Will models surpass humans across most capabilities, and should that prospect be welcomed?

Immad Akhund · May 12, 2025 · hedged

Advanced superintelligence is likely to arrive sooner than people expect

The pace of AI advancement over the past year has been relentless, which shifted him from skepticism

Scope: "I don't know if I've quite changed my mind all the way"; unsure whether it happens within five years

54:47 20VC Exclusive: Mercury Founder Launches First $26M Fund | Why Founders Should Take the Highest Price | Why Serial Entrepreneurs are Better | Why AI Is So Overhyped | The Future of Venture Capital with Immad Akhund

Fred Turner · Jul 18, 2026

Within a few years models will be ahead of humans on most capabilities, and this is exciting because it opens up unlimited intelligence

It opens up so many possibilities

64:26 20VC: $5BN in Revenue, 7 to 7,000 Employees in 9 Months, 206,000 Tests in a Single Day: The Craziest Story in Startups: Curative with Fred Turner

Julien Bek · Aug 24, 2026 · speculative

When AI moves from roughly 120 IQ today to 500 IQ, it will produce cures for diseases and transformations that make today's AI worries look insignificant

Vastly higher-intelligence systems will unlock discoveries in life sciences and biology that humans can't reach

Scope: he is not from the world of science

75:16 20VC: Inside Sequoia's Investment Committee: Lessons from Don Valentine, Doug Leone and Alfred Lin | How the SpaceX and Citadel Deals Went Down | What Sequoia Specifically Looks for in Founders with Julien Bek

Martin Casado · Jul 28, 2025

ASI is the most overhyped category in AI today

Scope: quickfire answer, no elaboration

58:17 20VC: a16z's Martin Casado on Anthropic vs OpenAI: Where Value Accrues | Cursor vs Replit vs Lovable: Who Wins and Who Loses | The One Sin in AI Investing | Why Open Source is a National Security Risk with China

Brendan Foody · Sep 15, 2025

The widely held belief that we will have superintelligence better than humans at everything within three years is totally wrong

52:48 20VC: Mercor: From $1M to $500M in 17 Months: The Fastest Growing Company in the World | How to Think About Margins and Revenue Sustainability in AI | Why Evaluation Benchmarks in AI are BS Today with Brendan Foody

Demis Hassabis · Apr 7, 2026

DeepMind's founding-era extrapolation of compute and algorithmic progress predicted AGI roughly twenty years out, and the field is pretty much on track against that prediction

The forecast was built by extrapolating compute and algorithmic progress, and timelines have not materially shifted for him since 2010

Scope: prediction originally made by cofounder Shane Legg in public blog posts

5:17 20VC: DeepMind's Demis Hassabis on Why AGI is Bigger than the Industrial Revolution | Why LLMs Will Not Commoditise & We Have Not Hit Scaling Laws | Bottlenecks in AI & The Energy Crisis Caused By AI | Whether AI Will Do More to Harm or Help Inequality

Bucky Moore · May 5, 2025 · hedged

AGI is effectively already here in certain pockets, because test-time compute lets models explore thousands of paths to an answer in ways no army of smart humans could

Test time compute brings the AlphaGo dynamic to many domains: humans predict the next token one path at a time, while these models can try a thousand attempts and then choose the best next token

Scope: 'in certain pockets' / 'in a lot of ways'; framed as how it feels to him

53:27 20VC Exclusive: Why Mega Platforms Will Win in VC | Why You Cannot Do VC If You Do Not Do Pre-Seed | Why Market Sizing is BS | Where Will Foundation Models Build/Buy Apps vs Where Will They Not with Bucky Moore

Anton Osika · Aug 18, 2025

AI is already smarter than humans when given full context or wrapped in a purpose-built system for its weaknesses — and most people wrongly disagree

Raw models often appear very stupid, but that is a context and scaffolding problem, not an intelligence ceiling

Scope: conditional on being given all the context or a purposeful system around it

56:10 20VC: Lovable CEO Anton Osika on $120M in ARR in 7 Months | The Honest Truth About Defensibility and Unit Economics for AI Startups | The State of Foundation Models: Long Grok, Short OpenAI, Why | Replit vs Lovable vs Bolt: What Happens

Ethan Mollick · Jul 31, 2024

AI is already profoundly better than most people think and will keep getting better — a view most people around him don't actually hold

61:45 20VC: Is More Compute the Answer to Model Performance | Why OpenAI Abandons Products, The Biggest Opportunities They Have Not Taken & Analysing Their Race for AGI | What Companies, AI Labs and Startups Get Wrong About AI with Ethan Mollick

Demis Hassabis · Apr 7, 2026

In most domains AI is ahead of where he and the field expected it to be, with video models and interactive world models as the clearest examples

Systems like Genie would have amazed him if shown to him five or ten years ago

Scope: critical capabilities such as continual learning are still missing

7:39 20VC: DeepMind's Demis Hassabis on Why AGI is Bigger than the Industrial Revolution | Why LLMs Will Not Commoditise & We Have Not Hit Scaling Laws | Bottlenecks in AI & The Energy Crisis Caused By AI | Whether AI Will Do More to Harm or Help Inequality

Edwin Chen · Jul 21, 2025 · hedged

If AGI has not arrived by 2040, the causes will be a shortage of further algorithmic/data-leverage breakthroughs and the slowness of gathering the necessary real-world data

Some breakthroughs, e.g. curing cancer, may require real-world experiments and studies that simply take time, and it's unclear whether simulation can speed them up

38:10 Agi delay would stem from slow real world data gathering and missing algorithmic breakthroughs

Richard Socher · Aug 18, 2023

People are overly optimistic about AGI because they extrapolate a burst of progress forward indefinitely, when progress is neither linear nor exponential

Aviation went from first powered flight to supersonic metal aircraft in decades and then stalled — we now often fly slower — and space re-entry has regressed from landing a space shuttle to dropping capsules into the ocean

31:32 Historical analogies like aviation and space show progress can plateau or regress undermining agi extrapolation

Richard Socher · Aug 18, 2023

There is a ceiling on how superhuman AI language can be, because language is a human construct and humans can only consume it sequentially

Superhuman language would mean thousands of parallel streams, which humans cannot register; you can't read five books or hold five conversations at once, so language beyond human comprehension is meaningless

33:05 Language cannot be superhuman beyond human sequential comprehension capacity

Jonathan Ross · Sep 29, 2025

LLMs are the telescope of the mind: they currently make humans feel small, but within a century we will see the vastness of intelligence they revealed as beautiful rather than threatening

Historical analogy to Galileo's telescope, which got him in trouble for revealing that the universe was far larger than imagined and made humanity feel small, yet we came to find that grandeur beautiful

76:53 Ai revealed vastness of intelligence will be seen as beautiful not threatening over time

Avishai Abrahami · Jul 13, 2026

The industry has moved the goalposts on AGI, redefining it to describe systems that still cannot answer the big scientific questions originally used to motivate it

A year ago people spoke of AGI as something you could discuss unsolved physics problems with; now people say AGI is here or near while those capabilities remain absent

50:42 Agi goalposts were moved the original capabilities are still missing

Harry Stebbings · Jul 13, 2026

Avishai is being too generous to current AI models, which still fail badly at basic tasks

He asked Gemini to remove people from the background of a beach photo and it replaced his mother's face with a different woman

51:27 Everyday basic task failures show capability is overstated