Today:

• Tencent ‘More Agents is All You Need’

• OpenAI Launches GPT-4 Turbo API

• Intel's Gaudi 3 Launches

• Microsoft Invests $2.9B in Japan

• South Korea's $7B AI Investment Plan

• Meta's to Release Smaller Llama AI

• Google Cloud Next 2024

"More Agents is All You Need" Paper | Is Collective Intelligence the way to AGI?

"More Agents Is All You Need" explores how the performance of large language models (LLMs) can be enhanced through a simple sampling-and-voting method. Authored by Junyou Li, Qin Zhang, Yangbin Yu, Qiang Fu, and Deheng Ye, the research demonstrates that the effectiveness of LLMs increases with the number of instantiated agents. 

Examples include using multiple agents to play Minecraft. However, concerns arise regarding potential misuse, such as in click farms or civil attacks. The future implications of AI advancement, including societal, economic, and ethical considerations, remain uncertain. 

WATCH THE VIDEO ON YOUTUBE

OpenAI makes GPT-4 Turbo with Vision generally available through its API

OpenAI has rolled out its GPT-4 Turbo with Vision model, now accessible through its API, facilitating integration into third-party apps. This enhancement combines text and image analysis in a single API call, boosting efficiency for developers. The model, boasting enlarged input context windows and improved speed, caters to diverse applications like autonomous coding, nutritional analysis in health apps, and website generation from drawings. 

Despite facing stiff competition from newer models, OpenAI's move aims to widen its appeal among enterprise clients and developers. The update underscores the company's commitment to advancing AI accessibility while awaiting its next-gen language model.

VENTUREBEAT

Intel’s Gaudi 3 launches to challenge Nvidia in the AI chip space with an open ecosystem

Intel's Gaudi 3 AI Accelerator, Intel's latest AI chip, aims to challenge Nvidia by offering a streamlined development process and improved performance for enterprise AI workloads. With increased computing power, bandwidth, and memory, it targets large language models. 

Gaudi 3 boasts higher efficiency and performance compared to Nvidia's options, emphasizing its suitability for AI tasks. Intel positions Gaudi 3 as part of a broader strategy to revolutionize enterprise AI, emphasizing an open ecosystem and collaboration. 

Additionally, Intel and Altera have unveiled new AI-optimized processors and FPGA chips for edge computing. These products aim to enhance various industries with advanced AI capabilities, enabling faster decision-making processes on-premises. The collaboration introduces a range of edge-optimized processors and FPGAs, catering to diverse industry needs. Intel's lineup includes Core Ultra Processors for Edge, Core Processors for Edge, and Atom Processors tailored for networking, telecommunications, and industrial applications. Altera's Agilex 5 SoC FPGAs offer high performance with lower power consumption, complemented by updates to its FPGA portfolio. 

VENTUREBEAT

Microsoft to invest $2.9 bln to expand AI, cloud infra in Japan

Microsoft announced a $2.9 billion investment over two years to expand its cloud and AI infrastructure in Japan. This marks their largest investment in the country in 46 years. The move aims to support the development of artificial intelligence and will also include skilling three million people in AI. 

Additionally, Microsoft plans to establish a Microsoft Research Asia lab in Tokyo. This investment reflects a broader trend of tech giants expanding globally to meet the growing demand for AI applications. Other players like Amazon and Google are also investing heavily in data centers worldwide.

REUTERS

South Korea to invest $7 billion in AI in bid to retain edge in chips

South Korean President Yoon Suk Yeol announced a $6.94 billion investment in artificial intelligence by 2027 to bolster the country's position in cutting-edge semiconductor chips. The plan includes a fund to support AI semiconductor firms, aiming to compete with the US, China, and Japan in semiconductor supply chains. 

Semiconductors are crucial for South Korea's export-driven economy, with recent chip exports hitting a 21-month high. Yoon aims for South Korea to be a top-three AI technology player and capture over 10% of the global system semiconductor market by 2030, focusing on AI chips and next-gen memory chips.

REUTERS

Meta may release smaller Llama AI model before the big version

Meta is set to release smaller versions of its Llama language model ahead of the launch of its flagship model. These smaller models, expected to be launched this month, aim to offer more cost-effective AI solutions. The move reflects a broader trend in the AI industry towards lightweight models, which are faster, more flexible, and cheaper to run than their larger counterparts. 

Such models are particularly suitable for specific projects or devices with limited computing power, attracting users who don't require the full capabilities of larger models. Meta plans to release its flagship Llama 3 model in July, with potential enhancements allowing it to answer controversial questions.

THE VERGE

Google Cloud Next 2024

Google Cloud Next 2024 highlights the company's commitment to democratizing AI tools for businesses worldwide. With a $36 billion annual revenue run rate in Q4, Google Cloud's momentum is fueled by AI investments. New releases include:

Google's Gemini 1.5 Pro update adds listening ability, enabling analysis of audio files without transcripts. It outperforms previous models and enhances AI applications.

Google is ramping up its semiconductor efforts with a new chip, Axion, designed to tackle diverse tasks from YouTube ads to big data analysis. Axion, commonly used in data centers, expands Google's decade-long pursuit of innovative computing resources, initially focusing on AI-specialized chips.

Google introduces Vids, a collaborative video tool for work. Simplifying video creation, it aims to replace traditional slides with easily shareable, editable videos. Launching soon.

Google's Imagen 2 updates bring text-to-live animated images and image editing features. Targeting media and advertising, it enhances user engagement and addresses security concerns.

Google and Nvidia collaborate to boost AI development for startups. Merging their startup accelerators, they offer cloud credits, technical resources, and market support for AI applications.

🧠RESEARCH

Ferret-UI: Grounded Mobile UI Understanding with Multimodal LLMs

Ferret-UI tackles the challenge of improving multimodal large language models' (MLLMs) comprehension of mobile UI screens. By incorporating enhanced visual features and meticulous training samples, Ferret-UI excels in tasks like icon recognition and text finding, outperforming both open-source UI MLLMs and GPT-4V. It boasts exceptional understanding and execution of UI instructions.

ByteEdit: Boost, Comply and Accelerate Generative Image Editing

ByteEdit revolutionizes generative image editing by addressing quality, consistency, instruction adherence, and efficiency challenges. Its feedback learning framework integrates reward models for aesthetics and coherence, along with an innovative adversarial and progressive strategy to boost inference speed. User evaluations showcase ByteEdit's superiority over leading editing tools like Adobe and Canva, achieving significant enhancements in quality and consistency.

BeyondScene: Higher-Resolution Human-Centric Scene Generation With Pretrained Diffusion

BeyondScene tackles the challenge of generating higher-resolution human-centric scenes with exceptional detail and naturalness. Overcoming limitations of existing text-to-image models, it utilizes a staged, hierarchical approach to produce exquisite scenes exceeding 8K resolution. By incorporating novel techniques like instance-aware hierarchical enlargement, BeyondScene achieves superior correspondence with text descriptions, enabling advanced scene creation beyond pretrained model capacity.

UniFL: Improve Stable Diffusion via Unified Feedback Learning

UniFL introduces a unified feedback learning framework to enhance diffusion models, addressing issues like visual quality, aesthetic appeal, and inference efficiency. It integrates perceptual, decoupled, and adversarial feedback learning components, achieving superior performance across various diffusion models. Experiments demonstrate UniFL's effectiveness in improving model quality and acceleration, surpassing existing methods like ImageReward and SDXL Turbo.

MagicTime: Time-lapse Video Generation Models as Metamorphic Simulators

MagicTime presents a novel approach to time-lapse video generation, addressing the lack of real-world physics knowledge in existing Text-to-Video models. By leveraging metamorphic generation and a unique training scheme, MagicTime learns from time-lapse videos to produce dynamic and varied outputs. The proposed ChronoMagic dataset further enhances model understanding and performance, highlighting the potential of time-lapse videos as metamorphic simulators.

🛠️TOP TOOLS

Coframe - Optimise your images with generative A/B testing

Modular - The programming language written in plain English

Section - Personal Productivity workshop completely free when you sign up

Texta - AI blog writer and article ideas generator

GPT LLM Trainer - Go from prompt to fine-tuned model automatically.

📲SOCIAL MEDIA

🗞️MORE NEWS

AlphaSense, a Goldman Sachs–backed AI research startup valued at $2.5B, gears up for IPO as it crosses $200M in annual recurring revenue

AlphaSense, a research startup backed by Goldman Sachs and valued at $2.5B, is set for an IPO after hitting $200M in yearly revenue. Its AI-driven search engine, favored by hedge funds and big corporates, is bolstered by generative AI, attracting investor attention. Expansion plans include new markets and leadership diversification. FORTUNE

Symbolica raises $31 mln to develop AI systems to compete with OpenAI

Symbolica, an AI startup challenging OpenAI, secures $31M funding led by Khosla Ventures. CEO George Morgan, a former Tesla engineer, aims to pioneer new AI architectures beyond transformers. First product, a coding assistant, slated for 2025 launch. Industry debates scaling limitations versus alternative models' potential. REUTERS

When an antibiotic fails: MIT scientists are using AI to target “sleeper” bacteria

MIT scientists leverage AI to target dormant bacteria resistant to traditional antibiotics. Led by former MIT-Takeda Fellow Jackie Valeri, researchers identify compound semapimod, originally an anti-inflammatory, effective against stationary-phase E. coli and A. baumannii. Semapimod disrupts Gram-negative bacteria's membranes, potentially broadening antibiotic effectiveness. MIT

Alibaba-backed Moonshot AI narrows gap with Baidu's Ernie Bot as China's generative AI rivalry heats up

Moonshot AI, a Chinese start-up, is gaining ground in the AI chatbot competition, challenging leaders like Baidu. Moonshot's Kimi surpassed Alibaba's Tongyi Qianwen, approaching Baidu's Ernie Bot. Kimi's popularity surged, reaching 12.6 million views, attributed to its specialized functions. However, global leaders like OpenAI's ChatGPT still dominate. YAHOO!

Klarna Boosts Profits With ChatGPT as BNPL Firms Tap AI

Klarna, a buy now, pay later (BNPL) firm, leverages generative AI like ChatGPT to enhance profits by $40 million, speeding up customer service and reducing labor needs. Other BNPL companies, like Afterpay and PayPal, are also integrating AI for personalized transactions, aiming to boost revenue and improve profitability in a competitive market. PYMNTS

Beijing launches AI public platform as demand mounts for computing power

Beijing has launched a massive artificial intelligence (AI) public computing platform in Yizhuang, Daxing district, offering 3,000 petaflops of computing power to support AI research and development. With the rise of AI, the demand for computational power is escalating, prompting China to bolster its computing infrastructure. This move aims to foster domestic alternatives to global AI leaders while leveraging China's vast manufacturing data for AI development. GLOBAL TIMES

The Aboard app is a totally different take on what an AI bot can do

Aboard, a novel AI app, combines elements of Pinterest, Trello, and ChatGPT to create versatile boards for tracking various tasks. Developed by web experts, it utilizes AI to organize data, research topics, and streamline workflows. Despite occasional glitches, it aims to revolutionize how users interact with AI, offering customizable solutions for personal and professional use. THE VERGE