Welcome to the AI Index Report, featuring practitioner analyst and entrepreneur Toni Witt. This series is intended to deliver the timely intelligence about artificial intelligence (AI) you need to get up to speed for an upcoming client engagement or board meeting.
OpenAI has been working on an innovative research project called Q*. Researchers reported that Q* could solve several math problems completely on its own. Although the math problems may have been simple, they still showed promise, particularly the ability of the machine to reason. Once these models reach sufficient reasoning abilities, can we define them as smarter than humans?
With reasoning also comes the ability to plan and act as an independent agent. If that agent doesn’t have good intentions, harm or risk components come in. As these companies continue innovating forward, researchers and developers must be aware of potential risks.
Six-month-old video AI startup Pika Labs announced that it raised $35 million in a Series A round of funding. Along with other rounds, the young company raised a total of $55 million.
Pika Labs also unveiled Pika 1.0, its first product. Pika 1.0 is a web-based platform enabling users to generate diverse video content, including 3-D animations, anime, cinematic, or live-action content. Pika 1.0 will support text-to-video, image-to-video, and video-to-video.
Other big names are building generative AI tools for creators. For instance, Adobe recently acquired Rephrase.ai to provide generative AI tools for video. While it may be a while before we see fully AI-generated videos come out, the development of these tools is happening faster than anticipated.
These tools can serve several purposes. Creative teams may use these tools for marketing use cases, whether they publish the final AI product or use it to test out ideas.
Solution of the Week (07:15)
Large language models (LLMs) are trained on reams of data that are mostly pulled from the internet. There is a distribution curve for certain languages that don’t have as much online content written in them. As a result, they don’t have much representation in training data, leading to LLMs not being as good at generating text in those languages. This makes generative AI tools less accessible to speakers of those languages.
A team of researchers in South Africa launched a new venture called Lelapa AI to tackle this issue for African languages. Its first product is Vulavula. It primarily works with voice-to-text and detecting names in four languages. The researchers work with both online and offline linguists as well as the broader community to gather data, annotate it, and train the models.