AI Karyashala logo Sign up

Unit V · Applications and the Future

21. Future trends in Generative AI

Generative AI changes every few months. This chapter lists the main directions. Some are already here. Some are coming. This is Lab experiment 20.

1. AI agents

Today you ask the AI one question and get one answer. Tomorrow you will give a goal, and the AI will do many steps on its own: search, read, write code, run it, fix it, send an email. These are called agents. Early agents already exist for coding, shopping and research. Engineers will spend more time checking AI work and less time doing it by hand.

2. Multimodal everything

Models now see, hear and speak. Soon they will live in glasses, cars, earphones and robots. You will point your phone camera at a machine and ask "what is this part?" You will talk to AI in Telugu and it will reply in Telugu, with the right accent.

3. Small models on your device

Big models run in data centres. Small language models (SLMs) run on a phone or a laptop with no internet. Examples: Gemini Nano on Android phones, Apple's on-device models, Llama and Gemma that you can run with Ollama. This is faster, cheaper and private. Good for villages with poor internet.

4. Indian language AI

India has 22 official languages. Most AI was trained mostly on English. This is changing fast.

  • Bhashini (bhashini.gov.in) — the government's AI translation platform.
  • Sarvam AI, AI4Bharat (IIT Madras), Krutrim — Indian companies and labs building Telugu, Hindi and other language models.
  • The IndiaAI Mission is funding GPUs and Indian models.

Engineers from Andhra Pradesh who know Telugu and English well will be needed for this work.

5. AI in science and engineering

  • AlphaFold (Google DeepMind) predicted the shape of almost every known protein. It won a Nobel Prize in 2024.
  • AI is designing new materials, batteries and medicines.
  • AI helps weather forecasting. This matters for cyclones on the Vizag coast.
  • In civil, mechanical and electrical engineering, AI is used for design, simulation and predicting when machines will fail.

6. Video and 3D generation

Text-to-video models (Veo, Sora, Kling) are getting good enough for advertisements and short films. 3D model generation is coming. This will change film, gaming and architecture.

7. AI and jobs

Some jobs will change a lot: writing, customer support, basic coding, design. New jobs will appear: prompt engineers, AI trainers, AI safety testers, AI product builders. The safest plan is to be the person who uses AI well in your own field.

8. Rules and safety

Governments are making laws. The EU AI Act is the first big one. India has the DPDP Act for data. Companies are adding watermarks to AI images and videos so people can tell what is fake. Expect more rules about deepfakes, copyright and AI in exams.

9. Open models

Some companies release their models for free: Meta's Llama, Google's Gemma, Mistral, DeepSeek, Qwen. Anyone can download and run them. This lets students, startups and colleges build without paying per use. See huggingface.co.

10. The cost is falling

In 2023, using a top model cost a lot. Now a model of the same quality costs 100 times less. This means AI will be in every app, even small ones made by students.

How to keep learning

Try it

This is Lab experiment 20. Pick one recent AI technology. Examples: NotebookLM Audio Overview, Gemini Live, Veo video, Ollama on your laptop, Bhashini, AlphaFold, an AI agent.

  • Use it yourself for 30 minutes.
  • Prepare a 5-slide demo or a one-page report with: what it does, how it works (simple), 3 uses, 3 benefits, and 3 responsible-AI concerns (Chapter 6).
  • Install Ollama and run a small model on your laptop with no internet. Ask it a question in Telugu.
  • Explore huggingface.co/spaces and find one demo from an Indian lab.

New words

  • Trend — a general direction in which things are changing.
  • Agent — an AI that can plan and act on its own to reach a goal.
  • On-device — running on your phone or laptop, not on the internet.
  • SLM — Small Language Model; a small version that runs on normal devices.
  • Data centre — a building full of computers where big AI models run.
  • Open model — a model whose weights are free to download and use.
  • Watermark — a hidden mark in an image or video that shows it was made by AI.
  • Simulation — using a computer to imitate a real process, like a bridge under load.
  • Hackathon — a short event where teams build a project in a day or two.
← previousAI workflow automationnext →Lab experiments guide