Spatial Models with LLMs Are Needed Now

Spatial Language Models have their own unique uses.

The same question people are bugging ChatGPT to solve, viz. “Drawing a map of the USA” can be addressed with the help of Spatial Models and LLM combined.

The reason is that SpatialModels understand object detection and identification of real-world entities. Therefore, we need to train it to recognize complex objects as well, such as fragmented regions on maps of countries and depiction of rivers and lakes within a state.

Once completed, it can also be useful in flood rescue operations with live satellite data.

Right now, Gen AI can’t draw and compute many things.

One challenge Gary Marcus gave to chatGPT is counting the apples and rounding to the correct answer. This, they say, is taken from Kindergarten books. This can be solved with object identification schemes in spatial AI.

So, why not collaborate?

Collaborate on what- Collaborate Spatial Models with LLM?

What can it solve? Apart from the examples mentioned below, think of a situation where there is a man in your garden who comes to the door and rings a bell. You need AI to tell you who is at the door. Can LLM alone, like ChatGPT, solve this? No, we are not there yet. But we want to get there. This is where spatial fragmentation of information and its conversion into human language through LLM come into play. Then, you can talk to LLM to decide what to do next. Or LLM can decide it for you if kids are inside.

Even the objects in prescriptions, charts, architecture, electric wiring, mathematical diagrams, and designs can be identified in human language with this combination. Then one can question GenAI about the knowledge it gained while parsing the information.

The values and numbers in human-made arrows in charts and diagrams, such as doctors’, architects or Professors’ notes, can be analyzed using spatial AI, which can segment and interpret figures and notes on pages. Currently, Gen AI can’t solve curve-based charts and their inference with full accuracy.

Analysis of human-made curved graphs can be made once this is understood.

Complex political and economic entities can be converted to text with the modified spatial LLM.

Meteorological, cyclones, and storms’ real-time satellite data can be understood once a modified spatial LLM is created.

Other areas include assistance in robotics, mining, forest services, and fire services, to name a few!

We might need to retrain the Spatial Models to improve their understanding, using services like Google Maps and Google Earth.

This can help identify where forest fires started, where roads are in poor condition, aid in crime detection, and analysis of design sketches.

It’s time to solve these problems!

So, time to tie laces and move on!

Thanks for reading.

Subscribe for updates.