GPT-5 Launches Soon

PLUS: Vercel Unveils SDK 3.1, SK Hynix Leads AI Chip Market and more.

Today:

  • GPT-5 Launches Soon

  • OpenAI Launching Microsoft-Powered Search"

  • Microsoft Bars Police Facial Recognition

  • Vercel Unveils SDK 3.1

  • ChatGPT Banned by National Archives

  • Microsoft new data center in Thailand

  • SK Hynix Leads AI Chip Market 

GPT-5 Launch Day is NEAR | The End of Privacy and "Digital People"

The upcoming release of GPT-5 by OpenAI has sparked widespread discussion and anticipation. Although its exact name and form remain speculative, significant developments in AI models are expected soon, including OpenAI's GPT-5 and Meta's advanced Llama 3 models. These models aim to enhance AI capabilities in reasoning and memory. 

Additionally, challenges in pre-deployment testing and model access, particularly with the AI Safety Institute and UK government, underscore ongoing regulatory and logistical hurdles. OpenAI continues to innovate in AI applications, suggesting a transformative impact on productivity and business management through increasingly sophisticated AI agents. 

OpenAI has been developing a web search product that would bring the Microsoft-backed startup into more direct competition with Google, according to someone with knowledge of OpenAI’s plans.

OpenAI, backed by Microsoft, is reportedly close to launching a new search engine to compete with Google's dominant position in the market. According to industry insider Jimmy Apples, the announcement is expected on May 9, 2024. The initiative follows OpenAI's significant recruitment of an events team, hinting at a major launch. 

The search engine would use Microsoft Bing's infrastructure, aiming not just to replicate Google Search but to innovate in how information is discovered and utilized. This development is part of a broader tech rivalry, with potential to reshape online search dynamics.

Microsoft bans U.S. police departments from using enterprise AI tool for facial recognition

Microsoft Store sign.

Microsoft has tightened its rules, now strictly forbidding U.S. police departments from using its Azure OpenAI Service for facial recognition purposes. This follows concerns about potential misuse and errors, like false facts and racial biases, which could be exacerbated by AI inaccuracies. These updated terms now also prevent global law enforcement from using real-time facial recognition on mobile devices in unpredictable settings. 

While these changes mainly affect U.S. police, the usage for facial recognition with stationary cameras in controlled environments like offices remains allowed. Microsoft's decision aligns with its recent strategic approaches to AI in law enforcement and defense sectors.

Vercel acquires ModelFusion, launches AI SDK 3.1 for enterprise AI development

Vercel, a key player in web development tools, has acquired ModelFusion.dev and released AI SDK 3.1, enhancing its capabilities in the enterprise AI development sector. This new toolkit supports the integration of AI into applications using TypeScript and simplifies working with various language models by providing a unified coding interface. AI SDK 3.1 includes AI SDK Core, which facilitates interaction with different AI models, offering functionalities like text and object generation. 

The acquisition of ModelFusion.dev extends Vercel's framework, adding functions that aid in automating and developing AI agents. This strategic expansion positions Vercel to significantly impact the growing market for enterprise AI solutions.

National Archives Bans Employee Use of ChatGPT

ChatGPT

The National Archives and Records Administration (NARA) has decided to block the use of ChatGPT on all its agency-issued devices, citing security concerns. Starting May 6, 2024, access to ChatGPT will be prohibited on devices connected to NARANet, including laptops, tablets, desktop computers, and mobile phones. 

The ban arises from worries about the potential for internal government data to be unintentionally shared or leaked via ChatGPT’s services, as it can incorporate and reuse data input by users. This measure reflects NARA's responsibility to maintain the security and integrity of its historical records.

Microsoft to open new data center in Thailand as it doubles down on AI and Southeast Asia

Microsoft has announced significant expansion efforts in Southeast Asia, revealing plans to establish a new regional data center in Thailand. This initiative, part of a broader commitment to the region, also includes investing $1.7 billion over the next four years to enhance cloud and AI infrastructure in Indonesia. 

These developments underscore Microsoft's drive to support the increasing demand for cloud computing in these markets and to promote AI-powered digital transformation. In Thailand, the company aims to train over 100,000 people in AI skills, supporting both the public and private sectors. These strategic moves are poised to leverage and accelerate economic and productivity opportunities related to AI across Southeast Asia.

SK Hynix Has Sold Most of Its AI Chip Capacity Through 2025

SK Hynix Inc., a major player in the semiconductor industry, has nearly filled its production capacity for high-bandwidth memory (HBM) chips through 2025, highlighting the strong demand for components crucial for artificial intelligence (AI) technologies. 

The South Korean firm plans to start mass production of its advanced HBM3E chips in the next quarter, aiming to maintain its competitive edge over rivals like Samsung Electronics. These high-performance chips are essential for AI platforms, especially those that use Nvidia Corp. accelerators, reinforcing SK Hynix's pivotal role in the ongoing global AI boom.

🧠RESEARCH

The paper discusses advancements in image editing using text-conditioned diffusion models. It addresses the challenge of seamlessly adding objects to images based on textual instructions. By leveraging the simplicity of removing objects compared to adding them, the study develops a method to effectively add objects into images using curated datasets and trained models.

The paper introduces SemantiCodec, an innovative audio codec designed for diverse sound types, including speech and music, compressing audio into fewer than a hundred tokens per second. Its dual-encoder architecture utilizes semantic and acoustic encoders, achieving ultra-low bitrates without compromising quality. Experimental results demonstrate superior performance over existing codecs, offering richer semantic information even at significantly lower bitrates.

The paper introduces Self-Play Preference Optimization (SPPO) for language model alignment, addressing limitations of traditional reinforcement learning from human feedback approaches. SPPO treats the problem as a two-player game to identify Nash equilibrium policy, achieving theoretical convergence. Experimental results demonstrate SPPO's effectiveness, outperforming other methods without additional external supervision.

The paper presents a novel approach for matching app images with search terms, crucial for effective app advertising. Utilizing a fine-tuned LXMERT model, it significantly improves matching accuracy over baseline methods like CLIP and transformer+ResNet. Evaluation results demonstrate notable performance gains, achieving high AUC scores for both advertiser-associated and human-labeled ground truths.

The paper introduces GS-LRM, a scalable large reconstruction model for 3D Gaussian splatting, enabling high-quality 3D primitive prediction from sparse images. Operating efficiently on a single A100 GPU, it utilizes a simple transformer-based architecture to decode per-pixel Gaussian parameters directly, accommodating scenes with varied scale and complexity. Outperforming existing methods, it demonstrates efficacy across object and scene captures, with potential applications in downstream 3D generation tasks.

🛠️TOP TOOLS

Gamma - Create a working presentation, document or webpage you can refine and customize in under a minute, using our powerful AI generator.

Getty Image - An AI image generator allows users to create custom visual content using text prompts

Squirrly SEO -  all-in-one SEO suite for WordPress powered by AI, designed to simplify and optimize the SEO process

Applicant AI - Review job applicants 10x faster with AI

Interview with AI - lets you practice with AI like an actual online interview with an HR professional

🗞️MORE NEWS

Providing further transparency on our responsible AI efforts

Microsoft's inaugural Responsible AI Transparency Report outlines the company's commitment to ethical AI practices. Since 2016, guided by six core values, they've developed products aligning with human-centered principles. The report details their generative AI development process, case studies, customer support, and community engagement, inviting global participation in responsible AI dialogue. MICROSOFT

2024 AI Readiness Report

The 2024 AI Readiness Report by Scale explores trends and best practices for optimizing and securing generative AI. It addresses challenges faced by companies in AI adoption and offers insights for building, applying, and evaluating AI models. The report aims to help businesses harness AI's potential for operational efficiency, customer experience enhancement, product development, and revenue growth. SCALE

Snapchat will soon let you edit your messages after sending

Snapchat introduces new features including post-sent message editing within a 5-minute window, emoji reactions, and AI-powered reminders. The update also includes AI-generated Bitmoji fashion and '90s-style selfie filters. These features enhance user experience and engagement, aligning with trends seen in other messaging platforms. THE VERGE

Stardog’s Karaoke offers on-premises, zero hallucination LLM solution for enterprises

Stardog introduces "Karaoke," an on-premises server hosting their Voicebox Large Language Model (LLM) platform, developed in collaboration with Nvidia and Supermicro. Karaoke allows users to query natural language into Stardog Cloud and receive responses from their structured data. It ensures data accuracy and compliance, catering to regulated industries' needs. VENTUREBEAT

Google Confirms It Will Join Global Experts at AI Safety Summit

Google's participation in the upcoming AI Safety Summit in South Korea signals a commitment to responsible AI development amid growing recognition of the need for such initiatives. The summit, set to explore AI's limitations and commercial impact, could have significant implications for global commerce and regulations. Despite ongoing efforts to enhance AI safety, observers remain cautious about the summit's outcomes due to continued governmental implementations and ongoing debates surrounding AI ethics and liability. PYMNTS

AI discovers over 27,000 overlooked asteroids in old telescope images

An AI-powered algorithm has recently discovered over 27,000 previously overlooked asteroids in our solar system, offering valuable insights into potential threats to Earth and enhancing our ability to track these space rocks. Most of these newly found asteroids reside in the asteroid belt between Mars and Jupiter, with some of them even crossing Earth's orbit. Utilizing AI technology, astronomers can analyze vast amounts of telescope images to identify moving objects, indicative of asteroids. SPACE.COM

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.