Newsletter: December 2023
The latest Deeper Insights blogs
Privacy in the age of ChatGPT: Understanding LLM data protection
Strategies for balancing privacy with AI and LLMs like ChatGPT. Learn about enterprise solutions, HIPAA compliance, Zero Data Retention, and the advantages of custom, locally-hosted AI models for optimal data security.
Using custom large language custom models to solve customer problems
Explore the benefits of Custom Large Language Models (LLMs) in customer service, contrasting with Open Source LLMs. Read real-world cases where Custom LLMs, tailored with industry-specific data, transformed customer service operations, ensuring consistency, reduced response times, and cost savings. [Read more]
The transformative power of MLOps: A comprehensive exploration
Discover the transformative power of MLOps, a unified discipline that streamlines the machine learning lifecycle from data collection to model deployment. Learn how MLOps can elevate your machine learning projects, ensuring long-term success, scalability, and sustainability. [Read more]
Featured GenAI news
Google in talks to invest ‘hundreds of millions’ into AI startup Character.AI - CTech.com
Google plans to invest in Character.AI, an AI chatbot startup, deepening their partnership. Character.AI, known for creating personalised AI companions, particularly appeals to younger audiences and is rapidly growing in the tech space, reflecting the rising interest in interactive AI technologies. [Read more]
Explaining: Generative AI - MIT News
Generative AI, exemplified by tools like OpenAI's ChatGPT, is making a significant impact with its ability to produce human-like text. This advanced form of AI, different from traditional predictive models, is designed to create new data that mirrors its training inputs. Despite recent excitement, this technology builds on over 50 years of machine learning and computational research, showing its deep-rooted evolution. It's a testament to AI's growing versatility and potential in various applications. [Read more]
Generative AI’s impact on white-collar work - Financialtimes.com
Generative AI is reshaping white-collar work, offering significant productivity and quality boosts, especially for less skilled workers in high-skilled occupations. While some jobs face automation risks, embracing AI as a collaborative tool can enhance performance, suggesting a positive, transformative impact on the workforce. [Read more]
What early adopters of generative AI have learned - Forbes.com
Generative AI is making a significant, positive impact in the business world, according to LTIMindtree's survey of 450 executives. It's enhancing customer experiences and driving revenue, with a majority reporting substantial growth. Despite some skill shortages, strong leadership and external expertise are paving the way for successful AI integration. Companies are proactively addressing challenges like operational costs and ethical considerations, focusing on responsible AI use and developing industry-specific models for better effectiveness and positive outcomes. [Read more]
GenAI news snapshots - Industry report
- After OpenAI introduced the DALL-E 3 AI image generator in ChatGPT, users noticed amusing bugs revealing internal prompts shared between the image generator and the AI assistant. These included all-caps commands, highlighting the increasingly human-like communication techniques in instructing computers, potentially shaping the future of programming and interactions. [Read more]
- Baidu introduced ERNIE 4.0, a potent AI chatbot comparable to OpenAI's ChatGPT. Demonstrations by CEO Robin Li showcased its capabilities, including creating car commercials, solving math problems, and generating novel plots. ERNIE 4.0 offers improved query understanding, complex response generation, and memory capabilities. [Read more]
- The creators of the data set used to train Stable Diffusion, a text-to-image model, are launching the Open Empathic project. This initiative aims to provide emotion-detecting capabilities to developers for free, promoting empathy and emotional intelligence in open-source AI systems. [Read more]
- President Joe Biden has issued an executive order, outlining guidelines for generative AI safety, in advance of any legislative action by lawmakers. [Read more]
- Amazon's new generative AI empowers advertisers to overcome creative challenges, making it easier to craft engaging ad content. With image generation in beta, brands can enhance their ads' performance by seamlessly integrating products into lifestyle contexts, increasing click-through rates by up to 40%. [Read more]
- Stability AI unveils Sky Replacer, a tool for changing the sky in photos and enhancing image aesthetics for professionals across industries, with a focus on real estate. This user-friendly feature offers an array of sky options to attract more potential buyers. [Read more]
- Meta is unveiling "Gen AI Personas," a range of generative AI chatbots geared towards engaging younger users. These personas, inspired by various characters, include a "sassy robot" and "Alvin the Alien." Meta intends to create multiple chatbots, some for celebrities and productivity purposes, aiming to tap into Gen Z's tech comfort and extend ad-serving opportunities. [Read more]
GenAI tools: LLM models
- Self-RAG: Learning to Retrieve, Generate and Critique through Self-Reflections - Remarkable development, excels in improving the quality and accuracy of LM outputs. It outperforms ChatGPT and retrieval augmented LLama2 on six different tasks by learning to retrieve, generate, and critique information. [Read more]
- MemGPT: Teaching LLMs Memory Management for Unbounded Context - MemGPT is a system that smartly handles various memory tiers in LLMs, allowing for extended context within the model's limited context window. It efficiently manages critical information by storing it in a vector database, making it retrievable for ongoing conversations. [Read more]
- Langchain Templates: Easy and Fast Way to Build a Production Ready LLM Application - LangChain Templates offer the quickest and simplest path to create production-ready LLM applications. These standardized templates serve as reference architectures for a variety of popular LLM use cases, easily deployable with LangServe.[Read more]
- JudgeLM: An Open-source LLM Judge for Evaluating LLM-generated Answers - JudgeLM, an open platform for scalable LLMs judges. Achieving over 90% agreement, it's designed to evaluate LLMs in various scenarios and offers a distributed multi-model serving system with a web UI. [Read more]
- OpenAgents: an Open Platform for Language Agents in the Wild - Designed with accessibility in mind for non-expert users, bringing language agents into everyday life with ease. It's a platform that opens up new possibilities for a broader user base to use and host language agents. [Read more]
- XAgent: an Autonomous LLM Agent for Complex Task Solving - An open-source experimental LLM driven autonomous agent, a versatile and powerful tool for a wide range of applications. [Read more]
- AlpaCare - A fine-tuned LLaMA-series model, demonstrates strong proficiency in both medical and general domains by using a dataset of 52,000 diverse machine-generated medical instructions, MedInstruct-52k. [Read more]
- Nous-Yarn-Mistral-7b-128k - A cutting-edge language model designed for handling long contexts. It's built upon Mistral-7B-v0.1 and pre-trained with the YaRN extension method, offering a remarkable 128,000 token context window. [Read more]
GenAI tools: Visual models
- Introducing GPT-4 with vision (GPT-4V), OpenAI’s latest capability, allowing users to instruct GPT-4 to analyse image inputs. [Read more]
- Segmind introduces SSD-1B, a groundbreaking open-source text-to-image model that outperforms previous versions, boasting a 50% smaller size and 60% faster performance compared to SDXL. [Read more]
- Matryoshka Diffusion Models is a framework for high-quality image and video generation that uses a NestedUNet architecture and progressive training, achieving impressive results at resolutions up to 1024x1024 pixels with limited data. [Read more]
GenAI tools: Everything else models
- SAM And MetaCLIP, a unique project where the capabilities of Segment Anything (SAM) and MetaCLIP were blended to craft an innovative Open Vocabulary Image Segmentation demo. [Read more]
- Meet Fuyu-8B, a compact multimodal model now available on HuggingFace. What's exciting about it? [Read more]
- Simple architecture and training for easy understanding and deployment.
- Tailored for digital agents, supporting various tasks and fast responses.
- Strong performance in image understanding benchmarks like visual question-answering and image captioning.
Let us solve your impossible problem
Speak to one of our industry specialists about how Artificial Intelligence can help solve your impossible problem