AGI by when?

AI PhD breaks down AGI predictions

Author

Omer Ben-Ami
April 06, 2025

This week I’m bringing to you our full breakdown of OpenAI’s AGI timeline.

I met Matt Baughman, University of Chicago PhD, to get his expert’s take on OpenAI’s prediction - AGI by 2027.

What is AGI and Why Does the Timeline Matter?

While the exact definition of AGI is still debated, it is generally agreed that it is a system that can outperform humans at most economically valuable work.

Unlike today's AI systems that are designed for specific tasks, AGI would have human-like ability to learn, reason, and adapt across many different areas.

So you’d definitely want to keep an eye on AGI’s timeline.
Researchers consider 2027-2030 to be the likely window.
In our conversation we went deep into the parameters that influence this timeline.

Key Expert Predictions

The episode covers different expert predictions on when we might achieve AGI. Here are the main viewpoints simplified:

Near-term Predictions (2025-2030)

Some experts believe AGI could arrive very soon - within this decade.
They point to:

  • The rapid progress we've seen with large language models like GPT-4

  • The increasing capability of AI to learn from fewer examples

  • Improvements in reasoning abilities of current AI systems

  • Continuous progress on compute power, and incresed model efficiency

These experts see current technology as already showing early signs of general intelligence that will quickly improve.

Factors Affecting the AGI Timeline

Several key factors could speed up or slow down progress toward AGI:

  • Computing Power: Bigger and faster computers help AI researchers train larger models

  • Data Quality and Quantity: AI systems need vast amounts of high-quality data to learn from

  • Algorithm Breakthroughs: New ideas and approaches that make AI learn better or more efficiently

What To Expect This Year

Before we reach true AGI, we're likely to see:

AI Systems with Narrow but Strong Capabilities

  • AI that excels in specific domains but can't transfer skills to new areas

  • Tools that automate specific professional tasks but need human oversight - will become way more ubiquitous. Capable agents fall into this category.

🔧 Tool of the week

Raycast is by far amongst the top 3 tools I use the most.

As you are an R4A reader, I wouldn’t be surprised if you’re already using Raycast, so here are some flows you might want to add to your Raycast’s configs:

  • Snippets - I configured “~”+shortcut for snippets I’m using all the time.
    Example - I’m tying into Cursor “~ncc” which stands for no-code-change, and it turns into
    without changing any code, explain to me . If there's specific area related in the codebase show it to me.

  • Clipboard search - remedy for all those times you copied something just to paste it elsewhere, and then you copied something else and now.. 🤷🏻‍♂️
    this completely solves this problem!

Raycast’s clipboard search

Your networks is your net-worth

You have a friend who loves AI, and tools.
Maybe you haven’t spoken in ages.
This is your excuse to poke them.
Share it with one person you need to catch up with.

You can copy this:

Hi!
It’s been ages.
I just read this newsletter and was thinking you might want to check it out.

Anyways, we should catch up! Should we pin something down before Passover?

Sayo-Nara 👋🏻
Omer