Author: Ashish Kumar, Chief Data Scientist, Indium Software
– Generative AI has emerged as one of the most promising applications of AI in real world. It was believed that creative fields would be tougher for AI to learn but opposite has happened; with advent of Whisper AI (for super-accurate multilingual voice to text transcription), chatGPT (LLM for human like conversation and content generation) and DALLE-2 and Stable Diffusion ( for generating image from text prompts) the creative tasks like art/photo design, marketing ad copies, video editing and subtitling look set to be done fully or highly assisted by AI in near future. Innovation in this field is happening at a breakneck speed led by likes OpenAI and Deep Mind
– Big Techs have done layoffs to make their EBITDA numbers look glossy which have come as an opportunity for startups like Open AI, Hugging Face etc.
– Synthetic data is another trend which is by-product of generative AI revolution and can have an impact on adoption of data science by removing the hurdle of availability of training data to create new models for small and medium sized tech companies.
– In ML world, a lot of focus has shifted to make the trained models work and operate continuously. This is critical as otherwise even the most accurate model is useless. A lot of independent products have started in this space which provide facilities to retain models, detect data and model drifts. MLOps will continue to see a lot of traction just that here it can’t be totally independent of model development part (as it is in usual DevOps)
– Emergence of graph databases and graph neural networks for next generation recommendation systems is another trend to watch out for
– A lot of ML workflows of large enterprises would happen on one of the cloud platforms like AWS SageMaker, GCP Vertex AI or Azure Databricks
– Drug discovery is one of the areas where AI is doing surprisingly very well
– AI explain ability remains one of the major prerequisites for AI adoption. It is well accentuated by the need for AI safety (content bias and biased decision-making) and AI auditing
– Hugging Face is leading the way in adoption of ML and AI models by creating a repository and ecosystem of pretrained models for tasks like sentiment analysis, question answering, paraphrasing, text classification etc
