🤝 The AI that turns hours of research into minutes
Dear curious mind,
The AI landscape is evolving rapidly, and today we take a look at its currently most exciting development: Deep Research Tools. From Google to OpenAI to Perplexity, these new AI assistants are transforming how we discover and process information. Let's explore what this means for the future of knowledge work.
In this issue:
💡 Shared Insight
Deep Research Tools: The AI Revolution in Knowledge Discovery
📰 AI Update
Google One AI Premium Now Includes Enhanced NotebookLM Features
Perplexity Joins AI Research Race with Free Deep Research Tool
GPT-4o's Stealth Upgrade: New Model Version Makes ChatGPT More Human Than Ever
🌟 Media Recommendation
Podcast: Naval Ravikant on Parenting, AI, and the Future of Work in the All-In podcast
💡 Shared Insight
Deep Research Tools: The AI Revolution in Knowledge Discovery
The world of AI is constantly evolving, and one of the most recent and exciting developments is the emergence of "Deep Research" capabilities. This new approach to knowledge discovery is going to revolutionize how we gather information and conduct research. While Google, OpenAI, and Perplexity all use the term "Deep Research," each company offers a slightly different implementation. This article delves into the core concepts of Deep Research, explores how it differs from traditional chatbots and reasoning models, and examines its potential impact on the economy and the future of work.
What is Deep Research?
Unlike traditional chatbots that provide quick answers or reasoning models that focus on logical deduction, Deep Research tools act as autonomous research assistants. They can search and dig through vast amounts of information, synthesize findings, and generate comprehensive reports with citations.
Imagine having a tireless research assistant who can identify and analyze hundreds of sources in minutes, extracting key insights and presenting them in a well-structured format. That's the power of Deep Research.
Using Deep Research
There are already various open-source realizations, but in this article I want to highlight three tools from major AI companies.
Google's Deep Research
This realization was already released in December 2024 and is based on the already outdated Gemini 1.5 Pro. Its non-reasoning model architecture limits the analytical depth of the reports, but I am still happy with the results in most runs. It shines when you give it detailed instructions. As a paying subscriber of the Google One AI premium plan, which is around $20 per month, you can access Deep Research in the Gemini app on your phone and on its website.
OpenAI's Deep Research
Released in February to the public, and it uses the reasoning model o3 which is currently not available as a standalone model. Expert evaluations indicate that its reports demonstrate exceptional quality and depth of analysis. So far it is available only in the $200 tier of ChatGPT. However, it is announced to become accessible through the $20 subscription and even the free tier with a strongly limited number of runs per month.
Perplexity's Deep Research
In contrast to the other two solutions, this version can even be accessed as a free user. It also uses a reasoning model, and the shared benchmark results look very promising. More information are shared in the news section of today’s issue.
Will Deep Research Replace Jobs?
The rise of AI technologies has sparked concerns about job displacement, which is also fueled by a recent post from Sam Altman, CEO of OpenAI, on 𝕏.
While AI will undoubtedly automate many tasks, it's more likely to augment human capabilities and create new opportunities and not replace jobs entirely. By automating tedious tasks and providing valuable insights, it can empower workers to be more productive and focus on higher-value activities that require human skills like creativity, teamwork, and leadership. Deep Research tools can greatly increase the productivity of persons in the workplace.
Conclusion
Deep Research represents a significant advancement in AI, offering a powerful new way to gather information and provide insights. While it's essential to address potential challenges, the benefits of Deep Research are undeniable. By embracing these tools and adapting to the changing landscape of work, we can unlock new levels of productivity, innovation, and economic growth. When integrated thoughtfully, AI can lead to enhanced efficiency and added value creation.
Deep Research is a valuable tool for both professionals and casual users. If you're eager to explore the possibilities of Deep Research, I encourage you to try out the tools offered by Google, OpenAI, and Perplexity (Sidenote: all three are accessible in the EU). You will be surprised at what you discover!
📰 AI Update
Google One AI Premium Now Includes Enhanced NotebookLM Features [Google blog]
NotebookLM is, from my point of view, the currently best tool to extract information from multiple documents. The tool is mostly known for the podcast like audio overviews it generates, but really shines when chatting with your own documents. With an active Google One AI subscription, you get now the NotebookLM Plus features. Besides higher rate limits, the main difference is the possibility to customize the length and style of the AI responses.
Perplexity Joins AI Research Race with Free Deep Research Tool [Perplexity blog]
Perplexity has joined Google and OpenAI as the third major player to launch an advanced research tool called Deep Research. While the core concept is the same, the version from Perplexity stands out by offering five queries per day for free users. The tool shows promising performance on benchmarks like Humanity's Last Exam and SimpleQA. To understand what it is capable of, take a look at Perplexity's announcement blog post and try it yourself.
GPT-4o's Stealth Upgrade: New Model Version Makes ChatGPT More Human Than Ever [@sama post on 𝕏]
OpenAI has silently rolled out a significant upgrade to the GPT-4o model in ChatGPT. Users have noticed significantly improved writing capabilities, with responses that feel more natural and human in tone. The enhanced model has climbed to the top spot in the lmarena leaderboard, sharing first place with two Gemini-2.0 models. This is a dramatic improvement from the previous GPT-4o version, which is currently placed at rank 16.
🌟 Media Recommendation
Podcast: Naval Ravikant on Parenting, AI, and the Future of Work in the All-In podcast
Naval Ravikant shared fascinating insights in a recent episode of the All-In podcast.
Key highlights from the discussion:
AI and Employment: Naval beliefs current AI job loss predictions are incomplete as they don't account for new industries AI will enable. He envisions AI unlocking large-scale projects previously deemed impossible, such as underwater and space habitation.
Future Skills: The skill’s landscape is shifting. While AI can replace technical abilities, soft skills like creativity, judgment, psychology, leadership, and communication remain crucial.
Parenting Philosophy: Naval advocates for "taking children seriously," treating them with adult-level respect and autonomy. He implements this through negotiation rather than enforced behavior, requiring only daily math/programming and reading while allowing freedom in other areas.
The discussion touched on the 4-point AI approach shared by JD Vance, Vice President of the United States, at the AI Action Summit:
Maintain American AI as the gold standard
Avoid excessive regulation
Keep AI free from ideological bias
Ensure AI benefits worker growth and job creation
My take: Naval's perspective on AI's future acknowledges opportunities and challenges. His insights on parenting in the AI age are particularly relevant as we prepare the next generation for a rapidly evolving technological landscape.
Disclaimer: This newsletter is written with the aid of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.