Has the Metaverse Bubble Burst? Here Are Some "Data Truths"

Many people are skeptical about the future of the metaverse

In recent years, the tech world has shown great enthusiasm for the concept of the "metaverse". This concept promises to transform the online experience and provide immersive virtual worlds in which we can work, play and socialize in unprecedented ways.

Tech giants and investors have invested billions of dollars in this vision, with Meta (formerly Facebook) being the leader. However, recent data suggests that the metaverse bubble may have burst, leaving many people skeptical about its future.

To understand the current situation, we need to take a step back and look at what the metaverse promised.

Meta CEO Mark Zuckerberg has become the spokesperson for this movement, reshaping his company and investing heavily in virtual reality technology.

Citibank researchers even predicted that the metaverse could attract 5 billion users and grow into a $13 trillion market.

These bold words triggered a gold rush, with companies and individuals scrambling to grab territory in this digital frontier.

01. The sharp decline of the metaverse

Today, the situation is very different.

Meta’s ambitious metaverse division, RealityLabs, has been bleeding money. In the last quarter alone, it lost $4.5 billion, bringing its total losses since inception to more than $46 billion. These figures are a far cry from the profitable future once envisioned.

Even more telling is the fate of Horizon Worlds, Meta’s flagship metaverse platform for adults. Despite extensive marketing efforts, the platform has struggled to attract its target audience. Ironically, it has become unexpectedly popular with children, which was not its original design.

02. The Collapse of the Crypto Metaverse

The metaverse concept is not limited to traditional tech companies.

An entire ecosystem of crypto-based virtual worlds has emerged, promising decentralized ownership and unique digital assets.

These platforms, built on blockchain technology, have been hyped and have astronomical valuations. However, they have also experienced a sharp decline.

Take The Sandbox, a virtual world once valued at more than $7 billion, but its daily trading volume has plummeted by 99.9%. At its peak, it had a trading volume of $117 million, but now it only averages $8,000 a day.

This is not an isolated case. Another pioneering Crypto metaverse platform, Decentraland, has seen a similar 99.9% drop in daily trading volume, from $2.5 million at its peak to less than $5,000 now.

03. The decline of digital assets

The most striking thing about these virtual worlds is that digital assets can be owned and traded, usually in the form of non-fungible tokens (NFTs). These tokens can represent anything from virtual real estate to in-game items.

At the peak of the metaverse craze, the prices of these assets were eye-popping. Now, their value has almost evaporated.

In Sandbox, NFTs once had a single-day sales volume of $10.2 million, but now they hardly exceed $10,000.

This pattern is repeating itself on other platforms, with AxieInfinity, once the poster child for “earn while you play” gaming, seeing its trading volume plummet from nearly $1 billion to less than $2 million.

Crypto associated with these “metaverse” projects has also fared less well. Tokens such as MANA (Decentraland), SAND (The Sandbox), and AXS (AxieInfinity) have all seen their value plummet by more than 90% from their November 2021 peaks.

This decline isn’t just a problem for individual projects; the entire metaverse crypto space has shrunk dramatically, with its total market cap falling from $50 billion to $16 billion.

04. What factors led to this collapse?

Several factors contributed to this rapid decline.

First, the initial hype created unrealistic expectations. The technology needed to deliver a truly immersive and seamless virtual experience is still in its infancy. Many users find the current offerings clunky and disappointing compared to the promised vision.

In addition, the concept itself may be too abstract for mainstream adoption. While technology enthusiasts are excited, average internet users have a hard time understanding how the Metaverse will meaningfully improve their digital lives. High entry costs, both in terms of hardware and learning curves, have further limited its adoption.

The broader economic recession and crypto market crash have also played a major role. As investment capital has become increasingly scarce and risk appetite has declined, many Metaverse projects have found themselves struggling to maintain development and user growth.

Despite these setbacks, it’s too early to write off the Metaverse concept entirely.

Technologies tend to go through cycles of hype, disillusionment, and eventual practical adoption. Some backers, like Mark Zuckerberg, believe in the long-term potential of the Metaverse and continue to invest heavily in Metaverse development.

History has shown that even after a major market correction, innovative ideas can reappear in a more practical form.

Just as companies like Amazon and eBay emerged from the dot-com bubble as tech giants, some Metaverse projects may find their footing and realize value in ways we haven’t yet imagined.

LLM mathematical performance increased by 168%, a 14-person Microsoft team worked on it, the secret of synthetic data 2.0 was revealed, and intelligent agent generation teaching

[Introduction] The secret of synthetic data 2.0 is exposed! Researchers from Microsoft proposed the agent framework AgentInstruct, which can automatically create a large amount of diverse synthetic data. The model Orca-3, which was fine-tuned with synthetic data, refreshed the SOTA on multiple benchmarks.

High-quality data in the world is almost exhausted.

AI scientists have racked their brains to solve this problem.

At present, synthetic data may be the future of large models and has become a recognized solution in the industry.

Even Nvidia scientist Jim Fan once said that synthetic data will provide the next trillion high-quality training tokens.

However, using synthetic data is not completely helpful for LLM training.

Some time ago, a Nature cover study showed that synthetic data will cause large models to crash after 9 iterations. Moreover, similar studies are everywhere.

So, what should we do?

Recently, the Microsoft team proposed an extensible agent framework-AgentInstruct, which can automatically create a large amount of diverse and high-quality synthetic data.

Its biggest advantage is that it can create complete prompts and responses using only the original data source.

Paper address: https://arxiv.org/pdf/2407.03502

In response, the researchers used AgentInstruct to create 25 million pairs of "post-training" data sets, covering a variety of usage skills, such as text editing, creative writing, tool use, coding, reading comprehension, etc.

Then, they used this data to post-train Mistral-7b to obtain the Orca-3 model.

Compared with the original Mistral-7b-Instruct, Orca-3 showed significant performance improvements in multiple benchmarks.

And in terms of mathematical performance, the performance directly soared by 168%.

When "synthetic data" meets intelligent agents

In the past year, we have witnessed the rise of intelligent agents.

Intelligent agents can generate high-quality data, and through reflection and iteration, their capabilities have surpassed the underlying basic large models.

In this process, intelligent agents can review solutions, self-criticize, and improve solutions. They can even use tools such as search APIs, calculators, and code explanations to expand the capabilities of large models.

In addition, multi-agents can bring more advantages, such as simulating scenarios and generating new prompts and responses at the same time.

They can also automate data generation workflows, reducing or eliminating the need for human intervention for certain tasks.

In the paper, the authors proposed the concept of "generative teaching".

This means using synthetic data for post-training, especially creating data through a powerful model to teach another model new skills or behaviors.

AgentInstruct is an agent solution for generative teaching.

In summary, AgentInstruct can create:

- High-quality data: using powerful models such as GPT-4, combined with tools such as search and code interpreters.

- Diverse data: AgentInstruct generates prompts and responses at the same time. It uses multi-agents (equipped with powerful LLMs, tools and reflection processes) and a taxonomy with more than 100 subcategories to create diverse and high-quality prompts and responses.

- Large amounts of data: AgentInstruct can run autonomously and can apply verification and data filtering processes. It does not require seed prompts, but uses original documents as seeds.

Generative Instruction: AgentInstruct

How do we create massive amounts of data? How do we ensure that the generated data is diverse? How do we generate complex or subtle data?

To this end, the researchers outline a structured approach to address these challenges:

Specifically, AgentInstruct defines three different automated generation pipelines:

Content transformation pipeline: Converts raw seeds into intermediate representations, simplifying the process of creating instructions for a specific goal.

Seed instruction generation pipeline: Consists of multiple agents that take the transformed seeds from the content transformation pipeline as input and generate a diverse set of instructions.

Instruction improvement pipeline: Takes the instructions from the seed instruction pipeline as input and iteratively improves their complexity and quality.

Next, the researchers implemented these pipelines for 17 different skills, each with multiple subcategories.

These skills include reading comprehension, question answering, encoding, retrieval-enhanced generation, creative writing, tool/API use, and network control.

The full list is shown in Table 1 below.

Next, the researchers explain how these workflows work through case studies of three skills.

Experimental Results

As mentioned at the beginning, the researchers used 25.8 million pairs of instructions to fine-tune the Mistral-7b-v0.1 model and then obtained Orca-3.

So how does Orca-3 perform after training it with AgentInstruct data?

The goal of AgentInstruct is to synthesize a large and diverse dataset with data of different difficulty levels.

On this dataset, baseline models like Orca-2.5, Mistral-Instruct-7b, and ChatGPT scored well below 10 points, showing their disadvantage relative to GPT-4 (which was designated as a baseline and scored 10).

The performance comparison depicted in Figure 4 shows a comparative analysis between the baseline models and Orca-3.

This figure shows the significant improvement in various capabilities during post-training with the support of AgentInstruct data.

Table 2 summarizes the average scores of all evaluation dimensions.

On average, including Orca-3 after each training round, the introduction of AgentInstruct data improves performance by 33.94% over the Orca 2.5 baseline and 14.92% over Mistral-Instruct-7B.

Refreshing multiple benchmarks SOTA

The results of all baselines for each benchmark are given in Table 3.

For example, it improves by 40% on AGIEval, 19% on MMLU, 54% on GSM8K, 38% on BBH, and 45% on AlpacaEval.

In addition, it continues to outperform other models such as LLAMA-8B-instruct and GPT-3.5-turbo.

For reading comprehension tasks, it is crucial for LLM. It is even more important for small models.

By using AgentInstruct for targeted training, it can be observed that Mistral's reading comprehension ability has improved substantially (see Table 4) - 18% over Orca 2.5 and 21% over Mistral-Instruct-7b.

Furthermore, by leveraging this data-driven approach, the researchers were able to improve the performance of a 7B parameter model on the reading comprehension section of the LSATs to match that of GPT-4.

For math, AgentInstruct successfully improved Mistral’s proficiency on math problems ranging from elementary school to college level, as shown in Table 5 below.

The improvements ranged from 44% to 168% on various popular math benchmarks.

It should be emphasized that the goal of generative teaching is to teach a skill, not to generate data to satisfy a specific benchmark. The effectiveness of AgentInstruct in generative teaching is demonstrated by significant improvements on various math datasets.

Table 6 shows the performance of the Orca-3-7B model and FoFo benchmarks on other open source and closed source benchmarks.

Additionally, AgentInstruct successfully reduced model hallucinations by 31.34% while achieving a quality level comparable to GPT-4 (Teacher).

Table 8 shows the results for all models on MIRAGE with and without RAG.

In summary, the AgentInstruct generative teaching approach provides a promising solution to the challenge of generating large amounts of diverse and high-quality data for model post-training.

10 billion unicorn goes bankrupt, investors collectively reflect

"Why can't we keep such unicorns?"

At the end of July, a US new energy unicorn went bankrupt due to a broken capital chain, triggering a collective reflection in the US venture capital circle. The most consensus view is that "the AI ​​gold rush has seriously concealed the problem of capital shortage of start-ups... and in order to survive, 'investor-friendly' terms are becoming more and more common."

However, "investor-friendly terms" are still too literary. A more straightforward statement is that the US venture capital market is now increasingly polarized:

For start-ups that are not in the limelight, VCs are increasingly focusing on bottom-up guarantees and starting to propose some harsh terms, which founders have to accept in order to survive; for AI companies in the limelight, VCs may continue the expectations and tolerance for innovation in the TMT period, but recently more and more large-scale model companies have been acquired, which also shows that VC's "stop loss point" has been advanced. Obviously, the original "patient capital" has also begun to be "less patient".

Coupled with the record-breaking fundraising scale of credit/private debt funds, this person said, doesn't this just show that the world is a huge grass-roots team? Don’t mention that venture capital is the spark plug of technological innovation. In the end, it’s all about lending.

We can’t draw conclusions so quickly. I think it’s more about asset allocation and risk/return considerations for American investors. In fact, even in the current domestic equity investment market dominated by state-owned assets, we have seen some changes, such as some cities have begun to truly separate financial purposes from the funds they attract, allowing the market to return to the market, and some provinces have begun to establish a sound fault-tolerant mechanism for state-owned assets.

Everyone has a bit of mean reversion.

01 “Investor-friendly” terms are becoming more and more common

In mid-July, Moxion Power, a manufacturer of energy storage batteries, held an emergency online meeting. Senior executives told all employees frankly that the company had been trying to seek financing of about US$200 million at a valuation of US$1.5 billion (more than RMB 10 billion) since the beginning of this year to maintain the normal operation of the production line, but due to “accidents at the last minute”, the negotiations eventually broke down and the company had to face the dilemma of “serious shortage of funds”.

In this case, the top management decided that all employees except the "core backbone" should enter the "vacation state" and enter the final self-rescue stage, and reminded all employees to be prepared for "everything is irreversible and collective layoffs on August 5, 2024".

According to the employees who attended the meeting, the meeting chaired by CEO Paul Huelskamp lasted only 5 minutes, revealing a "decisive and tragic" from beginning to end. Paul Huelskamp even said dejectedly, "Based on the current situation, I don't think our efforts will succeed."

However, at the time, this news did not attract people's attention-or more accurately, a considerable number of financial media believed that this internal meeting was just a small episode on the road to entrepreneurship, and Moxion Power would eventually turn the corner, because Moxion Power is not just a simple unicorn.

Paul Huelskamp and co-founder Alex Meek came from the well-known startup incubator Y Combinator and have an extremely prominent list of shareholders. In less than four years from its establishment to now, they have received investments from Tamarack Global, Amazon Climate Commitment Fund, and Microsoft Climate Innovation Fund, with a total financing scale of US$126 million.

Moreover, people still clearly remember that in May last year, at the opening ceremony of Moxion Power's California production line, the representative of the Democratic Young Forces and California Governor Newsom personally stood on the stage and talked about the ambitious "2045 Future Grid: California Clean Energy Transformation Plan" by borrowing the factory that was about to be completed behind him, praising Moxion Power's new factory for "marking the framework of the future", telling a good "American story", and even better telling the "California Made Story".

To use the words we are more familiar with, as a leading innovative enterprise that builds a circle and strengthens the chain and promotes high-quality economic development, Moxion Power should not consider whether it lives well or not, but only needs to think about whether it flies high or not.

However, everything said in this internal meeting is true, and the difficulties faced by Moxion Power have even been seriously "underestimated".

On July 26, local time, CEO Paul Huelskamp sent an internal letter entitled "The Last Day of Moxion" to all employees, "announcing in advance" that this new energy unicorn has entered the bankruptcy liquidation process, and the tone is almost humble to the dust:

"I know many people are curious about why we have ended up in such a situation, and question whether we could have made a better choice... I also hope to have a straightforward answer, but now I can only say that Moxion is a great vision, and it has been well executed."

So that now when we search for news related to Moxion Power, in addition to introducing their entrepreneurial journey, there are a large number of "reflection articles" with Moxion Power as the incision. Venture capital practitioners and business analysts collectively think about "why we can't keep such a unicorn."

02 Startups with exhausted cash flow

Let's talk about the first half of the reflection of the US venture capital circle: The artificial intelligence gold rush has seriously concealed the problem of capital shortage for start-ups.

Although "startups are facing increasing cash flow pressure" is a topic that has been talked about for a long time, in order to understand why a unicorn company with a well-known investor, in line with the national strategic positioning and the development trend of the times has fallen to this point, analysts have re-conducted a systematic review and obtained the following data to specifically restore the extent of the "cash shortage" that startups are facing:

-The interval between each round of financing for startups has expanded from 15 months in the era of large-scale funding in 2021 to 19 months;

-The median valuation of startups in the growth stage has dropped from US$33 million in the era of large-scale funding in 2021 to US$19 million; among them, the average valuation of the top ten companies in terms of valuation has also dropped from US$250 million in 2021 to around US$130 million, a drop of more than 40%;

-Investors generally raised the threshold for startups' business fundamentals and entered the stage of valuation re-adjustment, which led to not only longer time for startups to obtain new financing, but also greater difficulty in valuation growth - calculated by the annual percentage change in valuation, the annual valuation growth rate of startups has dropped from 14.5% to 9.6%;

- Among all venture capital investments in the first quarter of 2024, transactions completed at a flat or discounted valuation accounted for 27.4% - even if they entered the IPO stage, they were not immune. All startups that entered the secondary market in the first quarter of 2024 saw an average valuation decline of 28%, with the largest decline being the star company Reddit, which fell by nearly 40% compared to its peak valuation in 2021.

Behind all this, the exit channels are not smooth enough, and LPs no longer favor venture capital. The most intuitive reason is that the valuation of the artificial intelligence track has soared and too many positions of venture capital funds have been occupied, which is the most easily overlooked reason.

Taking the first quarter of 2024 as the statistical interval, while the valuations of startups in other tracks have dropped significantly, the median valuation of startups in the artificial intelligence track has successfully exceeded US$70 million, an increase of 51% compared with the data in 2023. If only companies in the late growth stage are counted, this figure will directly exceed US$100 million. Correspondingly, 35% of all unicorn-level financings that have appeared so far in 2024 have occurred in the field of artificial intelligence.

Of course, based on these data alone, everything is understandable. After all, capital is naturally profit-seeking. As the only fast-growing track in the current venture capital market, it is completely reasonable for artificial intelligence to become the main flow of funds.

But the problem is that most artificial intelligence startups are completely unable to provide commercial returns. At last year's TechCrunch Disrupt conference, Tomasz Tunguz, founder of Theory Ventures, estimated the overall development stage of the artificial intelligence track and believed that there were less than 25 startups with annual revenue exceeding US$10 million in the artificial intelligence track. If we further subdivide the track into the field of generative artificial intelligence, then about 95% of the start-ups cannot reach an annual recurring income of $5 million.

And as the track infrastructure such as computing resources becomes increasingly expensive, this "0 return" status cannot be changed in the short term:

The well-known unicorn Stability AI has an expected annual income of just $60 million, but the maintenance cost of the image generation system alone is as high as $96 million;

Inflection AI has accumulated a total financing amount of $1.5 billion, but the weekly income of its main product (AI personal assistant) is still "minimal" after one year of launch, basically zero, and has to seek to merge with Microsoft;

Anthropic is slightly better, able to generate revenue of about $150 million to $200 million, but the expenditure cost has reached a terrifying level of $2 billion, and it is supported by two giants, Google and Amazon.

So when such an industry status quo encounters the long-brewing fomo sentiment, the venture capital industry can be said to be "making the already poor family even worse", and is completely unable to think about revitalizing other tracks.

Giuseppe Stuto, co-founder and managing partner of 186 Ventures, said: "The venture capital industry seems to have entered the era of enclosure movement. Many investors have not considered the problem fundamentally. They just think about squeezing into the hot field at all costs."

Rajeev Dham, partner of Sapphire Ventures, is more pessimistic. He said: "The growth prospects of most startups are not optimistic because artificial intelligence transactions have monopolized the attention of venture capital, resulting in entrepreneurs now being divided into poor and rich... They can only rely on themselves."

03 Costly bridge financing

The background of the first half of the sentence is clear, and the second half is not difficult to understand: when venture capital cannot take care of other tracks, startups have made a lot of "compromises" in order to survive, and then a large number of "investor-friendly" clauses have begun to appear more and more frequently in the contracts of the US venture capital circle.

For example, according to Carta statistics, among the venture capital completed in 2023, nearly 10% of the transactions included a "preferred liquidation right of more than 1x", while this figure was less than 2% in 2022.

Andrea Schulz, audit partner at Grant Thornton, provided specific evidence for this set of data. She said that in a series of venture capital transactions she has handled in the past two years, more and more investors have begun to demand more seats on the company's board of directors and preferential terms, including "more than 1x liquidation priority rights", and many investors even claim 3x to 4x liquidation priority rights.

Along with the rise of "Nx liquidation priority rights", there is also the "right to participate in a new round of financing discount". Simply put, when negotiating, the investor requires the invested company to give itself a 25%-30% follow-up investment discount based on the valuation of the new round of financing.

This move is considered to be a response of the venture capital industry to the "discount financing for survival" of start-ups. This clause can ensure that early investors will not suffer a large floating loss on the books, and at the same time ensure that the capital structure of the invested company will not change so much as to "affect its own rights and interests" after introducing new investors.

In short, in today's US venture capital market, every participant can sense that the balance between entrepreneurs and investors is being broken. Don Butler, managing director of Thomvest Ventures, described this as: "When a company is about to complete the seed round and the A round, and claims in the documents that every participant is 'pari passu' (equal status), someone will soon come in with a long list of terms and say, 'No way, I still have a lot of requirements to add.'"

But this is not even more frustrating for entrepreneurs. Data shows that in order to survive, more and more entrepreneurs have to accept costly internal bridge financing.

Bridge financing does not require public disclosure, nor does it require interest to be paid in cash. Instead, it is paid in stock after the IPO or private placement is successful. It is very suitable for helping companies achieve cash flow and break-even or profitability in a short period of time. According to Pitchbook statistics, in 2023, when the exit market is sluggish, a large number of startups will try to make their financial situation slightly more "decent" through bridge financing.

However, since this move is equivalent to telling investors that "they have foreseen the difficulties they may encounter", the scale of bridge investment is generally not large, and therefore by 2024, the US venture capital market has entered a state of "large-scale maturity of bridge financing".

Under such premises, for entrepreneurs who are getting deeper and deeper into trouble, the emergence of more "investor-friendly terms" has almost become an irreversible trend. For example, do you still remember what Mr. Zhu Xiaohu mentioned at the beginning, "American investors also like dividends", Carta has already provided statistical data support - in all venture capital in the first quarter of 2024, 9.5% of the transactions included "cumulative dividends" clauses.

In addition, the increasing number and scale of credit fund raising seems to confirm the saying that "the end of venture capital is lending".

At the end of July, the well-known asset management company Ares Management completed the largest debt fund in its history, with a total scale of more than US$15 billion (equivalent to RMB 107 billion), mainly providing priority secured loans to companies in the North American market with EBITDA between US$10 million and US$150 million. According to reports, as of now, the fund has already had US$9 billion of funds pre-booked, with more than 160 clients.

In 2024, there are two debt funds with a total fundraising of more than US$13 billion (equivalent to RMB 92.9 billion), namely HPS Specialty Loan Fund VI established by HPS Investment Partners, with a scale of US$14.3 billion, and West Street Loan Partners V established by Goldman Sachs Alternative Asset Management, with a scale of US$13.1 billion.

With continuous accumulation, the annual fundraising amount of debt funds has also exceeded venture capital, becoming the second largest choice for investors in the private equity market.

The fact is that credit funds have been popular for several years. This is also a rational choice for global LPs as various risks continue to accumulate. According to "All the well-known PEs have gone into lending"

Apple is testing a new feature that may change the ad blocking experience

The "distraction control" Apple is testing may open up new ideas for browser ad blocking.

For any user with a little experience, "ad blocker" may be one of the indispensable tools in daily life.

With the help of these small and practical programs, you can greatly reduce the chance of being interrupted by advertisements when visiting web pages, and obtain a cleaner and safer browsing experience. Not to mention that because the ad blocker prevents ads and pop-ups from loading, it can even significantly increase the rendering speed of web pages and effectively reduce traffic consumption.

But I don’t know if you have thought about a question. Logically speaking, the ad blocker is used on your own computer (browser). It does not send data to the website. So how does the website know whether the user has used ad blocking? What about the device?

In response to this problem, we at Sanyi Life inquired about some "anti-ad blocking" information. According to it, anti-ad blocking plug-ins mostly involve detection scripts running on the website server. They will specifically detect whether the ad blocker is running on the current page. In addition, some information also mentions using CDN reverse proxy and other technologies to bypass the blocking of advertisements by ad blockers, so that advertisements can be "displayed normally" and so on.

It is not difficult to see that for websites that rely heavily on advertising revenue, if a large number of users use ad blockers, it may seriously affect the operation of the website. As a result, this will prompt website owners to "counteract" ad blockers forcefully, such as forcing users to turn off the blocker first, otherwise they will not be allowed to view website content, or use some methods to make ad blocking ineffective.

As a result, an "infinite loop" is formed, which actually returns the user experience to the level before the birth of ad blockers.

So are there any ways to "block ads" without being discovered by "anti-blocking solutions"? The new features Apple has recently added to their beta system may provide us with some inspiration.

01

According to information released by Apple, this new feature is called “Distraction Control” and is integrated into the Safari browser to provide users with a less disruptive web browsing experience.

Interestingly, Apple also specifically emphasizes that "Distraction Control" is not an ad blocker. It will not automatically identify ads or pop-ups on web pages, nor can it automatically block them. To use this feature, users need to specify which content on web pages they "don't want to see", and Safari will block the display of this content. The blocked object can be an advertising banner, a floating window, or even a cookie request that pops up on the website (in this way, the user can browse the website without answering "Do you allow cookies?").

At first glance, do you think that “Distraction Control” is essentially an ad blocker? It’s just that Apple hands over the rights to “designated advertising” to users, thereby avoiding possible infringement risks. But in fact, things may not be that simple. Because in the relevant instructions, Apple specifically mentioned that the interception effect of "distraction control" will automatically fail for web elements with content update mechanisms.

What is this concept? You must know that in traditional ad blockers, when you choose to block an ad column or floating ad window, what is actually intercepted is its corresponding web page or the entire page frame. After these web pages are blocked, they will be prohibited from loading, thus making It cannot be displayed. Of course, this is one of the reasons why traditional ad blockers are easily "detected".

But according to Apple, if we use "distraction control" to block an advertising banner and its content is updated (such as replacing a new advertisement), then the blocking will automatically fail and theoretically need to be blocked again. specified operation.

02

I have to say that this is very interesting, because it seems to mean that the "distraction control" function may not prevent the website ads from loading normally, but only blocks the rendering or display of specific content. To put it more bluntly, from the website's own "view", the advertisement is most likely loaded normally, but at this time the user cannot see its existence.

Of course, since Apple has not yet explained the specific principle of "distraction control", it cannot be simply inferred whether it "hides" the advertising content or "transparent" it, or even It is not rendered at the GPU level.

But in any case, judging from the functional performance of this obviously different from conventional ad blockers, Apple’s “distraction control” may open up a new idea of ​​​​browser ad blocking. Yes, although it is impossible for them to create a "perfect" and highly automated ad blocking function for various reasons, by "borrowing" such technical ideas, better-used ones may be born in the future. Third-party ad blocking plugins. And this may be what most friends are really looking forward to.

Exclusive Insights

Cutting-edge analysis providing a comprehensive perspective on news, technology, finance.

Sparknews provides cutting-edge information with exclusive perspective, allowing users to see the future first.

John Doe

★★★★★