Welcome to the latest edition of AI Fridays by HackerPulse in collaboration with Kusho. ML engineers from Kusho have prepared a roundup of interesting stories and news in AI you shouldn’t miss.
Here’s this week’s roundup:
🔌 LLM Powered Code Interpreters
💬 Chatbots from Eliza to ChatGPT
💡 The 1-Bit LLM Era
🤖 AI Search Engines Can’t Kill Google
❄️ Snowflake partners with Reka to support its suite of highly capable multimodal language models in Snowflake Cortex.
LLM-Powered Code Interpreters (🔗 Read Paper)
When discussing projects with E2B, developers often ask what exactly a "code interpreter" means in the realm of AI agents. In the traditional sense, a code interpreter is a program that reads and carries out instructions written in a programming language. However, with the rise of AI agents, we now have interpreters that include an AI intermediary, translating human prompts into code instructions. These AI-powered interpreters not only generate code but also execute it, performing tasks like creating files, downloading files, or generating charts based on provided data. While many popular AI products still don’t have code interpreter abilities, this new circuit of code generation is poised to shape the future of software. Some popular examples include Open Interpreter, Autogen agents by Microsoft, and ChatGPT's Code Interpreter feature.
From Eliza to ChatGPT (🔗 Read Paper)
From Eliza to ChatGPT, the journey of building chatbots spans over 60 years. Since the 1960s, developers have aimed to create more human-like ways to interact with computers. Eliza, created by Joseph Weizenbaum, acted as a therapist, showcasing the potential of conversational computing. Over time, we've witnessed numerous attempts at human-like chatbots, from SmarterChild to voice assistants like Siri and Alexa. Examples from pop culture, such as Star Trek’s computer and Scarlett Johansson's AI in Her illustrate our fascination with human-like interaction with technology. Now, tools like ChatGPT/Co-pilot represent the closest we've come to conversational computers, explaining why the chat interface has found AI Product-Market Fit (PMF).
The Era of 1-Bit LLMs (🔗 Read Paper)
Recent advancements in AI have led to the development of 1-bit Large Language Models (LLMs), such as BitNet b1.58, which represents a significant breakthrough in reducing the computational cost and energy consumption of LLMs while maintaining performance. BitNet b1.58 introduces a new scaling law and training approach that allows for high-performance and cost-effective LLMs. Unlike traditional LLMs, BitNet b1.58 utilizes ternary parameters {-1, 0, 1}, resulting in improved efficiency in terms of latency, memory, throughput, and energy consumption. Remarkably, BitNet b1.58 matches the performance of full-precision Transformer LLMs while being significantly more cost-effective. This groundbreaking approach defines a new standard for training LLMs, paving the way for future generations of high-performance and cost-effective models.
Why AI Search Engines Can’t Kill Google (🔗 Read Paper)
The evolution of AI-powered search tools is reshaping the way we access information online. AI search engines exhibit their strength in delivering more comprehensive responses to exploration queries aimed at understanding concepts or processes. They synthesize information from various sources and provide enriched content, supplemented with multimedia elements like images. However, despite their potential, AI search engines still grapple with speed and interface issues, highlighting the nuanced challenges of revolutionizing the search experience. Nowadays, Google remains unparalleled in efficiently handling navigational queries, where speed is paramount. Google's dominance also shines through in the realm of information queries, where accuracy and timeliness are crucial and the handling of live updates. Despite the potential of Generative AI to outperform search technology, modern search engines have evolved beyond basic link-based results. They now resemble miniature operating systems, equipped with multifunctional capabilities that enrich the user experience beyond mere search functionality.
Snowflake Partners With Reka to Support Multimodal Language Models In Snowflake Cortex (🔗 Read Paper)
Snowflake partners with Reka to support its suite of highly capable multimodal language models in Snowflake Cortex. These models, including Flash and Core, allow users to incorporate images and video alongside text for better contextual understanding. Recent testing demonstrates Core's impressive performance, comparable to top models such as GPT-4 and Gemini Ultra. Snowflake's collaboration with Reka and NVIDIA underscores its commitment to providing secure, easy-to-use AI solutions for customers. Customers can leverage multimodal AI within the Snowflake Data Cloud, unlocking value from diverse data types while benefiting from robust security measures.
And that’s a wrap! See you next time, humanoid. Make sure to send me to a fren 🐸🐸