Welcome to BIGDATA AI News Briefs. This timely new feature provides the latest industry insights and perspectives surrounding AI areas such as deep learning, large-scale language models, generative AI, and transformers. We work tirelessly to unearth the most timely and interesting information underlying the day's most popular technologies. We recognize that this field is rapidly evolving and want to provide regular resources to keep you informed. enjoy!
Intel demonstrated competitive AI inference performance in MLPerf inference benchmark results. His three products, the Habana Gaudi2 accelerator, the 4th generation Xeon Scalable Processor, and the Intel Xeon CPU Max series, have significantly increased the competitiveness of AI. These results are in line with the MLPerf Training 3.0 GPT-3 June benchmark results, which validated Gaudi2 as the only viable alternative to H100, and showed that Gaudi2 can outperform Nvidia's H100 on vision language AI models. Based on the Hugging Face performance benchmark.
Key takeaways from the results include:
- Gaudi2 offers attractive performance compared to Nvidia's H100, with H100 slightly outperforming Gaudi2 by 1.09x (server) and 1.28x (offline).
- Gaudi2 outperforms Nvidia's A100 by 2.4x (server) and 2x (offline).
- The Gaudi2 submission adopted FP8 and reached 99.9% accuracy with this new data type.
- For the task of summarizing news articles of approximately 1,000 to 1,500 words in GPT-J 100 words, the 4th generation Intel Xeon processor summarized 2 paragraphs per second in offline mode and 1 paragraph per second in real-time server mode .
- For GPT-J, the Intel Xeon CPU Max series, which offers up to 64 gigabytes (GB) of high-bandwidth memory, was the only CPU that could achieve 99.9% accuracy.
MLPerf results show that SiMa.ai, which provides ML at the embedded edge, outperformed NVIDIA in the closed edge power category. With frames/second/watt being the de facto performance standard for edge AI and ML, these results demonstrate that SiMa.ai's push-button approach continues its leadership in unparalleled power efficiency without compromising performance. It shows what is possible.
Tidio conducted research on AI illusions and what people think about them. Some of the findings are shown below.
- about 96% of internet users know about AI hallucinations, and approximately 86% have personally experienced them.
- around it 93% I believe that AI hallucinations can harm users
- Only 27% blame the users who created the AI hallucination prompts, while 22% blame governments trying to push their own agenda.
- About 48% of people want better user education about AI to combat AI hallucinations, and 47% would vote for stronger regulations and guidelines for developers.
RNDGen is a state-of-the-art random data generator designed to meet the diverse needs of developers, testers, data analysts, and data scientists. RNDGen is more than just a data generator. It is a comprehensive tool that provides over 100 types of dummy data templates, and was created by the company for internal use. Enables users to seamlessly generate large amounts of randomized synthetic test data. JSON, CSV, SQL, XML, and Excel formats are supported.
Transformers as Support Vector Machines – This new research paper establishes equivalence between self-attention optimization geometry and hard margin support vector machine (SVM) problems in transformers. This equivalence is used to characterize the implicit bias of his one-layer transformer, which is optimized with gradient descent. The main problem is understanding the optimization situation and the transformer's implicit biases. Specifically, we wanted to understand how the attention layer selects and composes tokens when trained with gradient descent. The proposed solution uses vanishing regularization to optimize the attention layer and converges towards the SVM solution. The concept of “Attendance-SVM” (Att-SVM) is introduced to separate and select the best token from each input sequence.
AI in TIME – First ever The most influential people in AI The list spotlights leaders, innovators, shapers and thinkers.
IBM rolls out GenAI – New features and models across the WatsonX data platform include data generation, privacy and ethics toolkits, and more. The data used to train the model will also be revealed.
d-Matrix NextGen AI Chip – The radical startup has designed an energy-efficient chip optimized for GenAI and expects an ARR and break-even of more than $70 million within two years. A $110 million Series B round was just backed by Temasek, Playground Global, and Microsoft.
Apple may have arrived late to the AI hype cycle, but don't discount the company just yet. With a history of visionary innovation and significant distribution advantages, the company could quickly break away from its competitors and become the leader of the pack. In fact, Apple started working on conversational AI four years ago. In the meantime, they've been quietly pouring “millions of dollars a day” and what he calls the human resources of four teams into developing language- or image-model-based features. Additionally, Apple plans to incorporate his LLM into Siri to allow users to automate complex tasks using voice commands. Additionally, an “edge AI” approach that runs facial detection AI models on iPhones rather than servers could solve some of the privacy, cost, and speed issues that plague today's LLMs. .
Open Interpreter: Allows Language Models to Run Code on Your Computer – Open Interpreter is an open-source platform that allows Large Language Models (LLMs) to run code on your local machine. Provides natural language interfaces for a wide range of common tasks, including:
- Edit and create photos, videos, PDFs, and more
- Manage Chrome Browser to run surveys
- Generate, clean, and analyze large datasets
Open Interpreter utilizes the function-call language model supported by OpenAI. Primarily, we employ the robust GPT-4 model, but other LLM variants such as Code LLaMA or any HuggingFace model are also available.
Llama 2 can now run for free on Graphcore IPU using Paperspace Gradient Notebook. Llama 2 is the next frontier in open source large-scale language models (LLM) developed by Meta. This will be a game-changer for adoption and commercialization, with performance comparable to larger models and a permissive open source license that allows use and distribution in commercial applications. You can try Llama 2-7B and Llama 2-13B on IPU for free via Paperspace's free tier environment using the Graphcore IPU-PodFour system. This is a great way to start, but you can scale up to a paid IPU-Pod to get the performance you need.16 A system that enables faster inference. It also introduced another powerful and efficient LLM, namely Flan-T5 XXL (and its smaller he 3B parameter, his Flan T5-XL) fine-tuned for Graphcore IPUs.
As competition intensifies in China, Tencent releases AI model for businesses – Chinese technology giant Tencent has released an AI model for businesses called “Hunyuan”. The news comes days after Baidu rolled out a number of AI-powered applications on Tuesday following more supportive regulations. Tencent said it is internally testing Hunyuan AI models for advertising and fintech.
OpenAI will hold its first-ever developer conference on November 6th, the company announced. At the one-day OpenAI DevDay event, which features keynote speeches and breakout sessions from members of OpenAI's technical staff, OpenAI said in a blog post that they would preview “new tools” and exchange ideas, but left the rest to them. Imagination.
AI Insights Forum – The U.S. Congress is back in session and getting serious about leveraging AI. Senate Majority Leader Chuck Schumer's first AI Insights Forum begins on Wednesday, and we'll likely hear a lot from the public over the coming weeks on various plans and positions on AI regulation. This and future forums will bring together top people in the AI field to discuss the risks and opportunities posed by advances in this technology and how Congress can legislate to address them. Masu. First of 9 6 hour sessions parliamentary forum Wednesday's event will feature guests Sundar Pichai, Mark Zuckerberg, Sam Altman, Elon Musk, Satya Nadella and Jensen Huang to define the US position on AI.
Sign up for the free insideBIGDATA newsletter.
Join us on Twitter: https://twitter.com/InsideBigData1
Join us on LinkedIn: https://www.linkedin.com/company/insidebigdata/
Join us on Facebook: https://www.facebook.com/insideBIGDATANOW
