Today:

OpenAI's "AGI Pieces" SHOCK the Entire Industry! AGI in 7 Months! | GPT, AI Agents, Sora & Search

OpenAI's latest marvel, Sora, is making waves with its text-to-video AI capabilities, impressing folks with its lifelike outputs. Yet, it's not just about cool videos; it's a glimpse into the fast-approaching reality of Artificial General Intelligence (AGI). The tech community, once skeptical about AGI, now takes it seriously, thanks to breakthroughs like ChatGPT. 

 Experts keep revising their AGI arrival estimates closer, hinting it might be around the corner, possibly within years. OpenAI's strategy involves assembling various AI components like Sora, ChatGPT, and autonomous agents into a cohesive AGI system. This blend of technologies aims to simulate, understand, and interact with the world in advanced ways, potentially taking over numerous tasks and professions.

WATCH THE VIDEO ON YOUTUBE

Microsoft develops AI server gear to lessen reliance on Nvidia

Microsoft is developing a new network card to boost its Maia AI server chip, potentially reducing reliance on Nvidia. CEO Satya Nadella has enlisted Pradeep Sindhu, co-founder of Juniper Networks, to lead the effort. The card, akin to Nvidia's ConnectX-7, aims to expedite OpenAI model training on Microsoft servers, cutting costs. Microsoft's investment in OpenAI aids its AI software sales. 

The move follows the introduction of Maia, designed for large language models and AI computing. The project could take over a year, signaling Microsoft's commitment to AI innovation and reducing dependency on external chip providers like Nvidia.

THE INFORMATION

NYT plans to debut new generative AI ad tool later this year

The New York Times is gearing up to launch a new ad-targeting tool powered by generative artificial intelligence (GenAI) later this year. This tech will help advertisers find the best placement for their campaigns based on content and target niche audiences effectively. 

The Times, with its massive audience, sees this as a strategic move to innovate in ad tech despite its legal tussle with OpenAI. Currently, the Times is testing the tool with select partners and aims to roll it out widely in the second half of the year. This move reflects a significant investment by the Times to stay ahead in the advertising game.

AXIOS

GPT-4 and GPT-4 Turbo

GPT-4 and GPT-4 Turbo models update their training data up to December 2023. GPT-4, a powerful multimodal model, boasts enhanced reasoning capabilities and broader knowledge. Meanwhile, GPT-4 Turbo aims to address instances of incomplete tasks with its updated training data. 

Both models offer significant improvements over previous iterations, particularly in complex reasoning scenarios. These updates signify OpenAI's commitment to advancing natural language processing capabilities, catering to diverse user needs.

OPENAI

Adobe Acrobat adds generative AI to ‘easily chat with documents’

Adobe Acrobat is introducing a new AI Assistant feature to simplify handling large documents. Described as a "conversational engine," it helps users find and understand information by summarizing, answering questions, and recommending content. Aimed at reducing time spent on tasks like research or report summarization, it's available in beta for paying Acrobat users. 

The AI Assistant suggests and answers questions, generates citations, and formats information for various purposes. Initially free for certain subscription plans, it will later become part of an add-on subscription. Future plans include integrating with Adobe's Firefly AI model and enhancing document handling capabilities.

THE VERGE

Chinese start-up Moonshot AI raises US$1 billion in funding round led by Alibaba and VC HongShan amid strong interest for OpenAI-type firms

Chinese AI start-up Moonshot AI secured over $1 billion in funding led by Alibaba and HongShan, reflecting sustained investor interest in innovative firms akin to OpenAI. Moonshot, valued at $2.5 billion post-funding, debuted its smart chatbot Kimi Chat, leveraging its Moonshot large language model (LLM). 

Founded by Yang Zhilin, a Tsinghua and Carnegie Mellon alum, Moonshot underscores China's prominence in generative AI, with 22 of 51 global start-ups funded in 2023. Despite criticism from Baidu's Robin Li Yanhong, who deems the proliferation of LLMs wasteful, Moonshot AI thrives amid competition, including a namesake in New York specializing in e-commerce automation.

SCMP

Microsoft to expand its AI infrastructure in Spain with $2.1 billion investment

Microsoft plans to invest $2.1 billion in Spain over two years to boost AI and cloud infrastructure, following a similar move in Germany. Vice Chair Brad Smith highlighted the investment's broader impact beyond data centers, emphasizing a commitment to Spain's development. Details on the Spanish investment are pending. 

The initiative aligns with Microsoft's longstanding engagement in Spain and its support for digital transformation. The investment underscores Microsoft's strategic focus on expanding AI capabilities globally, reflecting a broader trend in technology investment.

REUTERS

🧠RESEARCH

FiT: Flexible Vision Transformer for Diffusion Model

Flexible Vision Transformer (FiT), a new transformer architecture for generating images of any size and aspect ratio. Unlike older methods stuck with fixed image sizes, FiT treats images as sequences of tokens that can change size, making it adapt easily to different shapes during training and use. It's especially good at creating images outside its training range, showing top-notch performance across various resolutions.

Speculative Streaming: Fast LLM Inference without Auxiliary Models

"Speculative Streaming,"by Apple is a method that speeds up language model inference by predicting future text sequences without needing extra models. By focusing on predicting groups of words ahead, it achieves faster processing times (1.8 to 3.1 times quicker) across various tasks like summarizing or answering complex queries. This approach is also more efficient, requiring far fewer parameters than previous methods, making it ideal for devices with limited computing power.

GLoRe: When, Where, and How to Improve LLM Reasoning via Global and Local Refinements

"GLoRe: When, Where, and How to Improve LLM Reasoning via Global and Local Refinements," by Meta focuses on enhancing reasoning in language models, especially for tasks in math, science, or coding. It introduces Stepwise Outcome-Based Reward Models (SORMs) that use synthetic data to better detect incorrect reasoning steps, leading to more accurate refinements. By integrating global and local refinement models, the approach notably improves the accuracy of LLaMA-2 13B models on specific reasoning tasks, showcasing a strategic blend of corrections to boost performance.

FinTral: A Family of GPT-4 Level Multimodal Financial Large Language Models

FinTral is a cutting-edge multimodal large language model tailored for financial analysis. Built on the Mistral-7b model, it integrates various data types like text, numbers, tables, and images. FinTral is boosted with domain-specific training and excels in zero-shot performance, outdoing ChatGPT-3.5 in all tasks and surpassing GPT-4 in more than half of them. It's designed for real-time financial analysis and decision-making, marking a significant step forward in AI-driven financial tech. 

AnyGPT: Unified Multimodal LLM with Discrete Sequence Modeling

AnyGPT is a novel language model capable of handling multiple modalities (like speech, text, images, and music) through discrete representations. This approach allows it to process and generate content across different media without changing its architecture or training methods. By synthesizing a large multimodal dataset, AnyGPT achieves seamless any-to-any multimodal conversation capabilities, showing that it can match specialized models in performance across all tested modalities. This makes it a versatile tool for integrating diverse data types into language models​.

Learning to Learn Faster from Human Feedback with Language Model Predictive Control

Language Model Predictive Control (LMPC) is a new framework that enhances the learning speed of robots from human feedback. By treating interactions as a sequence of decisions, LMPC fine-tunes language models to better remember and learn from past interactions, leading to more efficient teaching. Applied to robot code-writing tasks, it improved teaching success rates by 26.9% and reduced the average number of human corrections needed. This method also significantly boosts the model's ability to learn new tasks in unfamiliar settings​.

🛠️TOP TOOLS

GrammarBot - AI app designed to correct grammar and spelling errors without needing an internet connection or subscription. 

Galileo AI - UI generation platform that speeds up design ideation, enabling the creation of interface designs quickly and efficiently.

Pecan - create personalized machine learning models with ease, targeting users proficient in SQL.

Distribute - provides a streamlined platform for managing product launches, offering tools for creating landing pages, organizing beta tests, and collecting feedback. 

Chatonai -  Access the ChatGPT language model directly from your search engine results to ask any question and get a natural language response.

📲SOCIAL MEDIA

🗞️MORE NEWS

Groq’s $20,000 LPU chip breaks AI performance records to rival GPU-led industry

Groq's $20,000 LPU chip sets new AI processing records, challenging GPU dominance. The LPU Inference Engine excels in language tasks, outperforming competitors with faster text sequence generation. GroqCard Accelerator offers superior performance for simpler data, while NVIDIA's A100 GPU leads in complex processing. The LPU's specialized design targets AI language applications, promising a new era in hardware innovation. CRYPTOSLATE

Ex Zillow exec launches AI-powered home search platform

Former Zillow executives launch Tomo, an AI-powered home search platform, promising faster, personalized results. Tomo stands out with its AI-driven free-text search feature, allowing users to describe their dream homes naturally. The platform provides insights into property listings and offers negotiation tips. Available initially in select U.S. cities. AXIOS

Self-Driving Startup Recogni Nabs $102 Million to Design AI Chip

Recogni Inc., a San Jose-based AI startup initially focused on autonomous vehicles, secures $102 million in Series C funding. Led by Celesta Capital and GreatPoint Ventures, the round also sees participation from HSBC Holdings Plc and Tasaru Mobility Investments. Recogni expands into designing chips for generative AI. Total funding reaches $175 million. BLOOMBERG

India’s AI market seen touching $17 bln by 2027, notes Nasscom-BCG report

India's AI market is forecasted to reach $17 billion by 2027, with a growth rate of 25-35% annually from 2024 to 2027, per a Nasscom-BCG report. Factors driving this growth include increased enterprise tech spending, a growing AI talent pool, and rising AI investments. India currently boasts over 420,000 employees in AI roles, with significant investments in training, startups, and technology expansion by IT companies. This surge in AI investment is expected to drive a 15% annualized growth in demand for AI talent in India until 2027. REUTERS

Scale AI to set the Pentagon’s path for testing and evaluating large language models 

Scale AI has been selected by the Pentagon's Chief Digital and Artificial Intelligence Office (CDAO) to develop a comprehensive framework for testing and evaluating large language models (LLMs) within the Department of Defense. This one-year contract aims to ensure the safe and reliable deployment of AI models, providing real-time feedback for warfighters and creating specialized datasets for military support applications. The initiative addresses the challenges associated with testing and evaluating generative AI, such as the lack of "ground truth" due to the complex nature of language models. Scale AI will work iteratively to develop holdout datasets and evaluation metrics, ultimately enhancing the robustness and resilience of AI systems in classified environments. DEFENSE SCOOP

This Boston Billionaire Is Building An Army Of Robot Construction Workers

John Fish, a Boston billionaire, pioneers automation in construction with robotic workers. His firm, Suffolk, deploys small robots for tasks like blueprint printing, aiming to boost efficiency. Fish also invests in construction tech startups to revolutionize the industry, addressing labor shortages and rising costs. His vision: embrace robotics for progress. FORBES

US House forms AI task force as legislative push stalls

The U.S. House forms a bipartisan task force to address AI concerns amid legislative stagnation. Led by Speaker Mike Johnson and Democratic Leader Hakeem Jeffries, the task force aims to produce a comprehensive report on AI regulations and potential threats, emphasizing the need for safeguards and guiding principles. REUTERS

AI can tell a person's sex from brain scans with 90 percent accuracy

A recent study utilizing dynamic functional connectivity fMRI and artificial intelligence reveals that men's and women's brains exhibit notable differences, despite past debate. With around 90% accuracy, AI can discern between male and female brain scans, shedding light on brain dissimilarities previously obscured by variations in brain size. NEWSCIENTIST