Quality over Quantity

Since its launch, OpenAI’s new model Q*, with its tremendous maths-solving ability and enhanced reasoning capabilities, has divided the entire AI community. The bone of contention is the heavy use of synthetic data to train Q* that might hinder the AI community’s dream of achieving AGI.

OpenAI is a big proponent of synthetic data. The team of Ilya Sutskever and former Tesla AI head Andrej Karpathy has been a significant force in leveraging synthetic data to build AI models. The development at OpenAI is testimony to the advanced growth of generative AI in the entire ecosystem, but not everyone agrees that they will be able to achieve AGI with the current methodology of model training.

Meta AI chief Yann LeCun stands as a prominent critic of LLM and synthetic data. According to him, a key obstacle in enhancing LLM reliability is the need to substitute auto-regressive token prediction with planning. LeCun has consistently advocated for an extended period that, in order to attain AGI, the focus should be on improving the reasoning capability of LLMs instead of merely increasing the volume of data.

LeCun loves to give the example of humans and animals, when it comes to the learning curve as they get smarter with immensely smaller amounts of training data. In his view, the current ML architecture is far from perfect. He advocates for the development of new architectures that could learn more efficiently, akin to humans and animals, which is a key area of research in AI.?

He is not the sole individual expressing disinterest in synthetic data. Bojan Tunguz, a machine learning scientist at NVIDIA, asserts, "For tabular datasets, where I have the most experience, synthetic data is not just ineffective, but detrimental. I've encountered similar feedback from those using it to train autonomous vehicles."?

Similarly, as per Jim Fan, senior AI scientist at NVIDIA, while synthetic data is expected to have a noteworthy role, blind scaling alone will not suffice to reach AGI.

Read the full story here.

GeoAI: From Maps to Impact

The integration of AI is propelling the rapid evolution of geospatial infrastructure, enabling automated data analysis, pattern recognition, and predictive modelling. In this dynamic landscape, emerges Esri (the Environmental Systems Research Institute), a comprehensive geographic information systems (GIS) solutions provider.?

Within ArcGIS, Esri provides GeoAI, furnishing pre-built models designed for seamless utilisation with various data types such as vector, tabular, imagery, and text. This integration of AI and spatial analysis empowers industries to simulate outcomes across a spectrum of scenarios. Globally, it has successfully delivered GIS solutions to over 350,000 organisations, boasting a clientele of more than 5,000 customers in India.

Read the full story here.

Pin vs Glasses?

During a recent Meta Instagram live session, Humane co-founders Imran Chaudhri and Bethany Bongiorno discussed their choice of the Ai Pin's form factor. Chaudhri emphasised the fading popularity of glasses, citing numerous attempts by others in the past.?

Despite Meta's introduction of Ray-Ban smart glasses offering similar functionality, the Ai Pin emerges as a unique alternative. Perceived as futuristic, the Ai Pin challenges the notion of glasses becoming obsolete. Chaudhri questioned the appeal of glasses, suggesting they are optional fashion accessories, unlike the Ai Pin designed for practical purposes.?

Read the full story here.

Govt Buckles Up

After a series of deepfake incidents, the Indian government has buckled up and is talking regulations. During the Digital India Dialogues, Rajeev Chandrashekar, Union minister of state for electronics & technology, underscored the right of Indian citizens to a secure and trustworthy internet, with intermediaries being held accountable to users.?

The discussions yielded significant outcomes: Firstly, all platforms and intermediaries committed to addressing deepfakes within the framework of existing laws while also considering potential new regulations. Secondly, within the next seven days, platforms pledged to update terms and contracts with users to explicitly prohibit 11 types of content outlined in the IT rules.?

Read the full story here.