Community Forums

 The rapid evolution of Large Language Models (LLMs) has ushered in a new era of AI-powered chat assistants, leveraging techniques like supervised instruction fine-tuning and reinforcement learning with human feedback (RLHF) to enhance their instruction-following and conversational abilities. These advancements have led to models that are increasingly aligned with human preferences, often outperforming their unaligned predecessors in user satisfaction. However, this progress has brought forth a significant challenge: how to effectively evaluate these sophisticated models. Traditional benchmarks, which typically focus on core capabilities such as knowledge retrieval and problem-solving, fall short in assessing how well these models align with human preferences in real-world applications. We've observed instances where models excel in benchmark tests yet underperform in practical scenarios that demand alignment with human expectations.While human ratings remain the gold standard for eval

 Hello Community Members!October has been an eventful month, filled with engaging activities, insightful news, and exciting events. Let's take a look back at the highlights, dive into our latest webinar recap, and learn how you can contribute your ideas to Boost and AI Studio initiatives. Here’s your October community recap!🎤 Last Webinar RecapZ by HP Boost Webinar On October 4th, we hosted a highly informative webinar focused on maximizing on-premise compute power for AI and ML model training. Akash James demonstrated how he leverages remote GPU access from the Himalayas to work on his multi-modal LLM project, VERONICA. His presentation not only highlighted innovative techniques but also inspired many attendees to explore remote compute solutions for their own AI projects. 🗞️ Latest NewsGartner’s AI PC Shipment Forecast A Gartner report predicts that AI-powered PC shipments will surge by 165.5% by 2025. This forecast has initiated discussions in our community about the future of AI

Gartner projects that AI-powered PCs will hit 114 million units in 2025, marking a massive 165.5% jump from 2024. These AI PCs, defined by their built-in neural processing units (NPUs), include models running on Windows, macOS, and Arm-based systems.By 2024, shipments are expected to reach 43 million units, nearly double the 2023 total. Laptops lead the way, with AI-enabled models predicted to make up 51% of all laptop shipments by 2025, outpacing desktops.Gartner forecasts AI PC shipments to reach 43 million units in 2024, a 99.8% increase from 2023.AI PC Shipments, Worldwide, 2023-2025 (Thousands of Units)     2023 Shipments   2024 Shipments   2025 Shipments AI Laptops 20,136 40,520 102,421 AI Desktops 1,396 2,507 11,804 AI PC Units Total 21,532 43,027 114,225  With nearly half of all PCs integrating AI by 2025, what industries or activities do you think wi

Stanford University research team explores gain cell memory technology to address the limitations of current SRAM and DRAM memory in GPUs. By combining features of both SRAM and DRAM, the goal is to speed up data access and reduce power consumption.Top Three Takeaways:Gain Cell Memory: hybrid gain cells combine the speed of SRAM with the capacity of DRAM, to overcome the memory wall problem in GPUs by reducing data transfer delays. Innovative Materials: combining different materials for transistors (ALD ITO FET and Si PMOS), the hybrid gain cells provide faster data access, non-destructive reads, and significantly longer data retention compared than traditional DRAM. Potential for SoC Applications: can serve as a drop-in replacement for SRAM, offering higher capacity at lower fabrication costs, making it attractive for SoC manufacturers in datacenter GPUs, CPUs, and embedded systems.Check out Chris Mellor article on Blocks & Files: “Stanford team proposes hybrid gain cell memory to

As we wrap up September 2024, we're thrilled to spotlight standout stories, articles, and posts from our vibrant community.  Celebrating Our Z by HP Ambassador of the Month: Javier Eluney Hernández 🎉 We were thrilled to honor Javier Eluney Hernández as our very first ever Z by HP Ambassador of the Month. Javier has significantly advanced AI research, particularly in Computer Vision and Generative AI applications within healthcare.With a sustained passion for exploring the frontiers of machine learning, Javier's work spans computer vision, AI in healthcare, and natural language processing. His notable projects include the Medical QA LLM, aimed at providing accurate, context-aware answers to medical inquiries. Javier currently holds the position of Data Engineer II at Brookfield Renewable U.S.. Webinar Announcement: Z by HP Boost Z by HP Boost brings enhanced performance to your data science teams by offering instance access to idle GPU resources. Don't miss our exclusive webinar on Oct

With the wrap-up of HP Imagine yesterday, AI and remote work took center stage in every new product announcement. One key highlight was the introduction of Z by HP Boost, a solution that enables teams to share on-demand GPU power from their combined fleet of local workstations. Additionally, HP is expanding AI Studio to strengthening LLM development with a “built-in layer of trust” by protecting against hallucinations and bias.HP also announced several exciting AI-powered PC, audio, and visual products, including the new HyperX Cloud MIX™ 2 headset, Poly Labs, and the HP OmniBook Ultra Flip 2-in-1 Laptop Next Gen AI PC, to name a few that will elevate both work and play experiences.🎉 Overall, these are exciting strides in AI for HP, coming in November 2024 and the first quarter of 2025.Check out the full press release here: HP Transforms the Future of Work | HP® Official Site

This week Anbu Valluvan Devadasan et all published a paper titled “E-SMOTE: Entropy Based Minority Oversampling for Heart Failure and AIDS Clinical Trails Analysis.”I reached out to Anbu to chat about the Entropy-based Synthetic Minority Oversampling Technique (SMOTE) which uses a subset of the dataset from the minority class to address overbias of a majority class. Anbu explained that the team used entropy as a metric to assess influence of a minority class on outcomes and outperform conventional SMOTE oversampling techniques in some cases. Results illustrated that both SMOTE and E-SMOTE techniques should be considered for addressing imbalanced data samples as they can be used to address class imbalance. Anbu is excited and looking forward to dig deeper into which data characteristics affect metrics and how.Anbu’s open to chat about his work. If you’re interested, ping him - linkedin.com/in/anbuvalluvan. S. Veerla, A. V. Devadasan, M. Masum, M. Chowdhury and H. Shahriar, "E-SMOTE: Ent

Hello everyone!I just created a blog post about exploring current Medical LLMs. If you are interested about AI in Healthcare and GenAI, and would like to learn how to create demos with Python and Gradio, I invite you to read it! https://medium.com/@miracfence/exploring-fine-tuned-medical-llms-cbb42dd4911f

 According to Gartner, by 2027, 40% of generative AI (GenAI) solutions will feature multimodal capabilities—encompassing text, images, audio, and video—up from just 1% in 2023. This transition from single-modal to multimodal models promises to improve human-AI interactions and create unique opportunities for differentiated GenAI-powered products and services.The Rise of Multimodal GenAIMultimodal Generative AI is set to revolutionize enterprise applications by introducing innovative features and capabilities previously unattainable. Its influence spans across various industries and applications, enhancing interactions at every point where AI meets human engagement. Currently, many multimodal models incorporate only two or three modalities, but this is expected to expand significantly in the coming years.Open-Source LLMsOpen-source large language models (LLMs) are transforming enterprise value by democratizing access to advanced generative AI. They enable businesses to tailor models for

As it is already well-known, for the last couple of years companies have been using AI in films in order to optimize multiple processes, from computer graphics, to animation, etc. Today, however, we can see it being hugely involved in the process, even in the whole automation of creating a film. For some artists, this may be a huge concern, for others it MAY be an opportunity to explore their creativity. In posts and programs like this https://runwayml.com/news/runway-partners-with-tribeca-festival we can see the potential in GenAI applications in the film industry, as long as the latest research https://runwayml.com/research/introducing-gen-3-alpha  What do you think that are the actual limits of this? Do you think it will eventually ceil and just become one more tool for humans to explore their own creativity? Or, do you think it will become something like that one Black Mirror’s episode “Joan is Awful” where content is being automatically generated without any restrictions, just for

Some years ago, one of the things that was hyped in the AI Community but that I still think has huge relevance for the future in AI is Federated Learning and Privacy Preserving Machine Learning (PPML). This comes from an idea that future devices will have more than enough power to run the majority of our biggest models, or that models will be efficient enough to run on them. This without losing the aspect of keeping our data private and secure.What do you think about this? Do you think this is where AI Development would head eventually? What others directions could be the most feasible and with the most usage our future? A couple of those that come to my mind are Embodied AI and Causal Machine Learning.

HP Data Science Ambassador Akash James uses a zero-shot approach to create Veronica to be aware of her spatial surroundings through visual, audio, and text interactions to describe what she sees. Check out his LinkedIn post to see a demo of how it works…  

OpenAI announced the launch of fine-tuning for GPT-4o, addressing one of the most frequently requested features from developers. This new capability allows developers to fine-tune the GPT-4o model with custom datasets, enhancing performance while reducing costs for specific applications.To support the rollout, OpenAI is offering 1 million free training tokens per day for every organization through September 23. This incentive is designed to help developers explore fine-tuning without immediate cost concerns.Developers can begin fine-tuning by visiting the fine-tuning dashboard. The process involves:Accessing the dashboard. Clicking "Create" to start a new project. Selecting gpt-4o-2024-08-06 from the base model options.GPT-4o Mini Also AvailableOpenAI also offers GPT-4o Mini for fine-tuning. This smaller model is available with 2 million free training tokens per day until September 23. To use it, developers should select gpt-4o-mini-2024-07-18 on the fine-tuning dashboard.Data Privacy

Maikel Ronnau et al will publish Automatic segmentation and classification of Papanicolaou-stained cells and dataset for oral cancer detection in next month’s Computers in Biology and Medicine.Their research uses a CNN U-Net architecture with ResNet encoder to achieve expert level performance by segmenting and classifying a trained on UFSC OCPap dataset of oral cytology images. The team uses transfer learning for image segmentation to train the model which is used to evaluate more than 1500 images from 52 patients labeled by specialists resulting in an average Dice score of 0.66 and Intersection over Union (IoU) of 0.65. “The CNN model architecture for encoding layers are based on DenseNet-169 while the decoding layers are LinkNet but to replace the regular softmax layer with a temperature scaling softmax with a temperature parameter value of 0.1 to increase the confidence of the predictions and avoid the bias towards the prediction of background pixels. The model’s prediction is furth

GPU availability and job interruption can be an issue for me, so this paper caught my attention. The paper is titled “Mirage: Towards Low-interruption Services on Batch GPU Clusters with Reinforcement Learning” and it proposes a method for reducing service interruptions on GPU clusters for deep learning jobs.The tool is called Mirage. Mirage is a Slurm-compatible foundational model that uses statistical and reinforcement learning (RL) techniques to proactively provision resources to mitigate interruptions. To accomplish this, researchers alter the decoder transformer architecture to create a Mixture of Experts (MoE) of neural networks. The neural networks share the same network architecture to support scaling up of model capacity and determine ‘best-fit experts’ that optimize job trace analysis outputs for resource allocations. What’s really cool about this paper is how it investigates the use of Deep Q-Learning RL techniques to reenforce rewards to improve GPU queue wait times for a f

At last month's Figma annual conference, Config, several new features were unveiled, including Figma AI. Designers like me have been eagerly awaiting updates from the collaborative design tool giant, especially after their deal with Adobe fell through. The question now is: how will Figma lead the way for creatives to design with AI as a supportive tool? How will it be different from what we’ve seen from existing tools?The new features are impressive. They include a text translation feature to visualize product designs in different languages and Figma Slides, a natural business and feature decision for Figma as designers are already presenting in Figma or creating slide decks rather than migrating to Google Slides or PowerPoint, as can be seen from the Figma Community templates available. Additionally, a redesign of Dev Mode allows developers to focus on details like new published updates.The most significant feature is Figma AI, where users can click a button to "Make a design," enter