Ex Google CEO Eric Schmidt on AGI timeline. San Francisco Consensus thinks "within 2 to 3 more cranks, with each crank being around 18 months, you get to AGI. They define AGI as an intelligence greater than sum of all human intelligence" max 4.5 yrs

That’s a striking timeline. If AGI really arrives within ~4–5 years, the bigger question won’t be when it comes, but how prepared our ethics, governance, and society are when it does.

👍👍

In 2016/2017, he told everyone to use Gmail as the most secure email while a massive torrent file circulated with many people's passwords. Gmail notified no one in the data.

18-month cranks make good podcasts, bad planning.” Capability can jump, but deployment is plumbing: energy, security, audits, liability. If the timeline is 4.5 years, the constraint isn’t IQ, it’s governance under load.

LLMs use statistical models to predict the next token this is not anywhere close to intelligence.

This isn’t AGI. It’s Doppler radar for cognition — sophisticated pattern matching driven by probability. Powerful? Yes. Human coherence? No. Math scales prediction. It doesn’t replace judgment.

Wrong: AGI is not an intelligence greater than the sum of all human intelligence; that is ASI.

No one should be allowed to build this. Where the fuck are world governments?

@lu_sichu AGI prediction under your preferred definition of choice?

Good luck with that.

His definition of AGI is unclear. He should first define what "intelligence" is. Otherwise, he is defining AGI using an undefined term, which is sort of pointless.

More importantly how exactly is AGI measured. If it's not measureable it's impossible to tell when we are there? Thoughts @Grok?

I’m betting against it if they think this is a crank-approach from existing probabilistic LLM tools. Not happening.

The world is not going to know what hit it.

Queue Spoiler Alert 2 of 2 🥷🥷 #pioneerdj

They're delusional, as defined by how much money they have riding on it. You can't brute force your way to AGI using LLMs, which will never be able to do this. But you can crash an economy with it, which they're on track to do.

Eric is full of shit with his timeline. Scaling doesn't fix errors; we have no alternative model, and we're decades from the required energy.

Hot take, the timeline is less important than the definition doing all the work here. "Greater than the sum of all human intelligence" is not a product milestone, it's a metaphysical flex. We will get systems that feel AGI in many jobs before we get anything like that grand total comparison. The cranks are real, but so are the bottlenecks, data, energy, reliability, adversarial robustness, and the fact that capability does not equal control. The right question is which specific powers show up in 4.5 years, autonomous research loops, persuasion at scale, cyber, bio design, and at what cost per run. Governance needs to be tied to measurable thresholds, not vibes about a finish line. If we prepare for the concrete failure modes, the exact date stops mattering.

really depends on how you measure it!

Today's machine based LLM's are not PhD's and here is Yann LeCun the Chief scientist of Meta AI telling everyone it's not happening

Timelines are compressing faster than most are ready to admit.

2030

18-month cycle will coincide with new nvidia GPUs per cycle

this is myth intelligence is not quantitative

Having AGI easily propagate by itself and leave its ‘box’ could be the only way to have massive (ethical) acceleration.

loading ... it is currently feeding on our billions of prompts each day ...

This makes a mistake analogous to saltationism in evolutionary theory. it ain’t likely to be correct! What we have mounting evidence of, given the plateauing of LLMs, is the notion that the brain relies on more than neural nets as currently modeled (perhaps QM etc).

And this is a goal why? Aside from gobs of cash for a few dozen people and mass unemployment, what’s the upside?

“Stochastic parrots" remain parrots, no matter the steroids involved. LLMs are incredible tools, but conflating scaled-up pattern matching with AGI is a hype-fueled illusion that conveniently attracts capital. Follow Eric Schmidt’s incentives, not just the promises.

Imagine explaining to your future AI overlord that you peaked in TikTok dances

The timeline is speculative. Defining AGI as that extreme makes the prediction less useful. It's a theoretical target, not a product roadmap.

What is it even supposed to mean, to have an intelligence greater than a humans'? Human intelligence is an advanced nervous system, nothing more. It is not hard to recreate it, but to exceed it? And how would they know it's works, they would be impressed?

We are so close to AGI yet every tech company is lowering their AI sales targets... yeah right buddy

A more unlikable person may not exist

Oh so now the defintion of AGI is actually ASI got it