Human vs AI research effort

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

> points to massive intelligence explosion

> paper claims AI could drive century's worth of tech progress in less than a decade

> AI research effort currently growing 600x faster than human cognitive research effort; predicted over the next decade:

• moderate scenario: >300x faster on avg

• rapid scenario: >1000x faster on avg

> claims AI-human cognitive parity could arrive in a matter of years

> after parity, AI research effort would still grow 5-25x per year (vs human growth of 1.04x/year)

> eventually, no. of human-level AI researchers would exceed all 8 billion humans

> China's models catching up to the US

> gap between countries closing on LMSYS arena:

• in Jan '24: best U.S. model outperformed best Chinese model by 9.26%

• in Feb '25: gap closed to 1.70%

> performance gap closing across many benchmarks (2023-2024):

• MMLU: 17.5 -> 0.3 percentage points

• MMMU: 13.5 -> 8.1 percentage points

• MATH: 24.3 -> 1.6 percentage points

• HumanEval: 31.6 -> 3.7 percentage points

> notable AI models: particularly influential AI models based on SOTA advancements, historical significance, high citation rates, etc.

> most from the US, then China

> in 2024, other top orgs include Apple (4), Anthropic (3), Mistral (3), ByteDance (2), DeepSeek (2), etc.