A few days ago, A16Z, the top VC in the United States, released a report entitled "Big Ideas in Tech for 2024" based on the views of more than 40 partners, which covers 8 major areas including biohealth, gaming industry, TOC applications, and crypto markets. Because the original text is too long, in order to facilitate everyone's silky reading, Shidao has translated and screened the report (removed the "American Society" part), and made some adjustments and rewrites, please feel free to eat. Here is the text:
In September this year, Sequoia Capital released a report "Generative AI in Healthcare", which showed that AI can handle patient interaction, documentation, pre-authorization, coding, and revenue cycle management well. For example, the dialogue between the doctor and the patient is automatically converted into an electronic medical record and coding, and so on. Sequoia Capital believes that artificial intelligence is now beginning to penetrate into many aspects of the medical industry, thereby greatly improving the efficiency and quality of the medical field, reducing costs and manpower.
So, in 2024, how will AI transform the "biology + health" industry?
Vijay Pande believes that the "handcuffed" healthcare industry is instead "jumping faster and better".
On the one hand, compared to other industries, the healthcare industry is still using old ways of data collection, such as manual input, which provides fertile ground for AI to be transformed
On the other hand, the healthcare industry is currently the only industry that is "regulated for the use of artificial intelligence". Back in April 2019, the FDA regulated some of the AI products used in healthcare to ensure the effectiveness and safety of these products within their regulatory scope, which will make the AI revolution in the healthcare industry easier.
By 2024, these leaps and bounds will be realized, and the quality of life for both doctors and patients will be several orders of magnitude higher.
Vineeta Agarwala believes that the role of AI is to free doctors from complicated tasks and allow them to focus their time and energy on caring for patients, but in terms of "care" itself, AI has not yet played a full role.
Jorge Conde gave a look at the drug development outlook. He believes that just as a traditional rocket can only last a day, traditional drug development has highly customized attributes, which is very time-consuming, high-risk, and expensive. In the future, there will be a batch of drugs that can reuse certain components, such as gene therapy.
In fact, gene ** is not a new thing, it refers to the introduction of exogenous normal genes into target cells to correct or compensate for diseases caused by defective and abnormal genes, so as to achieve "once**, for life**".
As for why Jorge Conde reiterated the gene**, Shidao believes that it may be related to the recent legal launch of CRISPR Cas9 gene editing technology.
This technique can be used to splice genes for a variety of diseases and also won the Nobel Prize in Chemistry in 2020. The turning point came in December, when in less than a month, the UK and the US approved gene editing** for marketing in quick succession, marking an important milestone for about a decade after the technology was discovered.
On one side is a GPT-4 teacher who can understand broken English and can point out language problems in Chinese;On one side is Siri, who only announces the weather and names your pets.
It's no wonder that Microsoft CEO Satya Nadella has called voice assistants of the past decade, including Microsoft's Cortana, "stupid as rocks."
Although speech is the oldest and most common form of human communication, it has never really played a role in "interaction", and all of this will change dramatically in the age of artificial intelligence.
Anish Acharya thinks:"The application of voice-first features will become an indispensable part of our lives." On the one hand, today's large models can communicate fluently with humansOn the other hand, the current related applications have not yet opened up supporting voice functions, which leaves room for AI voice to be transformed. By 2024, voice apps will be more deeply integrated into our lives.
Is it a big model?Or do you want to take a different approach and develop a proprietary model?Olivia Moore, a startup focused on the market, said:"In 2024, there will be more specialized, task-specific AI."
While a general-purpose model like ChatGPT is great, it's unlikely that it will "win" at every task. In 2024, we will see more professional and personalized AI solutions. For example, an AI platform built specifically for researchers, a writing generation tool for journalists, and a rendering platform designed for designers, to name a few.
In the long run, the products that people use every day will be customized for their use cases, whether it's a proprietary underlying model or a special workflow built around it. Some companies will have the opportunity to "own" data and workflows in the era of new technologies;They will excel in one category first and then expand their strengths.
For this problem, Shidao has been written many times in previous articles. As Sam Altman said: OpenAI's "imitators" and "shells" are a dead end. But you can specialize in long-tail areas that large models can't reach. Especially at present, the domestic large model is still in a relatively early stage, and startups can train vertical models in a short window period, build their own moats, and grab subdivided scenarios and users.
Because, for the initial product, the more professional the better.
Bryan Kim thinks"Code-free AI generators spawn new consumer behaviors":As AIGC reduces the marginal cost of creating art to near zero, new consumer behaviors will emerge. For example, on the multi-platform GLIF, users can generate artwork, comics, photos, and so on with a simple prompt.
For Bryan Kim's point of view, everyone starts with pika1The popularity of 0 can be glimpsed. Shidao believes that although the next killer application will be handed over to the test of time, from the perspective of AI startups that have already "died", user experience and computing power cost are all matters of life and death.
Zach Cohen gave **,"AI tools will educate children in 2024":In 2023, 30% of college students used ChatGPT-like tools to complete school assignments (the actual number may be higher). In 2024, AIGC will change the landscape of early education.
Because higher education also has to consider the issue of "academic misconduct", but early education can use AI to create a sandbox environment that can be explored infinitely. The key is to design a product that both engages and protects children, that is, satisfies "content moderation + user center + child-friendly interface". By 2024, groundbreaking AI tools designed for children will appear.
According to Lightspeed Venture Partners, AIGC is the "fourth industrial revolution" in the gaming industry.
When technologies such as "Wensheng Wen, Wensheng **, Wensheng Image" infinitely reduce the marginal cost of game components, a path of "reducing costs and increasing efficiency - UGC - XR headset" becomes clearer.
First of all, for AIGC, after the text and images, it will be 3D and **, combined with audio, interactivity, and the final development cost of the game will be 1 1000 of the traditional development cost.
Secondly, after solving the cost problem, users can create their own games – UGC games. The data shows that in the first quarter of 2023, developers of the UGC service Roblox earned 1$8.2 billion, an estimated 17% increase from 2022, forms a "creator flywheel" that encourages others to participate in creating games.
In this regard, Joshua Lu believes"A new generation of UGC game developers is on the rise":On the one hand, as competition on UGC platforms intensifies, developers are expected to benefit from greater incentives. It's worth noting that Meta's Horizon Worlds expands to mobile in 2023. On the other hand, UGC game developers also have access to more powerful, AIGC-powered tools. (Epic openly supports such a technology, and Roblox has announced a number of AIGC tools.) The combination of these two factors is likely to unleash millions more UGC game creators in 2024.
Again, in terms of enhancing the gaming experience, XR products have also found a product market fit. Andrew Chen argues"The next generation of headsets better double down",And in the process, it will attract millions of consumers, rather than thinking about jumping on productivity tools with low demand.
Doug McCracken thinks"The next Disney will be a gaming company":Global gaming revenue is expected to reach $188 billion in 2023, while the global box office is only expected to reach $34.5 billion. The younger generation of gamers is making gaming their go-to IP. Why?The game offers the deepest story and worldview, is interactive rather than passive, and you can socialize in the game.
Riot, Epic, Supercell, and a new generation of game companies are preparing to replace movies with games and become the "next Disney." This is already happening before the mainstream** is even noticing, and will accelerate in 2024.
There is a saying in 2023: "As long as you have the best model, users will naturally come".
So far, ChatGPT, Character, Bard, and Midjourney are all loved by users, and they are indeed the best models in their respective fields.
But Alex Immerman believes that the situation will change in 2024, ie"The battlefield will shift from a model to a user experience":Thanks to a number of factors – the chip shortage is likely to ease, most of the underlying models are available through APIs, and increasingly powerful open-source models – the foundation is laid for building blockbuster TOC applications on top of other people's models.
By 2024, AI applications will stand out by providing the best user experience around their unique use cases, rather than relying solely on model performance. I'm looking forward to the emergence of apps that are shared by multiple people, put multiple models in one interface, or build more solution-focused applications. At that time, the large model will become a source of differentiation. While today's large models still have a first-mover advantage, traditional barriers like network effects, high conversion costs, scale, and branding can still be the key to long-term success.
For the above point of view, Shidao recognizes "model ≠ product", and does not recognize "how can you be sure that the current winner will not be able to make a good product"?For example, when other platforms are still in line, Midjourney can generate photographer and concept artist level images with low latencyIn other platforms, it has enabled more than 14 million users to volunteer as "supernumerary employees" when designing social attributes, and the resulting image quality has been improved to the best in the industry under the flywheel effect.
In other words, users want the "best product". Under this premise, some of the already dominant "closed-source winners" will become the "king of volumes", and they also have more advantages in realizing the full commercialization and industrialization of large models. Therefore, as far as other start-ups are concerned, it is true that they do not have to "die" to close the source, but can use open source to open up differentiation and establish their own technology moat.
On the ** on AI applications, Sarah Wang, another partner of A16Z, believes"In 2024, AI will go beyond text-based chat to find a new way of storytelling":In the coming year, AI will evolve to multimodal models, and through user fine-tuning and personalization, this will deepen our engagement with AI and make the experience more exciting, entertaining, and engaging. And the task of creating these new narratives falls to startups.
From the above point of view, Inflection AI's AI Pi (Heypicom) takes the route of "bosom friend", which can not only reply with text, but also reply with accented voices. We might as well look forward to what imaginative storytelling will emerge in the future.
In 2013, when Bitcoin was still considered a "Ponzi", a16z set its sights on crypto and became a leading investor in the space. At present, A16Z's investment in blockchain-related projects covers NFT, DeFi, GameFi, stablecoins, Web30, DAO and other fields. So, what do A16Z veteran players think of 2024?
First, what happens when a large "centralized" AI model trained by a handful of tech giants meets a "decentralized" blockchain?
Andy Hall thinks"AI and blockchain combined":Crypto makes it possible to create multilateral, global, permissionless markets. Anyone can contribute—and get paid for it—whether it's computing power or new datasets, for those in need in the network. Leveraging the long tail of these resources will help reduce the cost of AI and make it more ubiquitous.
In addition, encryption can play a role in artificial intelligence – deepfakes. For example, to open a black box;Track down our *** stuff ** and so on. And web3 is the laboratory for decryption. Decentralized, open-source encrypted networks will innovate democratized (rather than centralized) artificial intelligence, ultimately making it safer for consumers.
Miles Jennings focused on presenting"Entering a New Era of Decentralization":In practice, decentralization at scale has always been difficult, especially when compared to the efficiency and stability of centralized systems. At the same time, most Web3 governance models involve DAOs and are based on governance models that do not lend themselves to the socio-political realities of decentralized governance.
However, thanks to the active labs of web3 over the past few years, best practices for decentralization have started to emerge. These practices include adapting decentralized models to applications with richer features;It also includes DAOs adopting Machiavellian principles to design more effective decentralized governance that holds leadership accountable for it. As these models evolve, we should soon see unprecedented levels of decentralized coordination, operational capabilities, and innovation.
Eddy Lazzarin believes that the 2024 annual meeting"Reimagining the User Experience of the Future":While the user experience in crypto has been criticized since 2016, not much has changed: keeping your own keys;Connect your wallet with dapps;Sending signed transactions to more and more network endpoints, and more.
We can't expect users to learn these complex processes in a matter of minutes.
But now, developers are actively testing and deploying new tools that can reset the crypto front-end user experience in mid-2024. One of these tools includes simplifying the login to the app and passing the password, which is an automatically generated encrypted password.
Other innovations include: smart accounts, which make the accounts themselves programmable and therefore easier to manageEmbedded wallets, built into apps, make getting started more silky;Multi-party computation, making it easier for third parties to support signatures without keeping user keys;Advanced RPC (Remote Procedure Call) endpoints that identify user needs and fill gaps, and more. All of this not only contributes to the wider adoption of web3, but also makes the user experience better and more secure than it would be in web2.
The financial industry is known for its large-scale, high-quality data;It has multi-dimensional and diversified application scenarios, providing fertile ground for the modeling, training and application of artificial intelligence.
Seema Amble thinks"[AI] software will greatly enhance financial professional services".This will lead to changes in professions such as accountants, tax advisors, wealth managers, etc.
Historically, software has been used primarily to track workflows, with at most some analytical tools. Now, with advances in generative AI and large models, more work can be automated, including administrative tasks, research processes (collecting data, searching for information), refining insights, and generating reports. This leaves humans with very limited work to do in addition to continuing to improve their profession, or even only reviewing and interacting with customers.
Similarly, D**id Haber and Marc Andrusko argue"AI will be key to improving ROE":In 2024, we will begin to see financial institutions adopt native AI applications across a variety of operational workflows. In addition to revenue generation, middle and back office functions, AI adoption will focus on use cases such as engineering, procurement, legal, compliance, and risk management.
Specifically, Joe Schmidt argues"Large model captures new 'base customer units'":Historically, it has been difficult for operating systems to scrape certain types of unstructured data. For example, in the field of insurance, Vertafore or Applied Systems have struggled to expand beyond tracking issued policies. And by 2024, startups that leverage large models will scrape data that is difficult to collect for existing operating systems, and automatically label and store it. If these startups capture the underlying customer unit upstream of the traditional platform, we may see a new era – the domain served by the software oligopoly.
In addition, Angela Strange believes"The 'developers' of the financial industry are becoming the 'buyers'":Historically, the procurement of financial services infrastructure has been dominated by buyers ("What's my ROI?").and business leadership ("Does this solve my use case?").Decided. But now a third influence group has emerged:Developers.
Fintech companies are prioritizing the creation of "developer sandboxes" that allow customers to "try before they buy" or even open source**. Developer-buyers are also more willing to learn about the product in advance. Therefore, it is important for the seller to consider how to attract developers, which may require improvements in the product architecture (including the latest documentation!).）。
Zeya Yang thinks"B2B AI Product Embedding Workflow":In 2024, I'm optimistic that we'll see native AI products embedded more deeply in workflows, performing tasks such as: proactively leaving comments, updating records, and completing action items after a simple user approval. For example, instead of waiting for the user to query a long document for relevant information, there are AI tools that proactively flag key parts.
Kimberly Tan thinks"Large models drive the progress of machine automation (RPA)":By 2024, I look forward to seeing the market for machine automation (RPA) driven by large models take off. RPA – deploying small "robots" to automate repetitive tasks, such as data entry, – is currently the best solution. However, RPAs are often still manual and often fail, so they often require a lot of work.
Now, with the big model, there is an opportunity to build a smarter RPA system. It understands the context and what it is doing, and can dynamically adjust to create a more robust solution on its own.
Within this, there may be multiple vertical solutions for specific types of automated tasks – whether it's processing invoices for a financial institution or responding to customer service inquiries – and buyers will always buy the solution that best suits their workflow and needs.