Kamal Shrestha (कमल श्रेष्ठ)

In a flurry of technological breakthroughs, the latest advancements in AI have been unveiled. Google introduces Med-Gemini, a groundbreaking multimodal model revolutionizing medical analysis, while Gradient’s LLama-3 wows with an unprecedented 1M context window. OpenAI partners with Stack Overflow to integrate technical wisdom into ChatGPT, and mysterious whispers surround the emergence of "gpt2-chatbot." Meanwhile, Claude debuts its iOS app and Team accounts, and GitHub launches Copilot Workspace, streamlining coding processes. OpenAI’s CEO, Sam Altman, shares insights on GPT-4, hinting at a transformative future. Rumors swirl about a potential ChatGPT-based search engine, while ethical debates ignite over OpenAI’s Sora in crafting Washed Out’s "The Hardest Part" music video. Stay tuned as AI continues to reshape industries and spark discourse worldwide.

Latest News:
The recent news in AI and technology highlights significant developments from major players like Apple, Microsoft, and Elon Musk’s xAI. Apple unveils OpenELM, a suite of open-source language models, while also exploring partnerships for AI integration in iOS 18. Microsoft introduces Phi-3 Mini, its smallest AI model, and Elon Musk’s xAI secures substantial funding. Additionally, there are releases of open-source tools from Cohere and Snowflake, along with Play.ai’s conversational Voice AI platform. ChatGPT introduces a new Memory feature, while OpenAI faces a GDPR complaint, raising concerns about data handling and corrections to their work.

Latest News:
Phi-3 Technical Report presents a suite of language models ranging from 3 billion to 14 billion parameters, with the 3B model showcasing superiority over the original ChatGPT. Meta introduces LLama 3, boasting sizes of 8B, 70B, and 400B parameters, trained on a colossal 15 trillion tokens, achieving remarkable performance on various benchmarks. Bigxtral unveils Instruct 0.1, a cost-efficient model with 39 billion parameters out of 141 billion, excelling in language tasks across five languages and offering strong math and coding capabilities. Meta gears up its AI assistant for widespread integration across its platforms, while Mistral seeks significant funding to further its open-source model development. Additionally, Google’s Infini-attention technique enables LLMs to process text of infinite length with consistent memory and compute demands.

AI Prompt Engineering Is Dead Long live DSPy

The blog post discusses the implementation of a retrieval-augmented generation (RAG) agent using langgraph, focusing on differentiating between langchain agents with multiple tools selections and langgraphs. Langchain operates with a sequential acyclic flow of execution, while langgraphs introduce a cyclic, stateful, multi-actor application framework. Langgraph enables re-execution of multiple steps, fostering an iterative decision-making approach. The post provides detailed steps for creating a RAG agent using langgraph, including installations, indexing, defining states, executing nodes, and constructing a workflow graph. Nodes such as agent, retrieve, rewrite, grade_documents, generate, and should_retrieve are explained, each serving specific functions within the agent’s workflow. A state graph is defined to manage the flow between these nodes based on conditions like relevance checking and user input. Examples demonstrate the agent’s functionality, including handling user queries, executing retrieval, assessing document relevance, generating responses, and transforming queries. The output showcases how the agent interacts with the user, retrieves information, and provides relevant responses, illustrating the effectiveness of the implemented RAG agent. Overall, the post serves as a comprehensive guide for implementing a RAG agent using langgraph, offering insights into its architecture, functionality, and practical applications in natural language processing tasks.

The blog explores the rapid advancements in Large Language Models (LLMs) and their implications, dubbed the "LLM race." It highlights the accessibility of LLMs beyond tech giants, their incorporation into various consumer-facing products, and their diverse capabilities, including text generation, data analysis, and even replacing humans in certain tasks. The blog discusses the expansion of LLMs to process multimodal data like images, videos, and audio, pushing the boundaries of what AI can achieve.It touches upon predictions for 2024, speculating whether LLMs will surpass human intelligence benchmarks and the implications for society. Companies like OpenAI, Microsoft, Google, and others are heavily investing in LLM development, with plans for Artificial General Intelligence (AGI). The blog introduces the "Olympics 2024" of LLM development, listing major participants and their advancements.The blog raises questions about the ethical and societal implications of increasingly advanced LLMs, including job market changes, understanding of LLM functionality, and privacy concerns. It concludes by acknowledging the ongoing evolution of LLMs and the need for continued observation and discussion of their impact.

With the rise in accessibility and popularity of various social media platforms, people have started expressing and communicating their ideas, opinions, and interests online. While these platforms are active sources of entertainment and idea-sharing, they also attract hostile and offensive content equally. Identification of hostile posts is an essential and challenging task. In particular, Hindi-English Code-Mixed online posts of conversational nature (which have a hierarchy of posts, comments, and replies) have escalated the challenges. There are two major challenges: (1) the complex structure of Code-Mixed text and (2) filtering the relevant previous context for a given utterance. To overcome these challenges, in this paper, we propose a novel hierarchical neural network architecture to identify hostile posts/comments/replies in online Hindi-English Code-Mixed conversations. We leverage large multilingual pre-trained (mLPT) models like mBERT, XLMR, and MuRIL. The mLPT models provide a rich representation of code-mix text and hierarchical modeling leads to a natural abstraction and selection of the relevant context. The propose model consistently outperformed all the baselines and emerged as a state-of-the-art performing model. We conducted multiple analyses and ablation studies to prove the robustness of the proposed model.

Citizen Journalism has boomed the opportunities for content marketing and public opinion. It is a golden rule of journalism, taught to any news reporter at the beginning of their career-your introduction should grab the reader straight away. Almost all of the online news media outlets completely depend upon the revenues generated from the clicks made by their readers,and due to the presence of numerous such outlets, they need to compete with each other for reader attention. So, In order to attract the readers to click on an article and subsequently visit the media site, the outlets often come up with catchy headlines accompanying the article links, which lure the readers to click on the link. Such headlines are known as Clickbaits. While these baits may trick the readers into clicking, in the long run, clickbaits usually don’t live up to the expectation of the readers, and leave them disappointed. In this work, we attempt to detect clickbaits based on features extracted from the headline of the news and then classify it accordingly. We were able to run a number of extensive cross checking to different news headlines from multiple platforms and were able to obtain an accuracy of 79% in detecting clickbait headlines.