OpenAI Leads $500B AI Push

PLUS: Perplexity Debuts Sonar: AI Search, Hassabis Reveals AI Drug Milestone and more.

January 23, 2025

In partnership with

Your daily AI dose

Mindstream is your one-stop shop for all things AI.

How good are we? Well, we become only the second ever newsletter (after the Hustle) to be acquired by HubSpot. Our small team of writers works hard to put out the most enjoyable and informative newsletter on AI around.

It’s completely free, and you’ll get a bunch of free AI resources when you subscribe.

Today:

  • OpenAI Leads $500B AI Push

  • ChatGPT Operator Launching This Week

  • Google Backs Anthropic with Another $1B

  • ByteDance Plans $12B AI Spend

  • Microsoft Loses OpenAI Exclusivity

  • Perplexity Debuts Sonar: AI Search

  • Hassabis Reveals AI Drug Milestone

AI Triggers UNPRECENDENTED Focus on "Tech Tree" Progression | $500,000,000,000 Project Stargate

OpenAI, SoftBank, Oracle, and UAE's MGX are investing $500 billion in "Project Stargate," a U.S.-based AI infrastructure initiative. Announced with presidential backing, the project aims to build colossal data centers and attract global funding to outpace China in AI advancements. 

While the initiative promises breakthroughs in healthcare and technology, critics question funding sources and geopolitical implications. With competition intensifying, AI development accelerates, supported by favorable U.S. policies and industry collaboration.

OpenAI is launching "Operator," a ChatGPT browser agent, capable of handling tasks like travel planning and online shopping. Users can view real-time browser actions, take control, and share tasks. Despite the innovation, safety concerns over "prompt injection" attacks remain. OpenAI sees AI agents as the next frontier, aligning with rivals like Anthropic, Microsoft, and Google, who are racing to automate workflows using intelligent assistants.

Google has invested an additional $1 billion in Anthropic, a San Francisco-based AI developer and notable rival to OpenAI. This follows over $2 billion in previous investments. Google’s partnership with Anthropic includes using its online tools and services. Amazon is also among Anthropic's key supporters, highlighting its growing influence in the competitive AI industry.

ByteDance plans to invest over $12 billion in AI infrastructure in 2025, doubling last year’s spending. This includes $5.5 billion for Chinese AI chips and $6.8 billion overseas, focusing on Nvidia GPUs for training foundation models. The move aims to strengthen ByteDance's AI capabilities amidst U.S. export restrictions and competition from Chinese tech giants like Baidu and Alibaba, while its social media business faces pressure in the U.S.

Microsoft is no longer OpenAI's exclusive cloud provider as OpenAI signs a new infrastructure deal with SoftBank, Oracle, and others. Microsoft retains a "right of first refusal" for hosting OpenAI workloads but allows flexibility to use rival providers. This shift comes amid strained relations over compute capacity. OpenAI continues to support Azure while exploring expanded capacity for AI model training and products, highlighting evolving dynamics in their partnership.

Perplexity has launched Sonar, an AI search API offering real-time web-connected capabilities, challenging Google, OpenAI, and Anthropic. With a dual-tier structure, Sonar delivers affordable, citation-backed responses, outperforming rivals in benchmarks like SimpleQA. Sonar Pro, tailored for complex queries, enhances citation density and accuracy. Already integrated with Zoom’s AI Companion, Sonar targets enterprises seeking real-time, verifiable data solutions. Backed by $500M funding, Perplexity aims to disrupt the AI search market with competitive pricing and innovative features.

Google-owned Isomorphic Labs, led by Sir Demis Hassabis, plans to have its first AI-designed drug in trials by year-end, targeting oncology, cardiovascular, and neurodegenerative diseases. The startup, spun out of DeepMind, collaborates with Eli Lilly and Novartis to revolutionize drug discovery by significantly reducing development timelines. Hassabis envisions AI agents negotiating on behalf of users within three years and advocates for cautious development of artificial general intelligence (AGI) to prevent misuse.

🧠RESEARCH

Agent-R introduces an iterative self-training framework enabling language model agents to reflect and recover from errors in real-time. By using model-guided critique and data splicing, Agent-R dynamically constructs training datasets for efficient learning. Experiments across interactive environments show a 5.59% performance boost, demonstrating improved error correction and reduced looping.

IntellAgent is a scalable, open-source framework for evaluating conversational AI systems, tackling the complexity of multi-turn dialogues, API integration, and policy constraints. It uses graph-based modeling and realistic simulations to create synthetic benchmarks, enabling fine-grained diagnostics and actionable insights. IntellAgent fosters optimization, reproducibility, and collaboration in advancing conversational AI.

MMVU is a benchmark for evaluating video understanding models across 27 subjects in Science, Healthcare, Humanities, and Engineering. Featuring 3,000 expert-annotated questions, it challenges models with domain-specific reasoning and expert-level tasks. Despite advancements, models like o1 and Gemini 2.0 still lag behind human expertise, offering room for future improvements.

This paper revisits Load-balancing Loss (LBL) in training Mixture-of-Experts (MoE) models. By shifting from micro-batch to global-batch LBL calculation, it encourages better load balance across diverse sequences, enabling improved domain specialization. Experiments with large-scale models show enhanced performance in pre-training perplexity and downstream tasks, highlighting this strategy's effectiveness.

TokenVerse introduces a multi-concept personalization framework for text-to-image diffusion models. It extracts and combines diverse visual concepts—objects, materials, poses, and lighting—from single or multiple images. Using modulation space for semantic control, TokenVerse enables precise and creative image generation. Its optimization-based approach outperforms existing methods, enhancing personalization versatility.

🛠️TOP TOOLS

ComfyUI - Open-source, node-based graphical user interface designed for working with Stable Diffusion models and other AI image generation tools.

Seneca Answers - AI-powered web application that allows users to interact with a virtual version of the ancient Stoic philosopher Seneca.

Spikes Studio - AI-powered video editing platform that revolutionizes content creation for social media.

Text To Song Converter - AI tool that transforms written text into fully-fledged songs. 

Chatmind -  AI-powered mind mapping tool developed by the Xmind team that transforms various content formats into clear, structured mind maps.

📲SOCIAL MEDIA

🗞️MORE NEWS

  • Google's Gemini 2.0 Flash Thinking AI has outperformed rivals like GPT-4o, excelling in math, science, and programming tests. With enhanced code execution and a larger context window, it sets a new standard in chatbot performance.

  • Tencent's Hunyuan3D 2.0 revolutionizes 3D modeling, transforming images or text into detailed models within seconds. Open-sourced on Hugging Face, it combines speed, accuracy, and versatility, enabling industries to simplify creating gaming, shopping, and entertainment assets.

  • OpenAI is training its next-generation model, o4, which promises significant capability improvements. Revealed at Davos, Chief Product Officer Kevin Weil highlighted faster iteration cycles, building on rapid progress from previous versions like o3, due soon.

  • Databricks secured $15.3 billion in funding, valuing the company at $62 billion. With Meta as a strategic investor, it plans to expand AI offerings, global operations, and acquisitions while delaying its anticipated IPO.

  • LinkedIn is facing a class-action lawsuit alleging it shared Premium users' private messages for AI training without consent. Plaintiffs claim LinkedIn violated privacy policies and laws, seeking damages and accountability for disclosed data use.

  • Santee Cooper plans to revive two halted nuclear reactors in South Carolina to meet AI-driven clean energy demand. With tech giants like Amazon and Microsoft seeking sustainable energy, interest in the project is expected to grow.

  • Clay, a data-driven AI startup, has doubled its valuation to $1.3 billion within six months. Focused on optimizing sales leads, it powers major firms like OpenAI and Anthropic, driving rapid growth and industry-wide innovation.

  • Elon Musk and Sam Altman clashed on social media over Stargate, a $500 billion AI infrastructure project backed by SoftBank, OpenAI, and Oracle. Musk questioned funding, while Altman defended the initiative's progress and national benefits.

What'd you think of today's edition?

Login or Subscribe to participate in polls.

Reply

or to participate.