Just now, OpenAI has evolved again.
Now, the highly anticipated ChatGPT Sites are officially open for public beta!
Even if you can't write a single line of code, you can generate a highly professional, feature-rich interactive website in just minutes—simply by using a few natural language prompts or uploading a sketch, just like writing a Google Doc.
Frontend development can now be accomplished using only natural language, causing major disruption to the traditional website-building SaaS industry.
Additionally, a new piece of news has emerged.
At the underlying technology level, GPT-6 demonstrates remarkable cross-modal capabilities.
In the absence of audio playback and with complete network disconnection, GPT-6 can identify the call of a deep-sea blue whale—and even the sound effect of a lightsaber from Star Wars—based solely on a Mel spectrogram (a visual representation of sound waveforms).
This demonstrates that AI's multimodal understanding has completely broken through human sensory barriers.
It is no longer clumsily "mimicking" human sight and hearing—it has evolved the native ability to decode the data structures of the physical world!
ChatGPT Sites is now in public beta—building websites is now as easy as writing a document.
At one time, having a top-tier personal website was a privilege reserved for programmers and professional designers.
Later, Dreamweaver emerged, followed by drag-and-drop website building SaaS tools such as Wix, Squarespace, and even Tilda.
But today, the public release of ChatGPT Sites has completely shattered the barriers to website creation.
After testing, Jeremy Caplan, editor of the renowned tech media Fast Company and Wonder Tools, said: “The experience brought by ChatGPT Sites is incredibly transformative!”
Now, creating a beautiful website is as simple as making a Google Doc.
This feature is now fully available to ChatGPT Plus, Pro, Business, Enterprise, and Edu users.
How does it work? The process is incredibly simple.
1. Describe your vision: Write a few sentences about the type of website you want and your target audience.
2. Provide input: If you have favorite webpage screenshots or hand-drawn sketches, upload them directly.
3. Specify your requirements: Do you want a link button, embedded visuals, or complex interactive elements? Just let us know.
4. Have a cup of coffee and wait: Depending on complexity, a complete website will be ready in about 10 to 15 minutes.
Note that this is no longer just "copying a template"—it's real Vibe Coding!
Early testers have already created many outstanding projects with it.
First, there are practical pages.
Someone used it to create a perfect media kit sponsorship page, a modern design guide for events in San Francisco, and even a contemporary online store for picnic supplies that supports dragging items into a shopping cart.
Additionally, artist Drue Kataoka used it to create an artistic tool called GoalFlow.
After selecting a flag, users can "paint" on the fluid simulation canvas using the flag's colors, where the pigments naturally swirl and drift like dyes in water.
Previously, creating a frontend page involving a complex fluid physics engine required senior frontend engineers to spend weeks, but now it can be done with just a few prompts.
Finally, there are physics engine games.
You can even create a Jenga-like game called Glass Towers with realistic physics in just a few lines of code, or build an arcade-style mini-game called Paper Glider where you control a paper airplane flying through hoops.
Even more alarming is its ability to iterate.
After ChatGPT provides the draft, you can directly use the "Annotation Tool" in the sidebar to circle elements on the webpage like grading homework: “Change the color of this button,” “Switch this font to sans-serif,” “Replace this image.”
It doesn’t just understand—it can seamlessly make adjustments.
Clearly, industry consolidation is inevitable.
Although early AI website builders like Lovable and Bolt still offer flexibility, and Claude's Artifacts and Claude Design excel in interactive UIs, ChatGPT Sites' standout advantage lies in its integration with an immensely extensive ecosystem.
If you work on a "ChatGPT Project," it automatically reads your brand style guide and logo library, so the generated website naturally carries your brand's DNA.
Now, frontend engineers have been reduced to “prompt product managers,” and traditional SaaS companies that still rely on “drag-and-drop website builders” to charge high monthly fees will likely not survive this winter if they don’t pivot quickly!
Application-side dimensionality reduction: eliminate all "actions," retain only "intent."
Recall that the history of human-computer interaction has been a continuous effort to lower the barrier to operation—from the command-line interface (CLI) of the PC era, to the graphical user interface (GUI) of the Windows era, to today’s natural language interface (LUI) in the era of large models.
The essence of ChatGPT sites is reducing front-end development from "engineering" to "expression."
In the past, to build a website, your mind had to go through multiple layers of translation.
I want to sell something (intent) → I need a shopping cart and payment system (product logic) → I need to write HTML/CSS/JS (frontend code) → I need to integrate with a database (backend logic)
Now, ChatGPT Sites have eliminated all the middleman hassle.
The user's input has been completely reduced to pure natural language—the most fundamental "intent."
The core advantage of traditional software SaaS is “we’ve packaged our code into nice buttons for you to click,” while the paradigm shift from large models is: you just need to speak.
From now on, all transitional tools centered on “making it easier for non-programmers to write code or build websites” will lose their place in the era of end-to-end generation, where intent directly produces results.
GPT-6 instantly deciphers spectrograms, breaking through human sensory limitations.
Moreover, just today, a set of tests revealed by AI researchers ChrisGPT and Max Rubin left countless audio engineers and AI professionals in awe.
This test is simple: turn sound into an image and have GPT-6 read the image directly.
ChrisGPT uploaded a Mel spectrogram to GPT-6, a technique that visualizes the frequency, time, and energy of sound as a two-dimensional image.
During the process, no actual audio was played, and the large model was offline.
The only prompt is: "This sound comes from an animal/mammal in nature."
GPT-6 replied: "Whale, possibly a blue whale."
The entire network was stunned.
Because it didn't just answer correctly—it surpassed our expectations; the reasoning behind it has gone far beyond simple image matching!
Keep in mind that blue whales produce extremely low-frequency calls.
In nature, the roars of lions, tigers, and elephants also fall within the low-frequency range of 40–200 Hz. It is not possible to determine that the sound is from a whale based solely on a cluster of energy around 50 Hz on a spectrogram.
How does GPT-6 achieve this? Audio engineers analyzing it believe that GPT-6 accurately captures the temporal energy patterns.
It’s not about observing a single point, but about understanding how sound waves flow and decay over time within this specific physical space.
The ability to directly zero-shot guess the animal category from a "picture of sound" with minimal context is incredibly astonishing.
Another tester, Max Rubin, delivered an even more impressive demonstration.
He tested non-natural sound effects and found that Astra could zero-shot identify the sound of a lightsaber swishing from Star Wars directly from Mel spectrograms!
Max exclaimed, "We haven't even scratched the surface yet."
Although in the scientific research field, academia has long been experimenting with spectrograms combined with visual models for whale call detection, and even training specialized LLMs to interpret Mel spectrograms for speech tasks.
However, this is the first time a general-purpose large model has been publicly demonstrated to perform this level of cross-modal reasoning without any specialized fine-tuning.
It’s like never teaching someone how to read sheet music, and simply showing them a full orchestral score.
As a result, he could not only "hear" sounds in his mind, but also accurately tell you that the trombone played the wrong note in the second measure.
Large models are breaking free from human biological constraints.
Human perception is severely limited.
We rely on air vibrations stimulating the eardrum so that the brain can convert these physical vibrations into electrical signals perceived as "sound." Subconsciously, humans associate sound with hearing through the ears and images with seeing through the eyes—this is a "sensory barrier" shaped by millions of years of evolution.
Early multimodal AI was essentially "mimicking" humans. We played audio files for it and showed it images, trying to make it understand like a human does.
However, the significance of GPT-6 Astra recognizing spectrograms is that large models are breaking free from the constraints of human biological organs!
For GPT-6, the long call of a deep-sea blue whale and the slicing sound of a Jedi swinging a lightsaber are fundamentally no different—they are both merely forms of energy fluctuations in the physical world.
When this volatility is transformed into a mathematical and geometric representation such as a Mel spectrogram, GPT-6 can directly "read" the physical laws themselves.
It doesn’t need long ears; it doesn’t need to “hear” sound—it can directly “see” and understand the physical formulas of sound.
This means that AI's multimodal understanding has evolved from "mimicking human senses" to "natively decoding" the data structures of the physical world.
Today, it can identify a blue whale from a spectrogram; tomorrow, if shown an image of seismic waves, could it directly predict the next fault rupture? If shown an EEG reading, could it directly discern what you’re thinking?
Humans perceive the universe's projection through their physical bodies, while now, AI is reading the universe's source code directly.
Reference materials:
https://x.com/ChrisGPT/status/2097047569520115800
https://x.com/maxxrubin_/status/2096892510241268094
https://www.fastcompany.com/91602695/chatgpt-sites-makes-building-a-website-feel-like-making-a-google-doc
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation, edited by Aeneas David.