OpenAI Introduces Deep Research

Also, DeepSeek’s true expenditure number is looking way higher than the reported $6 million that led to everyone freaking out

Author

Jack Lajoie
February 03, 2025 • read time ~ 9 minutes

⚡️ Headlines

🤖 AI

OpenAI Launches O3 Mini Model and ChatGPT API – OpenAI has released the O3 Mini model alongside a new ChatGPT API, aiming to make AI integration more accessible for developers. [The Verge].

Bill Gates Reflects on Coding and Generative AI – In a recent discussion, Bill Gates emphasizes the enduring importance of coding skills amidst the rise of generative AI technologies. [Axios].

ChatGPT Subscribers Nearly Tripled to 15.5 Million in 2024 – OpenAI's ChatGPT saw its subscriber base grow from 5.6 million to 15.5 million in 2024, reflecting a significant increase in user adoption. [The Information].

OpenAI Leadership Hosts Reddit AMA – Sam Altman, Mark Chen, and other OpenAI leaders participated in a Reddit AMA, discussing the O3 Mini model and the future of AI. [Reddit].

AI Tool Enables Conversations with Future Self – A new AI application allows users to simulate conversations with their future selves, providing insights into personal development and decision-making. [The Wall Street Journal].

Debate Over Secrecy in U.S. AI Development – Experts discuss whether America's AI initiatives should adopt secrecy akin to the Manhattan Project to maintain a competitive edge. [The Wall Street Journal].

DeepSeek's Success Amid China's AI Startup Challenges – DeepSeek's recent achievements highlight its unique position against the backdrop of struggles faced by other Chinese AI startups. [The Information].

🦾 Emerging Tech

Meta's Ray-Ban Smart Glasses Surpass 1 Million Sales in 2024 – Meta's Ray-Ban smart glasses achieved over 1 million units sold in 2024, indicating growing consumer interest in wearable tech. [The Information].

Parents Opt for Bitcoin Over Traditional College Savings Plans – A trend emerges as parents choose Bitcoin investments instead of traditional 529 plans for their children's college savings. [Bloomberg].

🤳 Social Media

Instagram Previews AI-Powered Editing Features – Instagram is testing new AI-driven editing tools to enhance user experience and content creation capabilities. [Social Media Today].

🔬 Research

Americans Unknowingly Use AI in Daily Products – A Gallup poll reveals that many Americans are unaware of the AI technologies embedded in everyday products they use. [Gallup].

Google X Spins Out Heritable Agriculture to Boost Crop Yields with AI – Google X launches Heritable Agriculture, a startup leveraging AI to enhance crop yields and agricultural efficiency. [TechCrunch].

MIT Chemists Use Generative AI to Model 3D Genomic Structures – MIT researchers employ generative AI to rapidly predict three-dimensional genomic structures, advancing genetic research. [MIT News].

⚖ Legal

Governor Abbott Announces Ban on Chinese AI, Social Media Apps – Texas Governor Greg Abbott bans Chinese AI and social media applications from state-issued devices, citing security concerns. [Office of the Texas Governor].

🔌 Plug-Into-This

OpenAI has unveiled "deep research," an AI agent designed to autonomously conduct complex, multi-step research tasks on the internet. Leveraging the upcoming o3 model, this tool analyzes and synthesizes information from various online sources, including text, images, and PDFs, to generate comprehensive reports within 5 to 30 minutes. Initially available to ChatGPT Pro users in the U.S., deep research aims to assist professionals in fields such as finance, science, and engineering.

  • Integration and Accessibility: Deep research is integrated into the ChatGPT interface, allowing users to input queries via text, images, and files. The AI processes these inputs to produce detailed reports, complete with citations and a summary of its research process.

  • Underlying Technology: The agent is powered by OpenAI's forthcoming o3 model, optimized for web browsing and data analysis. This model enhances the AI's ability to autonomously plan and execute research tasks, reflecting a significant advancement in AI reasoning capabilities.

  • Performance and Limitations: While deep research aims to match the output of a human research analyst, experts caution against fully trusting AI outputs without human verification. Potential limitations include distinguishing authoritative information from rumors and accurately conveying uncertainty.

  • Availability and Pricing: Initially, deep research is available to ChatGPT Pro users in the U.S., with a limit of up to 100 queries per month. The Pro subscription is priced at $200 per month. Plans for broader access or API integration have not been detailed yet.

⏳ This launch was likely moved up a bit from the DeepSeek craze last week, as OpenAI tries to recapture the public focus. Deep research ostensibly represents a significant step towards developing artificial general intelligence, though some are skeptical about how much of an advancement it really is.

Recent analyses suggest that Chinese AI startup DeepSeek's claims of developing advanced AI models with minimal resources may be overstated. Reports indicate that the company has invested approximately $1.6 billion in hardware and operates around 50,000 Nvidia Hopper GPUs, challenging the narrative of achieving high efficiency with limited means.

  • Infrastructure Details: DeepSeek's computing infrastructure includes 10,000 H800s and 10,000 H100s GPUs, with additional purchases of H20 units. These resources are distributed across multiple locations and serve purposes such as AI training, research, and financial modeling.

  • Financial Investments: The company's total capital investment in servers is around $1.6 billion, with an estimated $944 million spent on operating costs.

  • Origins and Funding: DeepSeek originates from High-Flyer, a Chinese hedge fund that adopted AI early and heavily invested in GPUs. In 2023, High-Flyer launched DeepSeek as a separate venture solely focused on AI. Unlike many competitors, DeepSeek remains self-funded, giving it flexibility and speed in decision-making (and deeper privacy on those numbers).

🧩 As plausible as this feels, it could just as easily be US led companies trying to downplay the rise of DeepSeek. Grand strategy from China to slow down venture capital spending in the US? Or real threat US companies want to sweep under the rug?

Security researchers from Wiz have discovered that DeepSeek, a Chinese AI platform, inadvertently left a critical database exposed online. This unsecured database contained over one million records, including system logs, user-submitted prompts, and API authentication tokens, making it accessible to anyone who encountered it.

  • Discovery and Response: The exposed database was identified by Wiz researchers, who promptly notified DeepSeek through multiple channels. Within approximately 30 minutes of the notification, DeepSeek secured the database, though it remains unclear if any unauthorized parties accessed the data prior to its protection.

  • Data Sensitivity: The leaked information encompassed user interactions with DeepSeek's AI services, such as chat prompts and system logs, as well as API keys used for authentication. This level of exposure could potentially allow malicious actors to manipulate the system or access further sensitive information.

  • Security Implications: Experts highlight that such a significant oversight indicates a lack of maturity in DeepSeek's security practices, suggesting that the service may not be suitable for handling sensitive data until more robust security measures are implemented.

🛡️ The incident intensifies scrutiny over DeepSeek’s murky origins and practices. No doubt the US government and companies would like nothing more than to see major leaks occur from DeepSeek to undercut it’s legitimacy at this juncture…

 🆕 Updates

📽️ Daily Demo

🗣️ Discourse