Variant Fund: How is the crypto world shaping more refined AI models?
Tasteful models have a large (and growing) potential market.Original Title: AI Models with Taste
Original Author: Alana Levin, Variant Fund
Original Translation: 深潮 TechFlow
In the past two years, new AI models have emerged one after another. These AI models are capable of performing many types of tasks—from finding information and answering questions to providing customer support, proofreading documents, generating content, and more.
Many of these tasks are objective, with clear optimization functions: finding the correct answer, identifying the most relevant information, detecting any errors or anomalies, etc.
However, there are also some models whose outputs are highly subjective, such as creating "excellent" artworks or developing "interesting" videos. I refer to these as "tasteful models." Taste-based models are often harder to optimize because they are a blend of collective and individual decisions; there are no obvious answers or outputs. Therefore, frequent feedback is particularly valuable in helping models understand the latest cultural preferences.
Today, there are roughly two ways to cultivate a model's "taste":
Based on user-generated content/data (such as feeds from Twitter or Reddit), which theoretically can reveal the latest trends that humans care about (thus serving as a representation of taste).
Utilizing a community of human "taste makers" to help the model train actively around their preferences.
The first method has many undesirable situations. The data may be isolated (for example, Reddit shutting down its API) or introduce biases (for example, sharing only partial data). The model may also overfit to the algorithms of specific platforms, especially if its data sources are limited. This may not seem important until people start imagining a vast amount of new media generated based on trending content from Twitter. This is not ideal.
The latter method, a network of human feedback, avoids many of the risks mentioned above. There may still be biases, but only in that it includes the preferences of community members who opt in to help train the model. Therefore, the key is to ensure that these community members, the ones who define "taste," are genuinely connected to the model's cultivation of good taste.
Cryptographic tracks can help facilitate this consistency. Providing ownership in the model/economic benefits from the model's outputs to participating members can incentivize their genuine involvement. Cryptocurrency also makes participation more open and accessible: anyone from anywhere in the world can contribute as long as they have an on-chain wallet and internet connection.
A notable example is the Botto project. Botto is an autonomous artist, and $BOTTO token holders have the ability to help train the model weekly. The training is simple: participants vote for or against various images, and Botto learns from the preferences of its members. At the end of the week, the most popular works are auctioned off, and participants who helped train Botto that week are rewarded.
Art is just one category of tasteful models. Others may include film, television, other forms of storytelling (novels, short stories), comedy, and advertising/branding campaigns. Even a few years ago, these tasteful models were impossible. These tools were less expressive, slower, and could not reliably expect models to produce cohesive or (in the case of video) realistic outputs. Only today has this become possible.
Importantly, tasteful models have a large (and growing) potential market. Art is a multi-billion dollar market. Online content consumption occupies trillions of hours of attention each year. If people are already planning to spend time and money on these forms of entertainment, it seems reasonable to give them a stake in the production, which would not only create a more positive user base but also a more satisfied user base. Imagine if the main participants in the Oscar for Best Picture were the audience who helped train and develop the storyline, or if a brand new award were established for films created for the community—that would be very cool.
I believe this is about creating a new category of content rather than replacing existing creations. It is similar to how smartphones and Instagram enabled everyone to become photographers; the existence of these new technologies did not eliminate the work of actual photographers, and in fact, it may have led to more appreciation for the work of photographers. Tasteful models are the same: they create a new form of participation by leveraging new technologies, here referring to cryptographic tracks for consumer ownership and economic connections, thereby expanding each of the aforementioned categories.
In the past few years, we have seen the emergence of thousands of new models. In the coming years, there may be millions (or even more) of new models, and at least some of these models should strive to engage stakeholders in new ways, from greater openness and accessibility to trying new ownership structures that incentivize participation. Tasteful models are particularly well-suited for this kind of innovation, but they are unlikely to be the last.
