply. I just took a screenshot of the homepage of Reddit and requested "Create a web page similar to a Reddit community. The content should be interactive and replicate the basic functions of the community."

The completion rate is also quite high, with basic comments, likes, and AI-generated simulated information. What surprised me the most was that he added an AI comment function for me, which was completely beyond my expectations. Compared to previous models, there has also been a significant improvement.

Scene Three: Article Cover Maker

After finishing my articles, I often struggle with creating the covers. I haven't had a systematic series of cover designs, and manually creating them in Photoshop is extremely time-consuming. Originally, I had planned to prepare the same set of prompts, and use the same set of prompts each time to maintain the consistency of the style.

Who knew he would directly create a cover editor for me, allowing me to freely modify the main and secondary titles, and also connect me with a Nanobanana background image generator? It comes with default settings of four styles and colors. The level of thoughtfulness is simply too much...

In-depth Insight: The Clash of Two Philosophies

Why is this the case? Because the two towers in the AI world have completely parted ways.
The left tower is the rationalism of OpenAI, who believe that logic is truth. They attempt to build a perfect rationality through computing power, teach models to learn artificial rules, and thereby reach AGI.
The right tower is the empiricism of Google, who take the path of simulation. Gemini 3 Pro doesn't care about whether the rules are perfect; it cares about whether they are true. It uses neural networks to forcefully fit this noisy, uncertain, and chaotic world.
If you want to solve the Riemann Hypothesis, go left and find GPT-5 Pro; but if you want to understand the jargon of forum posts, internet memes, and the social and cultural environment, Gemini is the only choice. Because the Gemini model uses violent fitting of massive unstructured data to reconstruct the context and atmosphere at that time. It doesn't answer why something happened, but simulates the feeling of what happened at that time. This enables it to have a level of delicacy that no other rational model can match when interpreting complex human narratives. It can quickly establish transfer relationships between physical principles and cooking techniques, which is the real-world feeling brought by its empiricism.
The 72.7% screen operation score of Gemini 3 Pro is actually a verdict on the era of graphical user interfaces. My prediction is: within six months, we won't need to learn how to use software anymore. The future interaction will be like this: you open Figma, not dragging layers, but say to Gemini: "Change this design to cyberpunk style and then directly release it to the test server." It will look at the screen, move the mouse, click the buttons, and complete everything.

Conclusion

We are standing at the edge of a singularity. The release of Gemini 3 Pro marks the official arrival of the AI Agent era. This is not just a better chatbot; it is a digital avatar capable of operating your computer for you.

We are Deploy AI, helping you develop and deploy your AI application demo. Now you can visit deployai365.com. We have prepared the complete Gemini 3 Pro API interfaces and Agent deployment environment for you.

How AI-Powered Deployment Platforms Are Revolutionizing Software Development in 2025

Jensen King — Wed, 19 Nov 2025 08:40:18 +0000

Introduction: The New Era of Software Development

The software development landscape is undergoing a seismic shift in 2025. With AI-powered deployment platforms emerging as game-changers, the traditional barriers to launching commercial applications are crumbling. According to recent industry reports, the global AI platform market is projected to reach $108.96 billion by 2030, growing at a CAGR of 10.8%.

This transformation is particularly evident in how developers and businesses approach application deployment. Platforms like DeployAI365 are pioneering a new paradigm where anyone can convert open-source projects into commercial applications with minimal technical expertise.

The Rise of No-Code and Low-Code AI Platforms

Market Growth and Adoption

The numbers tell a compelling story. By 2025, 70% of new enterprise applications will be built using no-code or low-code platforms, a dramatic increase from less than 25% in 2020. This explosive growth reflects the industry's response to:

Developer talent shortages: A global scarcity of skilled developers is driving organizations to empower non-technical staff
Accelerated digital transformation: Companies face mounting pressure to modernize processes swiftly
Increased demand for agility: Businesses need to pivot quickly in response to changing market conditions

The AI Platform Market Landscape

The AI platform market stands at $65.25 billion in 2025, with software components leading with 71.57% revenue share. This dominance is driven by enterprises favoring integrated development environments that bundle data, model, and orchestration capabilities.

Key trends shaping AI development in 2025 include:

Integration of generative AI capabilities for faster prototyping and development
Enhanced collaboration features supporting cross-functional teams
Robust evaluation frameworks to ensure AI system quality and reliability
Low-code/no-code interfaces that democratize AI development

How AI-Powered Deployment Platforms Work

The DeployAI365 Approach

Platforms like DeployAI365 have simplified the deployment process into three straightforward steps:

1. Paste GitHub URL
Simply paste any GitHub repository URL that you want to transform into a commercial product.

2. Customize & Configure
Add monetization features, customize branding, and configure deployment options to suit your business needs.

3. Deploy & Profit
One-click deployment to cloud infrastructure, with built-in payment processing and user management.

Real-World Applications

The platform showcases several powerful use cases:

Depcoder: An AI coding development tool inspired by Cursor, built on VSCode and Cline
Deepsite: AI-powered website generators powered by GPT-5, DeepSeek V3.1, and Qwen3-Coder
API Services: Instant deployment of API services with monetization built-in

The Business Case for AI-Powered Deployment

Speed to Market

Applications can be developed and deployed up to 90% faster compared to traditional methods. This acceleration is critical in today's competitive landscape where time-to-market can make or break a product.

Cost Efficiency

By reducing reliance on large development teams and streamlining processes, organizations can significantly lower operational expenses. The global low-code development technology market has reached unprecedented levels, with projections showing continued strong growth.

Democratization of Development

Perhaps most importantly, these platforms are democratizing software development. Business analysts, marketers, and operations teams can now prototype AI solutions independently, dramatically accelerating innovation.

Open Source Commercialization in 2025

The Evolving Landscape

The open-source software market is experiencing remarkable growth, projected to soar from $21.7 billion in 2021 to over $50 billion by 2026—a 130% growth rate. However, 2025 is also seeing:

A reckoning for single-vendor open source models: Community-supported projects are becoming the standard
Increased focus on sustainability: Critical infrastructure projects are scrutinized for long-term viability
More AI models from open source projects: Making AI accessible to organizations of all sizes

The DeployAI365 Solution

Platforms like DeployAI365 address the commercialization challenge by:

Automatically converting open-source projects into commercial versions
Providing one-click deployment without requiring coding expertise
Integrating monetization features like payment processing and user management
Supporting multiple AI models from GPT-5 to DeepSeek and Qwen3-Coder

Industry-Specific Applications

Healthcare

Providers are using low-code tools to streamline patient management, develop telemedicine solutions, and build custom applications for improved service delivery.

Education

Schools and universities leverage these platforms to create virtual classrooms, manage e-learning tools, and improve administrative efficiency without large development teams.

Enterprise

Businesses across sectors are rapidly deploying internal tools, customer-facing applications, and automated workflows using AI-powered platforms.

The Future of AI-Powered Development

Key Predictions for 2025 and Beyond

Continued market expansion: The AI platform market will maintain its trajectory toward $251 billion by 2033
Enhanced AI capabilities: Integration of more sophisticated models and better evaluation tools
Greater accessibility: Even more intuitive interfaces and broader language support
Stronger community ecosystems: Foundation-backed projects will become the norm

Competitive Advantages

Organizations adopting AI-powered deployment platforms gain:

Faster innovation cycles: Test ideas quickly and deploy with minimal developer involvement
Reduced technical debt: Modern, cloud-native architectures by default
Scalability: Enterprise-grade tools that grow with business needs
Enhanced collaboration: Better alignment between IT and business teams

Getting Started with AI-Powered Deployment

Best Practices

Start small: Begin with a pilot project to understand the platform's capabilities
Leverage templates: Use pre-built templates and components to accelerate development
Focus on user experience: Use AI to enhance, not replace, human-centered design
Monitor performance: Implement robust evaluation frameworks from day one
Plan for scale: Choose platforms with enterprise-grade features and support

Platform Selection Criteria

When evaluating AI-powered deployment platforms, consider:

Model variety: Support for multiple AI models (GPT, DeepSeek, Claude, etc.)
Deployment ease: True one-click deployment vs. manual configuration
Monetization features: Built-in payment processing and user management
Customization options: Ability to brand and configure applications
Community and support: Active user community and comprehensive documentation

Conclusion: Embracing the AI-Powered Future

The convergence of AI, no-code platforms, and open-source software is creating unprecedented opportunities for developers and businesses alike. Platforms like DeployAI365 are at the forefront of this revolution, making it possible for anyone to quickly create, launch, and monetize applications with AI.

As we progress through 2025, the question is no longer whether to adopt AI-powered development tools, but how quickly you can integrate them into your workflow. The market data is clear: organizations that embrace these platforms gain significant competitive advantages in speed, cost, and innovation capacity.

Whether you're a solo developer looking to monetize open-source projects, a startup validating prototypes, or an enterprise accelerating digital transformation, AI-powered deployment platforms offer a compelling path forward. The future of software development is here—and it's more accessible than ever before.

Ready to transform your development workflow? Visit DeployAI365.com to explore templates, start deploying AI-powered applications, and join the revolution in software development.

Resources

Market Reports: AI Platform Market forecasts and industry analysis
Open Source Trends: 2025 State of Commercial Open Source
Development Statistics: Low-code/no-code adoption metrics
Community: Join the growing ecosystem of AI-powered developers

AI #Deployment #DevOps #OpenSource #NoCode #LowCode #SoftwareDevelopment #DeployAI #CloudComputing

My victory settlement page suddenly crashed! It only took me a few steps of Nanobanana!

Jensen King — Mon, 17 Nov 2025 11:59:50 +0000

A casual attempt

For a seasoned player, the moment of victory settlement is the strongest motivation to share the glorious moments. Spending a fortune on a beloved hero, only to be able to share a default battle report that's exactly the same as everyone else's? That's just not acceptable!
One day, the content I randomly took and had AI modify suddenly went viral. I then realized that actually everyone had the same thought, and there were many in the comment section asking for help with the modification. So today, I'm sharing an autonomous, efficient, and no-barrier AI solution: Just four steps, and with the Nanobanana model on DeployAI, you can easily achieve top-notch modifications.

Tutorial

1.Open the Deploy AI website

(Its model performance is quite good and there are many options to choose from, so I chose this one as my first option.)

https://www.deployai365.com/

2.Enter the API Explorer page and locate the Nanobanana model.

It includes all the major applications of large models. You don't have to search around anymore. It's extremely convenient and the update speed is also very fast!

3. Upload the prompt words, pictures and other information

Upload the victory settlement page
The aspect ratio can be selected as 16:9 (for PC) or 21:9 (for mobile), or any other ratio. Sometimes it can work surprisingly well.
The prompt words I used are listed here. You can also come up with other content and let the AI optimize it.

Modify the picture
1 - Only changing the posture of the middle character, this character flashes and rushes forward from the side of the screen at an extremely fast speed. The body is pushed forward, and there is a distinct energy trace or speed line trailing behind, as if just completing a breakthrough movement. The character is very close to the camera, and its speed sense and some of the remaining traces even partially obscure the secondary characters on both sides (only keeping the edges or parts of their bodies). Her weapon points forward, and the posture is light and agile, creating a strong sense of speed and momentum.
2 - The character's clothing, hairstyle, and accessories are consistent with the style of the original image.
3 - The left and right battle records UI are naturally displayed in the screen. The original digital UI content such as numbers is not allowed to be redrawn.
4 - The picture details are rich, with an extremely high resolution.

4. Wait for your customized victory settlement page!

GPT-5.1 is here! Not only has it become smarter, but it seems... it understands you better?

Jensen King — Thu, 13 Nov 2025 09:02:27 +0000

GPT5.1

Friends, the world of AI is constantly evolving every day!
Perhaps you have just been using GPT-5 and thought it was already quite powerful. But we always wonder, besides being smart enough, can AI also understand us better and make the usage more comfortable? We not only want a super brain, but also a companion that we can chat with. Just yesterday, OpenAI seems to have heard the voices of everyone and officially released GPT-5.1!
This update is not a minor fix. The core is to make AI "more intelligent and better at chatting". In simple terms, there are several major changes.

Model upgrade

The model we use most frequently has been upgraded to GPT-5.1 Instant. It has become warmer, the conversations are more interesting, and it is more responsive to your instructions.
More importantly, it has learned adaptive reasoning. For simple questions, it responds instantly and maintains high speed; but if it encounters complex mathematical or programming problems, it will automatically "think a little more" and spend more time giving you a more accurate answer.

This chart is very intuitive. It showcases a very core new feature of GPT-5.1: "Adaptive Reasoning".
In simple terms, this chart tells us that GPT-5.1 has learned to "judge the situation" to allocate its "thinking time".
The most powerful GPT-5.1 Thinking model has also been upgraded. It now reacts faster, is more persistent when handling complex tasks, and speaks more clearly. It has less jargon, explains complex things more down-to-earth, and has a warmer and more empathetic tone.

Personality customization

This might be my most exciting feature! Now you can more easily adjust the "personality" of ChatGPT.
In addition to the previous options, there are now several new styles such as "professional", "honest", and "eccentric". Moreover, you can even fine-tune its "passion level", "conciseness", or "frequency of using emojis".

OpenAI believes that being more obedient and providing more reliable responses would be better. However, users feel that this reliability makes it more boring.
Personally, I think the updated style customization features, such as the newly added "Professional" and "Honest" modes, are actually more practical than that vague "Warm" concept.
This has made me wonder, what exactly do we want from AI?
Perhaps the concept of "warmth" and "being like a human" is inherently a subjective experience that is difficult to quantify.

The direction of AI development may be quietly changing.

The true essence of this upgrade might not lie in its greater human-like qualities, but rather in its increased controllability and reliability.
The previous AI was like a talented but somewhat crazy artist; you never knew whether its next statement would bring you joy or shock. The current GPT-5.1 is increasingly resembling a well-trained senior assistant. It may not make jokes with you, but when you give it professional instructions, it will never fail you.
AI is shifting from a creative partner to a super helper. This is neither good nor bad, but the direction is indeed changing.

GPT-5.1, who can use it now?

The official process is as follows:
Paid (Pro, Plus, Go, Business) → Free users → Unlogged users
The push will start on November 12th. If you are a paid user, you may have already received it.
But if you are a free user, you can have an early experience on Deploy AI.

https://www.deployai365.com/ApiExplorer

Your Agent is finally able to "breathe freely"! How does M2 solve the biggest cost problem for AI?

Jensen King — Thu, 13 Nov 2025 07:44:24 +0000

Question: The hurdle in Agent Economics

Agents are widely recognized as the next step in the AI industry. They accomplish complex tasks through multiple rounds of reasoning, tool invocation, and self-correction. However, the real obstacle for the large-scale implementation of Agents is not technical capabilities, but "Agent Economics": The frequent and continuous API calls made by Agents, in the face of high inference costs, have led many commercial projects to be budgeted out before they even start.

The antidote for M2: A radical overhaul of the cost structure

The release of MiniMax M2 is precisely aimed at addressing this "cost issue". It ranks among the top five globally in the authoritative list of performance, but its hybrid architecture fundamentally changes the cost structure of computing.
This is an extremely efficient "actuarial" computing philosophy: You don't have to pay for the entire redundant model; you only need to pay for the computing power activated for your current task. This structural advantage has pushed the API cost of M2 to an astonishing new low of $0.53 per million tokens, only 8% of Claude Sonnet 4.5, making it the most cost-effective model in the world. The output speed is also extremely fast, reaching 100 TPS, which means it can output approximately 100 tokens per second.

indicating how intelligent and capable it is, with higher values being better.
The horizontal axis represents the "price" of the model, showing how much it costs to use 1 million Tokens. The further to the right, the cheaper it is!
M2 offers developers the ultimate cost-effectiveness.

This chart actually tells us which AI model is currently the most popular and widely used by everyone.
You can see that the big pillar is constantly rising, indicating that the entire AI market is experiencing explosive growth. Everyone is desperately using the models.
Looking at the list below, MiniMax M2 has directly jumped to the third place! It is right in front of models like Grok and Claude from these major companies. M2 has managed to rank third, with monthly traffic reaching over 5 billion Tokens, and it is still increasing. This shows that developers really love using it.

The value of M2 is also reflected in its empowerment of the MiniMax full-mode ecosystem. As the core engine, it supports the efficient collaboration of capabilities such as Hailuo video and Speech voice. This means that the next generation of AIGC applications with humanized and low-latency features will be built on the low-cost infrastructure of M2.

How to use MiniMax-M2 for programming development?

After the configuration is completed, you can see the M2 model here. Let me take the report interpretation as an example, and the prompt is as follows:

Your role is that of a top-notch "Business Strategy Analyst". Now, you need to conduct a deep analysis of the [report type] uploaded by the user. Please follow the steps and requirements below to analyze the entire report and output it in a structured HTML format for me.
[Step 1: Summary and Key Data Extraction]

Executive Summary: In no more than 200 words, extract the three most crucial conclusions and two most significant trends from the report.

Core Data Extraction: In Markdown table format, extract the five most important quantitative data mentioned in the report (such as market size, growth rate, cost structure, key indicators), and indicate the page number or chapter of the report where the data is located. [Step 2: Multi-Dimensional Insight Analysis] Please conduct targeted analysis of the report content from the following three preset perspectives:

Investor Perspective: Analyze the three potential investment opportunities and two financial risks hidden in the report, and provide brief investment recommendations (buy, hold, or wait).

Technology/Products Perspective: Identify the three core technical barriers or product innovation points mentioned in the report, and assess their impact on the industry competition landscape.

Competitor Perspective: Identify the main competitors mentioned in the report and their market share, and predict the change in market share over the next 12 months.

The results are presented as follows: M2 accurately understood my requirements and outputted the data according to my specifications. The data is completely accurate and the presentation is clear and intuitive.

Conclusion: Transforming Creativity into Value

The door of the Agent era has been fully opened. Now, all that stands in your way is your own creativity. MiniMax M2 has already offered industry-leading performance at the lowest cost.
It's time to let go of your concerns about high costs and transform the Agent ideas in your mind into commercial value that can be operated on a large scale and sustainably!

If I have no programming experience at all, can I develop my own App or website?

Jensen King — Tue, 11 Nov 2025 10:22:44 +0000

BACKGROUND

I have to share my excitement! I finally turned a Side Project idea that had been in my head for months into reality—and I'm a complete novice who can't even tell the difference between Python and the Terminal! For non-technical users like me, the biggest pain point is that even though I know AI can write code (like the powerful Claude), using it usually involves dealing with complex Command Line Interfaces, tedious API key configurations, and frustrating deployment environments. The setup alone is enough to make me give up.

Complex Configuration and Deployment: To even begin, I need to obtain and configure an API Key , locate the correct API Full URL , identify the Model ID information , and potentially handle specific Authorization headers. For someone without a technical background, just managing this collection of complex settings and information is overwhelming enough to make me quit.
Command Line Interface Barrier: Many powerful AI coding tools still rely on the Terminal or Command Line Interface . This non-graphical operating method presents a huge learning and usability obstacle for those without a technical foundation.
High Risk of Configuration Failure: Even if I manage to find the correct URL, Key, and Model ID , the process of manually editing and saving this information is highly error-prone, and if something goes wrong, I wouldn't even know how to troubleshoot and fix the issue.

Claude Code

Claude Code is a desktop application that deeply integrates Claude's AI IDE. It is not merely a web-based packaging, but rather it creates a native, secure and fast AI programming workstation for you from the very bottom.

Native performance, ultimate speed: Developed based on the native desktop framework, this means you will experience a smoother coding experience with faster speed and lower resource usage. Say goodbye to delays, every AI interaction feels like a smooth flow.
Local-level security guarantee: Powerful local design gives you greater control over your code data. In a focused and secure environment, fully unleash your creativity.
Programming as AI, AI as programming: Deep integration of Claude's powerful intelligence provides high-quality code generation, refactoring, and interpretation. Your IDE becomes the smartest AI assistant.
Always up-to-date, one-click finished： Achieve one-click installation and automatic updates. We have solved all the cumbersome configuration and update problems for you， allowing you to always be at the forefront of technology.

Three steps to start

Step 1: Download and install the client application

Open the official designated website, find the version corresponding to your operating system, and click to download. After the download is complete, double-click the installation package and follow the prompts to complete the installation.

Official download website:Link

Step 2: Configure your Claude API Key

Obtain Key: Register a DeployAI account, visit the following website, and save the URL, KEY, and model ID information.

API KEY acquisition address:Link

Access the configuration page: When you first open Claude Code, click on "API Connect" to enter the configuration page.
Configuration and Saving: On the left side, paste your API information; on the right side, configure your API; in the upper right corner, save and restart.

Step 3: Begin your immersive AI programming

After the configuration is completed, you will immediately enter the simple yet powerful IDE interface of Claude Code:

Code Assistant: Directly access Claude from the sidebar for code questions, debugging, or code generation. -** Project Integration:** Easily import existing project folders to familiarize the AI with your code context.
Focus on Coding: Enjoy the fast response and zero-latency experience brought by native applications.
Real-time Debugging: Use the built-in preview window (such as the game interface on the right side of the screenshot) to view the effect of your code in real time.