Alex McFarland

This framework is a shift toward practical AI-human collaboration in research, prioritizing efficiency while maintaining human expertise in the research process.

⚠️ Key limitations

- Generated papers score 1.45 points below average NeurIPS submissions

- AI reviewers consistently rate papers 2.3 points higher than human reviewers

- Requires active human oversight for optimal results

🏆 Model performance

- o1-preview: Highest scores in usefulness and clarity
- o1-mini: Best experimental quality scores
- GPT-4o: Most cost-efficient, fastest processing

📊 Performance metrics

- Speed: Complete workflow in 1,165.4 seconds using GPT-4o
- Cost: $2.33 per paper (84% reduction from traditional methods)
- Quality: Papers score 4.38/10 with human collaboration

👥 A virtual research team

- PhD Agent: Handles literature reviews and research planning
- Postdoc Agents: Refine experimental approaches
- ML Engineer Agents: Manage technical implementation
- Professor Agents: Evaluate research outputs

🎯 What makes it different

- Creates a virtual research team with specialized AI agents
- Focuses on enhancing (not replacing) human researchers
- Enables flexible compute allocation based on resources and needs

AMD and Johns Hopkins University have released something interesting - an open-source framework on how AI and humans can collaborate on research.

⬇️

AI School #1: AI agents and their tools Using Google's whitepaper "Agents" to understand how these systems interact with our world.

Ever wondered what it would be like to learn about AI... from AI itself? Welcome to my AI School, where artificial intelligence isn't just the subject – it's your teacher.

aidisruptor.ai/p/ai-school-...

Most people rely on a single source when trying to understand tech events - whatever summary article appears first in their feed.

But the real insights come from combining three distinct perspectives that each serve a unique purpose.

My three-source framework for NotebookLM ⬇️

How to digest big tech events like CES with NotebookLM Watch now (18 mins) | A three-source NotebookLM framework for understanding any event.

aidisruptor.ai/p/how-to-dig...

Pro tip: Start small.

Pick a recent announcement you care about and try the process. You might be surprised by what you discover when you combine these different sources.

3️⃣ The community insight: Collective wisdom

By feeding in structured community discussions (I'll show you exactly how in the video), you tap into the collective expertise of developers, industry insiders, and power users who spot implications that journalists miss.

2️⃣ The analyst perspective: Critical context

I specifically look for articles with unique angles - not basic summaries.

This second layer helps you understand market implications and read between the lines.

1️⃣ The raw source: Your foundation

When you feed a keynote transcript into NotebookLM, you're creating a base layer of knowledge that lets you verify claims, spot patterns, and catch details.

This becomes your ground truth - especially important when everyone's racing to interpret what was said.

Most people rely on a single source when trying to understand tech events - whatever summary article appears first in their feed.

But the real insights come from combining three distinct perspectives that each serve a unique purpose.

My three-source framework for NotebookLM ⬇️

All of this NVIDIA/CES talk got me working on a new guide for AI Disruptor.

How to use NotebookLM to understand any big tech event. You will need three types of content. Then you don’t have to worry about trying to keep up with all of this. It’s too much!

Hopefully will be out later today.

Our Discord: sidestack.io/aidisruptor

A clear guide to understanding AI agents Using Google's research to explore how agents think and act.

If you've been following AI developments, you definitely know about AI agents by now. Google dropped a great white paper back in September explaining agents, but for some reason it is now just circulating.

aidisruptor.ai/p/a-clear-gu...

The Dispatch #1: OpenAI's o3; DeepSeek's rise; xAI's $6b round Exploring test-time compute, China's AI ambitions, and the future of model scaling.

This week, I did something different with the podcast.

It is now "The Dispatch" - a weekly roundup where I hand-pick and analyze the most impactful AI developments I think you should know about.

Think of it as an educational resource.

A weekly technical briefing.

aidisruptor.ai/p/the-dispat...

If 2024 taught me anything, it's that AI's true potential isn't in the headlines or the hype.

It's in the quiet wins happening in home offices, classrooms, and small businesses across the globe. It's in watching someone transform their work process using tools they thought were "too technical."

Before the holidays: A letter to 4,000 AI enthusiasts An introduction long overdue.

Two days before Christmas, and here we are – nearly 4,000 of you. Whether you're celebrating the holidays or simply enjoying the year-end calm, I wanted to take this moment to pause, reflect, and finally share something I've been meaning to tell you.

aidisruptor.ai/p/before-the...

How to create a podcast with 2 AI tools Combine Claude + ElevenLabs to create personalized AI podcasts.

My next guide is here.

How to create a podcast with 2 AI tools.

aidisruptor.ai/p/i-turned-m...

Just finished up a new guide for tomorrow. I recently discovered a new tool for generating AI podcasts. It is way better and more natural than my previous workflows (NotebookLM or Claude + ElevenLabs).

BYOAI:Employees are hiding their AI usage New data shows 78% of workers are bringing AI to work.

aidisruptor.ai/p/byoaiemplo...

Most companies still think they can control how their employees use AI.

Here's the reality: 75% of knowledge workers are already using AI tools, whether their companies approve or not. And that number isn't going down - it's accelerating.

BYOAI = Bring Your Own AI

How I extract insights from 1000+ Reddit comments in minutes Turn messy Reddit threads into structured data for AI analysis.

This AI hack was sitting in front of me for so long and I didn’t even realize. It’s not an expensive tool or complex API.

I used to spend a lot of time analyzing reddit threads to understand what the community was saying on a given topic. Now I can do it in seconds.

aidisruptor.ai/p/how-i-extr...

You don't need ChatGPT Pro Why you won't see me dropping $200/month on an AI anytime soon.

Unless you give me a really good agent that can take over a few of my tasks autonomously, you won’t see me dropping $200/month for an AI tool right now.

90% of you do not need ChatGPT Pro. You would be better off spreading that $200 across several other AI tools.

aidisruptor.ai/p/you-dont-n...

The first AI Disruptor podcast We will create the best AI podcast out there.

I will create the best AI podcast out there.

Our first AI Disruptor podcast is out now. Check it out and share. Each week I'm running polls and feedback to make adjustments.

aidisruptor.ai/p/the-first-...

For a long time, I was crafting prompts by carefully choosing every word, agonizing over structure, testing and retesting until things worked just right.

Then I discovered something that changed everything: Anthropic actually built tools to handle all of this for us.

Stop writing prompts from scratch A guide (with videos) to using Anthropic's prompt optimizer tools.

You all asked, and it’s here. My in-depth guide (with videos) to using Anthropic’s prompt optimizer tools.

Don’t listen to anyone who tells you prompt engineering is complicated.

aidisruptor.ai/p/stop-writi...