jakiro
Author: Alex xu, Mint Ventures
introduction
As of now, this round of the Cryptic Bull market is the most boring round of commercial innovation. The lack of a phenomenal boom track such as DEFI, NFT, and Gamefi in the last round of bull markets has led to the lack of industrial hotspots in the overall market.The growth of developers is relatively weak.
This is also reflected in the current asset price. From the perspective of the entire round of cycles, most Alt Coins continues to lose blood on the exchange rate of BTC, including ETH.After all, the valuation of the smart contract platform is determined by the degree of prosperity of the application. When the development of the application is lacking, the valuation of the public chain is difficult to lift.
As the newer encrypted commercial category in this round, AI benefits from the development speed and continuous hotspots of the explosive development of the external business world. It is still possible to bring a good incremental increase in the AI track project in the encrypted world.
In the IO. Net report released by the author in April, the necessity of combining AI and Crypto, that is, the advantages of encrypted economic solutions in certainty, mobilizing allocation of resources and relief.One of the three challenges for Human -machine.
In the AI track in the field of encrypted economy, the author tries to discuss and deduce some important issues through an article, including:
What else is the sprouts of the encrypted AI track, or the narrative that will explode in the future
· The catalytic path and logic of these narratives
· Narrative -related project standard
· The risk and uncertainty of narrative deduction
This article is the stage of thinking as of the time when I published it. It may change in the future, and the views have strong subjectivity. There may also be errors in facts, data, and reasoning logic. Do not as a reference for investment. Welcome to criticize and discuss the criticism of the industry.Essence
The following is part of the text.
The next wave of narratives of the encryption AI track
Before forming the next wave of the encryption AI track, let’s take a look at the main narrative of the current encryption AI. From the perspective of market value, more than 1 billion US dollars are:
Becetic power: Render (RNDR, 3.85 billion in circulation), AKASH (circulating market value of 1.2 billion), IO. Net (the last round of first -level financing valuation of 1 billion)
Algorithm network: Bittensor (TAO, the market value of circulation 2.97 billion)
AI proxy: fetchai (fet, 2.1 billion before merging circulation)
Data time: 2024.5.24, currency units are US dollars.
Except for the above areas, which AI track will be over 1 billion in the next single project?
The author feels that it can speculate from the two perspectives: the narrative of “industrial supply end” and the narrative of “GPT time”.
The first perspective of AI narrative: From the supply side of the industrial, see the opportunity of energy and data track behind AI
From the perspective of the industrial supply side, the four driving forces of AI development are:
· Algorithm: High -quality algorithms can perform training and reasoning tasks more efficiently
· Computing power: Whether it is model training or model reasoning, GPU hardware is required to provide computing power. This is also the main industrial bottleneck at the moment.
· Energy: The data computing center required by AI will generate a lot of energy consumption. In addition to the power required by the GPU itself to perform the calculation task, it also requires a lot of energy to deal with GPU heat dissipation. A large data center cooling system accounts for the total.About 40% of energy consumption
· Data: The improvement of the performance of large models requires expanding training parameters, which means massive high -quality data requirements
For the driving force for the above four industries, algorithms and computing tracks have encrypted projects with a market value of more than 1 billion US dollars in circulation, and no projects with the same market value of energy and data tracks have not yet appeared in the energy and data track.
In fact, the shortage of energy and data supply may soon come, becoming a new wave of industrial hotspots, thereby driving the boom in related projects in the encryption field.
Let’s talk about energy first.
On February 29, 2024, Musk said at the 2024 conference of the Bosch Interconnection World: “I predicted the chip shortage more than a year ago, and the next shortage will be electricity. I think there will be no enough electricity next year toRun all chips. “
From the perspective of specific data, the HUMAN -Center Artificial Intelligence Institute (Human -Center Artificial Intelligence) led by Li Feifei will release the AI Index Report every year.The evaluation believes that the scale of AI energy consumption accounted for only 0.9%of global power demand at that time, and the pressure on energy and environment was limited.In 2023, the International Energy Agency (IEA) summarized the 2022: Global Data Center consumed the power of about 460 Taiwa (TWH), accountingThe lowest energy consumption will also be 620 Taiwa time, and the maximum will reach 1050 Taiwa time.
In fact, the estimation of the International Energy Agency is still conservative, because there are already a large number of projects around AI, and the scale of its corresponding energy demand far exceeds its 23 years of imagination.
For example, the Stargate project that Microsoft and Open AI are planning.This plan is expected to be launched in 2028 and is completed around 2030. The project plans to build a super computer with millions of special AI chips to provide OPEN AI with unprecedented computing capabilities and support its artificial intelligence, especially large language models, especially large language modelsIn terms of research and development.The plan is expected to cost more than $ 100 billion, which is 100 times higher than the current large data center cost.
The energy consumption of a project of the interstellar gate alone is as high as 50 Taiwa.
Because of this, the founder of OpenAI, Sam Altman, said at the Davos Forum in January this year: “In the future, artificial intelligence will need energy breakthroughs, because the power consumed by artificial intelligence will far exceed people’s expectations.”
After computing power and energy, the fast -growing area of the AI industry is likely to be data.
In other words, the shortage of high -quality data required by AI has become a reality.
At present, in the evolution of GPT, human beings have basically figured out the law of the ability to grow in large language models -that is, by expanding model parameters and training data, the ability to index level levels can be improved -and this process can not be seen in the short termTechnical bottleneck.
However, the problem is that high -quality and disclosed data may become scarce in the future, and AI products may face the same contradiction between supply and demand as chips and energy in terms of data.
The first is the increase in disputes in data ownership.
On December 27, 2023, the New York Times officially sued OPENAI and Microsoft to the Federal Federal Court, accusing them of using their own hundreds of articles to train GPT models without permission, asking themValue works bear billions of dollars of legal and actual damage compensation “, and also destroy all models and training data including the New York Times copyright materials.
At the end of March, the New York Times issued a new statement that not only pointed to Open AI, but also aimed at Google and Meta.The New York Times statement says that Open AI transcribes a large number of voice parts in YouTube videos through a voice recognition tool called Whisper, and then generates text to train GPT -4 as a text.The New York Times stated that the use of small companies to use thief touches when training AI models is now very common, and said that Google is also doing such things.In essence, the rights and interests of video content creators are infringed.
The New York Times and Open AI, as the “first case of AI copyright”, considers the complexity of the content of the case and the profound impact on the content of the content and the future of the AI industry, may not be able to get a result soon.One of the possible results is that the two parties reconcile outside the court. Microsoft and Open AI who are rich in wealth pay a large amount of compensation.But in the future, more data copyright friction will inevitably increase the comprehensive cost of high -quality data.
In addition, as the largest search engine in the world, Google also revealed that the search function is considering its own search function, but the charging target is not an ordinary mass, but AI.
>
Source: Reuters
Google’s search engine server keeps a lot of content, and it can even be said that the content of all Internet pages has appeared on all Internet pages since the 21st century.The current AI -driven search products, such as Perplexity overseas, and domestic such as Kimi and Secret Tower, are processed through AI and output them to users.The search engine’s charges will inevitably increase the cost of data acquisition.
In fact, in addition to public data, the AI giants also follow the non -public internal data.
>
Photobucket is an old -fashioned picture and video custody website. In the early 2000s, it had 70 million users and nearly half of the US online photo market share.With the rise of social media, the number of Photobucket users has decreased significantly. At present, only 2 million active users are left (they pay a high fee of 399 US dollars per year), and the agreement and privacy policy signed according to the user registration will beThe account will be recycled, and it also supports the right to use the pictures and video data uploaded by the user.Ted Leonard, CEO of Photobucket, revealed that the 1.3 billion photos and video data it owned were very valuable to the training generation AI model.He is negotiating with a number of technology companies on sale of these data. The scope of the quotation ranges from 5 US dollars to $ 1 per photo. Each video exceeds $ 1, and its estimated data provided by Photobucket is worth more than 1 billion US dollars.
EPOCH, which focuses on the development of artificial intelligence, based on the use of data and the generation of new data according to the 2022 machine learning, and then considering the growth of computing resources, it has published an article on the status of the data required by machine learning “Will we run out of data? An Analysis of the Limits of Scal in G Data Sets in Mach in E Learn InThe image data will be exhausted from 2030 to 2060.If the efficiency of data can not be significantly improved, or new data sources appear, the trend of large machine learning models that currently depend on massive data sets may slow down.
According to the current situation of the AI giants to buy data at high prices, free high -quality text data has been basically used. Epoch’s predictions 2 years ago were more accurate.
At the same time, the solution to the needs of “AI data shortage” also appears, namely: AI data provides services.
DEFINED. AI is a company providing AI with customized and high -quality data for AI.
>
Data types that defined.ai can provide examples: https://www.defined.ai/dataSets
Its business model is: AI provides its own needs for data for DEFINED. AI. For example, as far as pictures are concerned, the quality requires more than how much resolution, avoiding fuzzy, overexposure, and real content.In terms of content, AI companies can customize specific themes according to their own training tasks, such as photos of nights, cone barrels, parking lots, signs, and signs to increase AI’s recognition rate in night scenes.Volkswagen can receive tasks. After the shooting is uploaded, the company is reviewed, and then settled the part that meets the requirements according to the number. The price is about $ 1-2 in a high-quality picture.High-quality films for more than 10 minutes are $ 100-300, and the text is 1,000-character 1 US dollars. Those who take the split mission can get about 20% of them.Data provision may become another crowdsourced business after the “data mark”.
The global task allocation, economic incentives, the pricing of data assets \ circulation and privacy protection, and everyone can participate, it sounds like a commercial door suitable for the web3 paradigm.
AI narrative target under the perspective of the industrial supply end
The focus of the chip’s shortage penetrates into the encryption industry, making distributed computing power the most popular and highest market value AI track category as of the currently.
So what are the contradictions between the supply and demand of the AI industry in energy and data. What are the current narrative-related projects in the encryption industry in the next 1-2 years?
First look at the target of energy.
The energy items of the head CEX have been launched very rare, only Power Ledger (token POWR).
Power Ledger’s item in 2017 is a comprehensive energy platform based on blockchain technology. It aims to achieve the decentralization of energy transactions, promote direct transaction power of individuals and communities, support the widespread application of renewable energy, and ensure it through smart contracts to ensure thatThe transparent and efficient transaction.Initially, Power Ledger was running based on the alliance chain transformed by Ethereum.In the second half of 2023, Power Ledger updated the white paper and launched its own comprehensive public chain. The public chain was transformed based on Solana’s technical framework, which facilitated the high -frequency micro trading in the distributed energy market.The main business of Power Ledger currently includes::
· Energy transactions: Allow users to buy and sell electricity directly, especially the power from renewable energy.
· Environmental product transactions: such as carbon credit and renewable energy certificates, and financing based on environmental products.
· Public chain operation: attract applied developers to build applications on the Powerledger blockchain, and the transaction cost of the public chain is paid at the POWR token.
At present, the market value of the Power Ledger project is 170 million $, and the market value of the full circulation is 320 million $.
Compared with the encryption label of energy, the number of encryption bids of the data track is richer.
The author only listed what he currently pays attention to, and has been launched at least the data track project of one of the CEX, OKX, and Coinbase, and arranged from low to high according to FDV:
1. StreamR – Data
The value proposition of StreamR is to build a decentralized real -time data network that allows users to freely trade and share data, while maintaining a complete control of their own data.Through its data market, Streamr hopes that data producers can sell data flow directly to consumers who are interested without intermediary agencies, thereby reducing costs and improving efficiency.
>
Source: https: //streamr.network/hub/projects
In actual cooperation cases, StreamR cooperates with another web3 vehicle hardware project DIMO to collect temperature, air pressure and other data through DIMO hardware sensors loaded on the vehicle to form weather data and pass to the required institutions.
Compared with other data items, StreamR focuses on the data of the Internet of Things and hardware sensors. In addition to the DIMO vehicle data mentioned above, other projects also include Herbiny’s real -time traffic data flow.Therefore, Stream’s project token Data also created a single -day increase in a single day when Depin’s concept was the hottest.
At present, the market value of the StreamR project is 44 million $, and the market value of the full circulation is 58 million $.
2. COVALENT -CQT
Unlike other data projects, COVALENT provides blockchain data.The COVALENT network reads the data through the Blockchain node through the RPC, and then processs and organizes these data to create an efficient query database.In this way, users of COVALENT can quickly retrieve the information they need, without having to directly query from the blockchain node. Such services are also called “blockchain data indexes”.
COVALENT’s customers are mainly B, including DAPP projects, such as various DEFIs, including many centralized encrypted companies, such as CONSENSYS (Metamask’s parent company), Coingecto (well -known encryption asset market), Rotki (tax tools), Rainbow, etc. In addition, the giants in the traditional financial industry, Welfare, and the Four Cultural Firm Ernst & Young, and are also customers of Covalent.According to data disclosed by COVALENT, the income of the project from the data service has exceeded the Graph in the same field.
Due to the integrity, openness, authenticity, and real -time nature of the Web3 industry, it is expected to become a high -quality data source of segment AI scenes and specific “AI models”.COVALENT, as a data provider, has begun providing data for various AI scenarios, and launched verified structural data specifically for AI.
>
Source: https://www.covalenthq.com/solutions/decentralized- ai/
For example, providing data for smart trading platform Smartwhales on the chain, and use AI to identify profitable transaction models and addresses; ENTENDRE Finance uses the structured data of COVALENT to process AI processing for real -time insights, abnormal detection and predict analysis.
At present, the main scenarios of the chain data services provided by COVALENT are still dominated by finance, but with the generalization of Web3 products and data types, the use of data on the chain will also be further expanded.
At present, the market value of the COVALENT project is 150 million $ and the market value of the full circulation is 235 million $. Compared with the blockchain data index project The Graph of the same track, it has a relatively obvious valuation advantage.
3. HiveMapper -Honey
Among all data materials, the unit price of video data is often the highest.HiveMapper can provide AI with data including video and map information.HiveMapper itself is a decentralized global map project, which aims to create a detailed, dynamic and accessible map system through blockchain technology and community contributions.Participants can capture map data and add it to the open source HiveMapper data network by driving recorders, and obtain the award of the project token Honey based on contribution.In order to improve the effect of the network and reduce the cost of interaction, HiveMapper was built on Solana.
HiveMapper was first established in 2015. The original vision was to use drones to create maps. However, it was discovered that this model was difficult to expand, which turned to use driving recorders and smartphones to capture geographical data and reduce the cost of global map production.
Compared with Google Map and other street views and map software, the Hive Map Per can more efficiently expand the map coverage, maintain the freshness of the map, and improve the quality of the video through the incentive network and crowdsourcing mode.
Before AI’s demand for data broke out, the main customers of HiveMapper include the autonomous driving sector, navigation service companies, governments, insurance and real estate companies of the automotive industry.Today, HiveMapper can provide extensive roads and environmental data for AI and large models through APIs. Through continuously updated images and road characteristic data flow input, the AI and ML models will be able to better transform data into capabilities and implementTasks related to geographical location and visual judgment.
>
Data source: https://hiveMapper.com/blog/diversify–computer-vision-models-with-global-magery-data/
At present, the market value of HiveMapper -Honey project is 120 million $, and the market value of the full circulation is 496 million $.
In addition to the above three projects, the projects of the data track also include The Graph -GRT (the market value of 3.2 billion yuan, FDV 3.7 billion).Protocol -Ocean (circulation market value of 670 million $, FDV 1.45 billion $, this project is about to merge with fetch. AI and SingularityNet, token converted into ASI), a open source agreement, aimed at promoting the exchange and monetization of data and data -related services.Connect data consumers with data providers to share data on the premise of ensuring trust, transparency and traceability.
The second perspective of AI narrative: GPT reappears at all times, general artificial intelligence comes
In my opinion, the first year of the “AI track” in the encryption industry is the 2023 that GPT shocked the world. The skyrocketing AI project is more, which is more of the “heat wave” brought about by the explosive development of the AI industry.
Although the capabilities of GPT4, Turbo, etc. have continued to upgrade after GPT3.The cognitive impact is weakening, and people are beginning to use AI tools, and large -scale job replacement seems to have not yet happened.
So, will the AI field of “GPT” in the future, and the AI leap -up development that has shocked the public will make people realize that their lives and work will be changed because of this?
This moment may be the advent of General Artificial Intelligence (AGI).
AGI refers to the comprehensive cognitive ability similar to humans, which can solve various complex problems, not limited to specific tasks.The AGI system has high degree of abstract thinking, extensive background knowledge, common sense reasoning and causal understanding of the whole field, and cross -professional migration learning.The performance of AGI is no different from the best humans in various fields. In terms of comprehensive ability, it completely surpasses the best human group.
In fact, no matter the presentation of science fiction, games, games, film and television works, or after the rapid popularization of GPT, the public expectations of the public about AGI, which beyond the level of human cognition.In other words, GPT itself is an AGI pioneer product, which is a prediction version of GM artificial intelligence.
The reason why GPT has such a large industrial energy and psychological impact is that the speed and performance of its landing surpassing Volkswagen’s expectations:People did not expect that an artificial intelligence system that could complete Turing test really came, and it was so fast.
In fact, artificial intelligence (AGI) may have reappeared the suddenness of “GPT time” again in 1-2 years: people have just adapted to GPT’s assistance, and found that AI is not just an assistant, it can even complete the extreme independentlyCreative and challenging work, including those who have trapped the top scientists of human beings for decades.
On April 8 this year, Musk accepted an interview with Nicolai Tangen, chief investor of the Norwegian Sovereign Fund Fund, talked about the time when AGI appeared.
He said: “If AGI is defined as smarter than the smartest part of humans, I think it is likely to appear in 2025.”
That is, according to his inference, at most it takes one and a half years, AGI will come.Of course, he added a prerequisite, that is, “if both electricity and hardware can keep up.”
The benefits of the advent of AGI are obvious.
It means that the level of human productivity will be a step on the ground, and a large number of scientific research problems that have trapped us for decades will be solved.If we define the level of “the smartest part of humans” as the level of the Nobel Prize winner, it means that as long as the energy, computing power, and data are sufficient, we can have countless tireless “Nobel Prize winners”, and those who are in close weather are those who are the most the most.Difficult scientific issues.
In fact, the Nobel Prize winner is not as precious as one -hundred -hundred -hundred -in. They are mostly the level of professors of top universities in terms of ability and intelligence.The person who is equal to him, his equally good colleagues may also win the Nobel Prize in the parallel universe of scientific research.However, it is helpless that people with top universities and participating in scientific research breakthroughs are still insufficient, so the speed of “traversing all the correct directions of all scientific research” is still very slow.
With AGI, in the case of full supply of energy and computing power, AGI, which can have an unlimited “Nobel Prize winner” level, explores in any possible scientific research breakthrough direction, and the technical improvement will be dozens of times faster dozens of times faster.EssenceThe improvement of technology will cause us to think that it is quite expensive and scarce now increased by 100 times in 10 to 20 years. For example, grain production, new materials, new drugs, and high -level education, etc.We can support more people with less resources, and the per capita wealth increases rapidly.
>
Global GDP Total Trend Chart, Data Source: World Bank
This sounds a bit sensational. Let’s look at two examples. These two examples have also used the research report on IO. Net:
· In 2018, the Nobel Prize winner Francis Arnold said at the awards ceremony: “Today we can read, write and edit any DNA sequence in practical applications, but we cannot create through it.”Five years after he spoke, in 2023, researchers from AI startups from Stanford University and Silicon Valley, Salesforce Research, published papers in “Nature-Biotechnology”.0 created a new 1 million protein, and found two proteins with different structures that have different structures, but they all have bactericidal ability, and hope to become a bacterial confrontation solution other than antibiotics.In other words, with the help of AI, the bottleneck of “creation” of protein breaks through.
· Previously, the artificial intelligence Alphafold algorithm was predicted by almost all 214 million protein structures on the earth within 18 months.This result is hundreds of times the results of all human structural biologists.
The change has occurred, and the advent of AGI will further accelerate this process.
On the other hand, the challenge brought by the advent of AGI is also very huge.
AGI will not only replace a large number of brainpower workers, but now it is considered that the physical service operator is considered to be “less impacted by AI”.The proportion of labor replacement will increase rapidly.
At that time, the two problems that have been very distant will emerge quickly:
1. Employment and income of a large number of unemployed people
2. In a world where AI is ubiquitous, how to distinguish AI and humans
The WorldCoin \ WorldChain is trying to provide solutions, that is, using UBI (basic income) system to provide basic income to the public, and distinguish between people and AI based on iris -based biological characteristics.
In fact, UBI, which gives the whole people, is not an empty pavilion without practical practice. Finland, England and other countries have carried out basic revenue of the people. Canada, Spain, India and other countries have also actively proposed related experiments.
The advantage of UBI allocation based on biological characteristics recognition+ blockchain is the global nature of this system, which has a widerfolding population. In addition, it can be based on the user network expanded through income distribution to build other business models.For example, financial services (DEFI), social, mission crowdsourcing, etc. to form a collaboration of business in the network, which is exactly exactly
One of the corresponding targets brought by the impact effect brought by the advent of AGI is WorldCoin -WLD, which has a market value of 1.03 billion $, and the full circulation market value is 47.2 billion $.
Risk and uncertainty of narrative deduction
This article is different from the many projects and track research reports previously released by Mint Ventures. It is of great subjectivity to the deduction and prediction of narrative. Readers are requested to use the content of this article as a divergent discussion, rather than predict future predictions.The author’s above narrative deduction faces many uncertainty, leading to wrong conjecture. These risks or influencing factors include but are not limited to:
Energy: The energy consumption of the GPU update is reduced
Although the energy demand around AI has increased, chip manufacturers represented by Nvidia are through continuous hardware upgrades to provide higher computing capabilities with lower power consumption.The new -generation AI computing card G B200 of GPU and a Grace CPU. Its training performance is 4 times that of the previous generation’s main AI GPU H100. The reason for reasoning is 7 times that of H100, but the energy consumption is only 1/4 of H100.Of course, people want the power to gain from AI. The desire is far from the end. With the decline in unit energy consumption, with the further expansion of AI application scenarios and demand, the total energy consumption may increase.
In terms of data: Q * plan to realize “self -produced data”
There has always been a rumored project “Q *” inside Open AI. The internal information sent by the project to employees in Open AI has been mentioned.According to Reuters’s opinion, this is a breakthrough that OPEN AI has obtained on the road to pursue the road of pursuit of super intelligent / General Artificial Intelligence (AGI).Q * Not only can it solve the mathematical problem that has never been seen before before, but also can create data used for large models through self -creation without the need for data feeding in real world.If the rumor is true, the bottleneck of AI model training limited by high -quality data will be broken.
Agi comes: OPENAI’s hidden concerns
It is really unknown whether the time of the time of AGI comes, as Musk said, it will arrive in 2025, but this is just a matter of time.But as the direct beneficiary of the AGI’s approach to narrative, Worldcoin may come from Openai. After all, it is recognized as the “OpenAI shadow tokens”.
In the early morning of May 14th, the latest GPT -4O and another 19 different versions of large language models performed in the comprehensive task score in the new spring product launch conference.It seems that it is a lot higher than the latter, but from the perspective of the general score, it is only 4.5%higher than the second GPT 4 Turbo, which is 4.9%higher than the fourth Google Gemini 1.5 Pro, which is the first one than the first, which is the first, which is the first.The five Anthropic Claude 3 OPUS is 5.1%higher.
>
The moment when GPT3.5 was shocked at the beginning of the world, it only passed for more than a year. Openai’s competitors have chased a very close position (although the GPT5 is not released and it is expected to be released this year), whether the Openai will be maintained in the future can be maintainedIt seems that this answer is becoming blurred.If OpenAI’s leading advantages and dominance status are diluted or even overtaken, then the narrative gold content of the shadow tokens as the OpenAI shadow tokens will also decline.
In addition, in addition to Worldcoin’s iris certification scheme, more and more competitors have also begun to enter this market. For example, the palm scanning ID item Humanity Protocol has just announced that it has completed a new round of financing for $ 30 million for $ 1 billion. Layerzero LabsIt is also announced that it will run on Humanity, and add its verified node network to use ZK to prove to verify the credentials.
Conclusion
In the end, although the author has carried out the follow -up narrative of the AI track, the AI track is different from the encrypted native tracks such as DEFI.Not running, many projects are more like the AI theme MEME (such as RNDR similar to Nivine Meme, WorldCoin similar to Open AI meme), readers should be cautious.
