NexaQuanta Weekly Newsletter
Welcome to the NexaQuanta weekly newsletter! In our ongoing mission to keep you informed about the latest advancements in AI and technology, we’ve gathered key updates from industry leaders.
This week, we explore IBM’s new Mistral AI model for on-premises deployment, enhancing enterprise data security. We also highlight OpenAI’s latest moderation model, designed to analyse harmful content more effectively, and Google’s updates to its Gemini models, which bring improved performance and significant price reductions.
Stay tuned as we delve deeper into these exciting developments!
IBM Launches Mistral AI for On-Premise Deployment with Watsonx
IBM has introduced Mistral AI with IBM, allowing customers to deploy Mistral AI’s advanced foundation models on their infrastructure.
This launch helps enterprises maintain control over their proprietary data by deploying AI models on-premises rather than relying on third-party cloud services.
This is particularly crucial for finance, healthcare, and government industries, where data privacy and compliance are top priorities.
Foundation models like Mistral AI are trained on billions of parameters, typically using publicly available data.
While this data is helpful in general scenarios, many enterprises need to fine-tune these models with their proprietary data to achieve higher accuracy for specific use cases.
On-premise deployment ensures the data remains secure while allowing businesses to fine-tune models more easily.
IBM’s Mistral Large 2, a powerful general-purpose model, is now available on Watsonx.ai and can be used on-premises. This model is optimised for various complex tasks like Retrieval-Augmented Generation (RAG), non-English languages, and coding support in over 80 languages.
It also supports up to 128k tokens in a single context window, making it highly efficient for enterprise applications.
This new release aligns with IBM’s open, multi-model strategy, providing customers worldwide—especially in Europe—with more flexibility in deploying and scaling their AI solutions responsibly.
With Mistral AI available as a Software-as-a-Service (SaaS) and on-premises, IBM offers innovative solutions for enterprises looking to build and deploy generative AI models efficiently.
Check this link for more details.
IBM Cloud Introduces NVIDIA H100 Tensor Core GPU Instances for AI Workloads
IBM has launched new cloud instances featuring NVIDIA H100 Tensor Core GPUs, designed to accelerate AI and High-Performance Computing (HPC) workloads.
These instances are ideal for businesses scaling their Generative AI projects from pilot to production. They offer high-performance solutions tailored for various stages of the AI lifecycle.
Depending on their specific use cases and budget, customers on IBM Cloud can now choose from a range of NVIDIA accelerators, including the H100, L40S, and L4.
The H100 instance, known as gx3d-160x1792x8h100, features eight GPUs connected with NVLink technology, paired with Intel Sapphire Rapids CPUs, and provides 61TB of storage.
This setup offers significantly improved performance over previous NVIDIA A100 instances.
To enhance accessibility, these H100 GPU instances are now available in 9 multizone regions across the Americas, APAC, Europe, and Japan. IBM Cloud also ensures multi-level security for AI workloads, safeguarding against data leaks and privacy violations.
In addition to GPU advancements, IBM is strengthening AI governance with its watsonx.governance tool.
This platform helps manage AI model lifecycles by ensuring models are validated for risks before deployment and continuously monitored for fairness and quality once live.
This transparency aids regulators and auditors by providing detailed documentation of AI models’ behaviour.
IBM Cloud further simplifies AI-powered deployments by automating essential services like lifecycle management, storage, and security, reducing manual configuration and minimising errors.
This allows businesses to operationalize AI applications securely and efficiently.
Click here for more details.
IBM Launches Reddit Megathread on Foundation Models and Granite AI
IBM has released a detailed mega thread on Reddit, providing valuable insights into how foundation models transform businesses.
This discussion focuses on IBM’s Granite family of models, which are designed to simplify complex AI tasks. It also helps enterprises maximize their AI investments.
Key takeaways from the thread include:
1). Foundation Models Explained: A simplified breakdown of foundation models and how they work.
2). Introducing IBM Granite: An exploration of the features and benefits of IBM’s Granite models.
3). Model Selection Tips: Best practices and guidance on choosing the right AI model for your business needs.
This mega-thread serves as a comprehensive resource for businesses curious about integrating AI models into their operations. Visit Reddit to dive deeper into the conversation.
OpenAI Launches New GPT-4 Based Moderation Model for Text and Images
OpenAI has introduced a new AI moderation model called “omni-moderation-latest,” based on GPT-4.
This model can analyse text and images for various types of harmful content. It also improves the accuracy of previous text-only moderation models.
It is especially effective in moderating non-English languages and includes new harm categories, such as “illicit” content, which covers advice on committing wrongdoing, violent or not.
This update, available through OpenAI’s Moderation API, is designed to help developers create safer applications as the amount of generated text and images grows rapidly.
The new model is a free upgrade, further advancing the tools for moderating harmful content across platforms.
Read details by visiting this link.
Google Updates Gemini 1.5 Pro and Flash with Improved Performance and Lower Prices
Google has announced updates for its Gemini 1.5 Pro and Gemini 1.5 Flash models, delivering performance enhancements and price cuts.
The new versions show a 7% boost in MMLU-Pro scores, about 20% improvement on MATH and HiddenMath benchmarks, and 2-7% gains in visual understanding and Python code generation tests.
Along with these performance improvements, Google has slashed the price of Gemini 1.5 Pro by 50%.
Both models now feature increased rate limits and faster output speeds, enabling developers to process longer documents. Also to analyse large codebases, and generate content from hour-long videos more efficiently and affordably.
Get further updates here.
Microsoft Launches “Correction” to Address LLM Hallucinations and Errors
Microsoft has unveiled “Correction,” a new Azure AI Content Safety feature to reduce ungrounded AI-generated content.
This system employs a two-model approach: a classifier model first identifies potentially incorrect or irrelevant text snippets, and then a language model rewrites these sections based on specified grounding documents.
Correction is compatible with various text-generating AI models, including Meta’s Llama and OpenAI’s GPT-4. It is particularly beneficial in fields like medicine and science, where accuracy is vital for reliable outputs.
However, some critics argue that this approach does not tackle the underlying issue of AI hallucinations. They caution that it may create a false sense of security and introduce new problems, as the correction models may also be prone to errors.
For details, visit the link.
Stay Connected: Subscribe to NexaQuanta Weekly Newsletter!
Thank you for reading the NexaQuanta weekly newsletter! We hope you found the updates insightful and valuable for understanding the rapidly evolving AI landscape.
Subscribe to our weekly newsletter to stay informed about the latest trends, technologies, and innovations in the AI industry.
By joining our community, you’ll receive exclusive insights, expert analyses, and tips to help you navigate the complexities of AI.
Don’t miss out on our future updates—subscribe today and join the NexaQuanta journey!