OpenAI has been trained by datasets obtained by scraping the entire internet numerous times. This allows the model to be readily trained on text-based inputs and terms. The latest trend is towards multimodal inputs, in other words instead of asking ChatGPT a question in writing, a prompt could instead be a photo of a menu with a voice note that says choose an order for me that meets my macronutrient goals.
The internet has a wealth of text-based and video data, however in this new era of multimodal AI inputs it will be all the more important to go out into the real world to gather fresh, real-time data points. The most reliable way to do this currently is logically via people’s phones; OpenAI recently announced that it is collaborating with former Apple Head of Design Jony Ive and Softbank to build an “iPhone of artificial intelligence”.
Alongside this specialized phone, it seems likely there will be an enormous boom of wearable tech to accompany this: smart glasses, necklaces, earpieces, health trackers, and even cars will be gathering as much data as possible and transmitting it back over the air to the ‘mothership’ to continue refining the model. This trend will only accelerate as low-orbit satellite internet e.g. Starlink becomes available in all regions globally.
If AI is an intelligence boom, then crypto is a coordination boom.
The biggest and best product of crypto thus far has been the ability to enable ownership, align incentives, and tokenize experiences to mobilize millions of people towards a shared cause or objective.
The more the well of internet data gets scraped the more it becomes table stakes. Companies like OpenAI are already demonstrating the need to go into the real world to acquire new datasets. Crypto tokenomics, particularly DePIN (decentralized physical infrastructure) tokens can be applied to great effect here, and we are already seeing the early examples of this trend with WorldCoin.
It is clear crypto tokens will be used to coordinate, reward, and incentivize certain behaviors to train LLMs with a voracious appetite for multimodal inputs. In the not-so-distant future, it seems very likely that high-throughput blockchains will play a key role in distributing this economic value to hundreds of millions of users all over the world to feed the models powering the AI boom.