Revolutionizing Gaming with AI
Recently, Framework Venture’s Vance Spencer revealed a fascinating development in the gaming industry. The game, AI Arena, combines the thrill of competitive gaming with the complex world of AI modeling. As an investor in the game, Spencer is captivated by the potential of this novel merger. Framework Ventures has backed winning technologies before, considering their early investment in Chainlink, Synthetix, and a platform similar to the NBA Top Shots, before it was an established venture.
AI Arena, also backed by Paradigm, cleverly marries the concepts of popular games like Super Smash Brothers and Axie Infinity, with the futuristic applications of AI modeling and blockchain. The AI models within the game are tokenized as NFTs (non-fungible tokens). This means they can be trained by players to become increasingly adept and then either flipped for a profit or rented out to beginner players. Tokenizing these AI models presents limitless endeavors for not only gaming but also for business applications.
Spencer hypothesizes that these tokenized AI models may become some of the most valued assets in blockchain. AI Arena’s COO, Wei Xie, described how the formation of the gaming concept came to be. Xie and his cofounders, Brandon Da Silva and Dylan Pereira, recognized the potential of integrating gaming with AI models and blockchain when NFTs became mainstream. The shared vision resulted in the creation of AI Arena where players have the unique opportunity to contribute to AI model development while partaking in an engaging game.
Training AI Models through Gaming
Xie explains the training procedure within the game. Initially, the player demonstrates the task the AI model is expected to perform in the game, similar to a parent instructing a child. Secondly, the context must be provided to the model under different conditions, for example, instructing the AI when to pass or when to shoot for a goal in the game. The final step involves the AI demonstrating its learned skills in the game and the player diagnosing areas for improvement. The cycle then repeats making the AI increasingly adept in the game.
The uniqueness of the AI Arena lies in the game dynamics. It is not solely reliant on the quantity of resources for the AI model. The game fosters creativity, making it a crucial determining factor for the player’s victory. AI Arena, currently going through closed beta testing, is planned for its mainnet launch in the first quarter of next year.
AI Models as a Cryptocurrency
Following the roots of the game in blockchain couldn’t have been complete without the introduction of a token. The token will be distributed to players competing in the launch tournament and later used to enter subsequent competitions. Xie envisions a future for this technology where AI models can be transferred and utilized in varying setting such as a first-person shooter game or a soccer game. This allows the best models to be deployed in different situations, creating a marketplace for crowdsourced AI models.
Limitations in AI Models
A recent analysis by AI startup Vectara revealed a well-known truth about AI models: they are not entirely reliable. Earlier, there was no tangible method to measure the extent of inaccuracies in the model’s generation of information. However, Vectara’s analysis revealed that even the most accurate model, GPT-4, produced around 3% of inaccurate information. Other models like Meta’s Llama and Anthropic’s Claude 2 generated 5% and 8% of incorrect information, respectively. A more shocking revelation was Google’s PaLM, generating an astonishingly high amount of inaccurate information in its predictions, roughly 27%. These models have been criticized for falsely reproducing and amplifying information sourced from non-reliable platforms like Reddit.
AI Technologies and their Role in Warfare
The implications of AI technologies have been greatly debated and critically analyzed, especially their role in warfare. Real footage of atrocities in the Israel-Gaza war has shown some AI-generated images, but these have not been the dominant source of public imagery. For example, an AI-generated image of a child gazing at a missile attack with a chilling caption became widely circulated. Another image depicting children in the rubble gained traction after being shared by a Tunisian journalist. Furthermore, questionable AI-generated images of an “Israeli refugee camp” have been used by Arabic news outlets. With growing scrutiny over the authenticity of images, tools are being designed to help users distinguish real photographs from AI-generated ones. Several tech giants, including Google, Facebook, and Microsoft, are incorporating symbols or watermarks to differentiate AI-generated imagery.
OpenAI Developer Conference and Announcements
OpenAI has recently unleashed GPT-4 Turbo, a significant upgrade to the GPT-4 due to improved speed and the ability to handle longer text inputs. Developers may find access to the updated model at a third of the current cost. OpenAI also announced the launch of the GPT Store, an innovative platform for users to create personalized GPTs. The recent advancements were not without criticism, as the absence of GPT-5 was noted by industry experts who debated the technical limitations leading to this delay.
Grok: The New AI Kid on the Block
Elsewhere, Elon Musk has introduced the beta version of Grok AI. Though not as powerful as GPT-4, it is more current due to being trained on recent tweets. The new AI model provides accurate and contemporary information while also maintaining a judicious answer algorithm. This threshold, as Musk points out, is akin to the availability of information on the internet from reasonable browser searches.
Other Noteworthy AI Developments
Several other significant developments have taken place in the AI arena. Samsung has introduced an AI model named Gauss, expected to be incorporated into its devices soon. YouTube has rolled out AI features for premium subscribers including a chatbot summarizing videos and categorizing comments. Google DeepMind has also released a tier list classifying AI models based on their capabilities.
In conclusion, AI technology’s implications are vast, from gaming to warfare, there are few areas left untouched. Nonetheless, the challenge and the focus going forward should be to develop accurate, reliable, and ethical AI models.