Artificial Intelligence (AI) is still transforming many facets of society, with natural language processing (NPL) leading the way. The development of Large Language Models (LLMs), which use deep learning algorithms to comprehend contextual subtleties and comprehend text inputs, is one noteworthy advancement.

Understanding Large Language Models: These vast neural networks with millions or billions of parameters that have been fine-tuned over large datasets are known as LLMs. Their remarkable communication talents stem from their ability to comprehend intricate grammatical structures and linguistic nuances due to their sheer size.

Key Features of LLMs:

• Context-awareness: Unlike simpler NPL models, LLMs comprehend broader contexts rather than treating each input independently.
• Multilingualism: Several LLMs handle multiple languages, broadening their applicability worldwide.
• Customizability: LLMs can be fine-tuned according to specific downstream applications, making them versatile and adaptable.

Applications of LLMs:

• Chatbots and virtual assistants: LLMs facilitate lifelike conversations, delivering superior user experiences compared to rule-based counterparts.
• Content generation: From drafting emails to composing blog posts, LLMs demonstrate potential as automatic writers for various formats.
• Text classification: LLMs prove adept at categorizing vast volumes of documents efficiently.

New Developments: Research on reinforcement learning with human feedback (RLHF) allows model outputs to be aligned with society ideals, reducing negative reactions. As this is going on, there is increasing interest in multi-modal LLMs, which increase the usefulness of text inputs in multimedia contexts by including visual and auditory signals. 

Handling Ethical Obstacles: Although the future looks bright, there are still a lot of ethical questions around responsibility, openness, justice, and privacy with reference to LLMs. Building reliable and accountable LLMs involves addressing biases present in training datasets, bringing attention to potentially discriminatory outcomes, and using third-party audits. 

In short, large language models show remarkable potential for contextually rich conversations and versatile industry adaption, making them an interesting new direction in natural language processing. Taking proactive measures to manage ethical concerns and keeping up with advancements guarantees that they can be fully utilized to drive significant innovations.