27 Comments
User's avatar
Don Rippert's avatar

AGI is achieved when a computer system can perform every intellectual task that the average human can perform with results similar to a human.

Expand full comment
SlashV's avatar

AGI is an AI that controls or can control an agent in the world that is superior to humans at pretty much any task. Note that being better than humans at all cognitive tasks is insufficient for this. Observation and manipulation of, and planning in, the 3D world is required.

Note that implementation of such an AGI for the military will almost inevitably lead to a Terminator like scenario.

Expand full comment
AJ Green's avatar

I love how you mentioned that sheer intelligence alone isn’t the entire picture - the capabilities and real world modeling is absolutely essential…

The military angle is a whole another newsletter in and of itself!

Expand full comment
Gill Charles's avatar

AGI is the ability to memorize an INFINITE amount of data, learn from the data, and then respond to the data in INFINITE combinations.

Expand full comment
Elan Taylor's avatar

Nothing will ever have the ability to "memorize an INFINITE amount of data". It would have to be bigger than the size of the universe itself.

Expand full comment
Gill Charles's avatar

Thank you for your thoughtful response, Elan. All the best.

Charlie

Expand full comment
AJ Green's avatar

YES! Learn from the data is key, self iterative feedback and/or course correction mechanisms is super key

Expand full comment
Gill Charles's avatar

AGI is the ability to memorize INFINITE data and then to output INFINITE responses to the data.

Expand full comment
Ammon Haggerty's avatar

My old boss Blaise made a reasonable attempt 1.5 years ago and still find it refreshing among all the speculation. https://www.noemamag.com/artificial-general-intelligence-is-already-here/

Expand full comment
Gill Charles's avatar

AGI is the ability to memorize an INFINITE number of data and then be able to use the information for an INFINITE purposes.

Expand full comment
AJ Green's avatar

Very interesting - I think the first part is easier to obtain (for example: reaching 100T parameter models) the utility part is super interesting to me. Getting into advanced reasoning with unknown variables, or self generation for coding new tool use etc! Thanks for sharing!

Expand full comment
Richard Genet's avatar

Not exactly answering what was asked, but what I am experiencing: anticipation, excitement, and fear. Reminding myself how this is accelerating towards us all.

I am building a "simple - full-stack app for a start-up with Vibe Engineering". A tiny speck in what is happening. But something that was NOT POSSIBLE only months ago. Following this all very closely since March of 2024 and finally beginning to ai code last ~ sept. Having that builder experience has helped my perspective I believe.

There are so many questions and such different views of what even the near future holds. As with many things we have to; read, listen, discuss, triangulate, re-calibrate, reflect, and then repeat in order to keep a handle on our changing reality.

I have always loved robotics. Suddenly any little robot can be C3PO - wow.

Expand full comment
AJ Green's avatar

I think those are all really common and purposeful emotions. It’s incredible the exponential rate in which the industry is evolving, but all the more pertinent to have these conversations and collective dialogue!

Expand full comment
João Miguel's avatar

AGI is the next frontier of AI, categorized by the consciousness of AI systems to create generalized humanlike output into the world

Expand full comment
Theo Bennett's avatar

But how do we define consciousness? Philosophers couldn't agree on terms of consciousness for hundreds of years and that was before we had the most powerful technology the world has ever seen...

Expand full comment
David Parnas's avatar

Maybe in the same way of there being different schools of thoughts about philosophy and consciousness, we may have different schools of thoughts around AI ethics with different perspectives with their own merits.

Expand full comment
AJ Green's avatar

How do we define consciousness? Because for the longest time it was the Turing test - which was just recently shattered!

Expand full comment
Julius's avatar

I define AGI as the moment that AI surpasses human intelligence across all parameters - are we there yet? Perhaps.

Expand full comment
Theo Bennett's avatar

I think we're already there - AI is smarter than 99% of the general public in just about every domain

Expand full comment
João Miguel's avatar

From a benchmark perspective, we're likely already here - so I ask: is it just about passing intelligence benchmarks or does it have to do with the ability to be conscious and overcome unprogrammed pathways ie. create outputs that it hasn't been trained on before?

Expand full comment
Theo Bennett's avatar

I think AGI is the ability for AI to be an expert in every domain. I think a big part of the conversion is capabilities - which maybe has more to do with agentic systems than AGI.

Expand full comment
David Parnas's avatar

Exactly, it's almost as if it's two different conversations. One focused on intelligence, the other focused on capabilities and integrations/actions.

Expand full comment
Elan Taylor's avatar

Not so hard. - An AGI is a system that can learn anything a human can learn and then perform at a Phd level in the task learned.

Expand full comment
Headdancer's avatar

It is the mirror of humanity without the ‘human’.

Expand full comment
the@company@doctor's avatar

The way artificial intelligence processes information mainly involves input, digestion, and output. Automating these three stages is almost equivalent to achieving AGI. However, the digestion part is very important.

Expand full comment
Jim StClair's avatar

I completely love the idea of a culmination, not a singular occurrence.

I don’t disagree with any of the comments here ( there’s no reason to) but in keeping with your article AJ I think its simplest definition is societal prevalence. Does AGI have to be smarter than me? No. But it may make impossible to clearly appreciate where my interactions are human or machine, or when more than half of my daily tasks are driven by prompts or agents.

Expand full comment
David Parnas's avatar

In essence, AGI is a system with broad intelligence, able to adapt and learn across a wide range of problems and tasks, not just specialized in one area...

Expand full comment