OpenAI Launches Deep Research for Comprehensive Web Analysis

OpenAI has unveiled its new tool, Deep Research.

It conducts multi-step research across the internet for complex tasks.

Powered by an optimized version of the upcoming o3 model, Deep Research excels at web browsing and data analysis. Just provide a prompt and let ChatGPT work. It scours text, images, and PDFs online.

Within minutes, it synthesizes hundreds of sources into a report akin to that of a research analyst.

The company claims it accomplishes in tens of minutes what would take a human many hours.

However, OpenAI admits the tool is still in early stages.

It may struggle with distinguishing authoritative information from mere rumors and often fails to convey uncertainty accurately.

Deep Research is currently available on the web version of ChatGPT.

Mobile and desktop apps will follow in February.

This launch marks OpenAI’s second foray into autonomous agents this year, following the preview of Operator in January—a tool designed for tasks like to-do lists and vacation planning.

Built for intensive knowledge work, Deep Research targets professionals in finance, science, policy, and engineering. It’s also ideal for discerning shoppers in need of detailed, hyper-personalized recommendations.

Every output comes fully documented, complete with clear citations and a summary of its reasoning. This makes verifying and referencing information a breeze.

Using Deep Research is straightforward. Select the “deep research” option in ChatGPT, enter your query, and attach any relevant files for context. A sidebar displays the steps taken and sources used, with the process taking between 5 to 30 minutes.

Trained with end-to-end reinforcement learning on complex browsing tasks, Deep Research plans and executes multi-step data searches.

It can also process user-uploaded files, generate graphs with Python tools, and embed images from websites.

This new tool signals a significant stride toward OpenAI’s broader ambition of developing AGI.

Google Removes AI “ Harm” Clause

Google has removed a key passage from its AI principles that once barred the use of its technology in harmful applications, including weapons.

The original guidelines—titled “AI applications we will not pursue”—explicitly mentioned avoiding “technologies that cause or are likely to cause overall harm.” Screenshots viewed by Bloomberg confirm that this language is now missing.

In a blog post published Tuesday, a Google spokesperson responded to inquiries about the change. Senior figures at Google, including James Manyika and Demis Hassabis, wrote: “We believe democracies should lead in AI development, guided by core values like freedom, equality and respect for human rights.”

They stressed that collaboration among companies, governments, and organizations sharing these values is essential for building AI that protects people, promotes global growth, and supports national security.

Critics say the removal could signal a worrisome shift in priorities. Margaret Mitchell, formerly of Google’s ethical AI team and now chief ethics scientist at Hugging Face, was blunt: “Having that removed is erasing the work that so many people in the ethical AI and activist spaces had done at Google.

This change comes amid a broader realignment among major tech companies. Earlier this year, Meta and Amazon scaled back diversity and inclusion programs, igniting debates over the balance between ethical guidelines and competitive pressures in AI development.

Tracy Pizzo Frey, who once led Responsible AI at Google Cloud, noted that these principles were more than just policy—they built trust. “Responsible AI is a trust creator,” she said.

Internal debates over ethics have long simmered at Google, especially as the race to regain AI leadership intensifies.

Amazon’s $100B Plan for 2025

Despite last week’s buzz that DeepSeek might trigger lower AI budgets, Big Tech isn’t slowing down at all.

In fact, spending is ramping up dramatically.

Amazon is leading the charge.

The company announced plans to invest over $100 billion in capital expenditures for 2025. CEO Andy Jassy explained during the Q4 earnings call that most of this spending will boost AI capabilities in AWS.

He pointed out that Q4 2024’s $26.3 billion capex is indicative of an annual spend around $105.2 billion—a huge leap from the $78 billion in 2024.

Amazon dismisses worries that cheaper AI will hurt revenue. Jassy argues that lower prices will only fuel higher demand. With its wide range of AI services, AWS is set to benefit significantly.

Other tech companies are following suit.

Meta’s CEO Mark Zuckerberg declared the company would invest “hundreds of billions” in AI over the long term. Meta is set to spend at least $60 billion on capital expenditures in 2025, focusing largely on AI.

Alphabet is also stepping up. It’s capex for 2025 has surged by 42% to $75 billion. CEO Sundar Pichai explained that lower AI costs will enable a broader range of applications.

Microsoft isn’t far behind either. Last month, the company announced it will invest $80 billion in AI data centers in 2025. CEO Satya Nadella even tweeted the Wikipedia page for Jevons Paradox—illustrating the idea that lower costs can drive higher demand—as discussions about DeepSeek heated up.

Whether Jevons Paradox will fully apply to Big Tech remains to be seen.

For now, one thing is clear: AI investments are far from slowing down.

Meta Teams Up with UNESCO on Open-Source AI Language Initiative

Meta is launching a new program with UNESCO. The Language Technology Partner Program aims to gather speech recordings and transcriptions to fuel openly available AI.

The initiative calls for collaborators to contribute over 10 hours of speech recordings with matching transcriptions. It also welcomes large volumes of written text and sets of translated sentences in diverse languages.

Partners are already stepping forward. For instance, the government of Nunavut—where residents speak Inuktut—is on board.

Meta’s blog post explains, “Our efforts are especially focused on underserved languages, in support of UNESCO’s work.”

The goal is clear: create intelligent systems that understand and respond to complex human needs, regardless of language or culture.

Alongside this, Meta is releasing an open-source machine translation benchmark.
This benchmark is crafted by linguists, supports seven languages, and is accessible via the Hugging Face platform.

Anyone can contribute. Meta frames both initiatives as philanthropic. Yet the company stands to benefit from improved speech-recognition and translation models.

These upgrades will enhance Meta AI and support features like automatic translation for creators. Last September, Meta began testing a tool for translating voices in Instagram Reels, allowing for dubbing and auto-lip-sync.

However, Meta’s handling of non-English content has faced criticism.

Reports claim nearly 70% of Italian and Spanish COVID misinformation went unflagged compared to 29% for English.

Leaked documents also reveal that Arabic posts are often mistakenly flagged as hate speech.

Meta insists it is taking steps to improve its translation and moderation technologies.