The Numbers Museum

Every specific figure ever claimed on the show. 76 match this view. Red rows are numbers the cited sources contradict.

All Dollars Multiples Percentages Big numbers

Figure	As spoken	The claim	Who	When	Checked?
10,000×	“10,000 times”	Alexandr Wang: High-quality frontier data is 10,000x more valuable than enterprise data.	Alexandr Wang	May 22, 2024	—
1,000×	“1000 x”	Brown says AI models optimized his PhD poker algorithms by 1,000x	Noam Brown	Jun 26, 2026	—
750×	“750 times”	Prince: OpenAI and Anthropic Refer Traffic 750x and 30,000x Less Than Google	Matthew Prince	Aug 7, 2025	Supported
240×	“240 X”	Gil: GPT-4 Equivalent Token Costs Dropped 240x in 18 Months	Elad Gil	Nov 7, 2024	Supported
200×	“200 X”	Gil: LLM API costs dropped roughly 200x in 18 to 24 months	Elad Gil	Sep 12, 2024	Supported
200×	“200 times”	Liang: Google beat Stanford's medical AI with a model 200x larger	Dr. Percy Liang	Apr 25, 2023	Supported
180×	“180 X”	Gil: GPT-4 equivalent inference costs dropped 180x over 18 months	Elad Gil	Feb 7, 2025	Supported
150×	“150 X”	Gil: GPT-4-level token pricing dropped 150x in 21 months	Elad Gil	Feb 19, 2026	Supported
100×	“hundred x”	Complex task compute costs fall 10x to 100x per model release	Noam Brown	Jun 26, 2026	—
100×	“100 X”	Taylor: Nuclear is a fake industry charging 100x margins	Isaiah Taylor	Aug 15, 2026	—
100×	“hundred X”	Human-in-the-loop oversight will fail as AI agent actions scale exponentially	Maxim Bar Kogan	May 28, 2026	—
100×	“hundred times”	Tiwari: Blackwell GPUs achieve 90 to 100x inference efficiency over Hopper	Neil Tiwari	Feb 26, 2026	Supported
100×	“hundred X”	Gil: GPT-4 equivalent token costs dropped over 100x in 2024	Elad Gil	Jan 8, 2026	Supported
100×	“hundred X”	Davis: Democratizing AI Compute Will Increase AlphaFold-Level Breakthroughs 10x to 100x	Jared Quincy Davis	Aug 22, 2024	—
100×	“hundred times”	Glyman: Preventing spending is 100x better than earning 1% cash back	Eric Glyman	Aug 15, 2024	—
100×	“hundred X”	Guo: AI agents cannot automate early-stage VC due to missing data	Sarah Guo	Mar 21, 2024	—
100×	“hundred X”	Gil: Existing models yield 10x to 100x gains without base model scaling	Elad Gil	Oct 5, 2023	—
100×	“hundred times”	Generative Image Models Will Soon Achieve 25 Frames Per Second	Emad Mostaque	May 3, 2023	Held up
100×	“hundred times”	Zipline operates at 100 times the scale of its closest competitor	Keller Rinaudo Cliffton	May 10, 2023	—
88×	“88 X”	Gil: o1-equivalent model inference costs dropped 88x in 11 months	Elad Gil	Feb 19, 2026	Supported
85×	“85 times”	Tan: Cadence delivered an ~85x shareholder return under his leadership	Lip-Bu Tan	Jun 18, 2026	Supported
76×	“76 times”	Tan: Cadence delivered an ~85x shareholder return under his leadership	Lip-Bu Tan	Jun 18, 2026	Supported
50×	“50 times”	Zaharia: 6B parameter models can achieve instruction following with 50x less data	Matei Zaharia	Apr 25, 2023	—
30×	“30 X”	Guo: Baseten Has Grown 30x Over the Past Year	Sarah Guo	May 1, 2026	—
30×	“30 times”	Tiwari: Blackwell GPUs achieve 90 to 100x inference efficiency over Hopper	Neil Tiwari	Feb 26, 2026	Supported
20×	“20 X”	Fang: DoorDash internal AI spend jumped 20x from January to June	Andy Fang	Jul 23, 2026	—
20×	“20 X”	Feldman: Cerebras is 15 to 20 times faster than GPUs at inference	Andrew Feldman	May 21, 2026	Partly supported
20×	“20 times”	Nadler: OpenEvidence is used daily by 40% of US doctors	Daniel Nadler	Sep 5, 2025	—
10×	“10 times”	Taylor: Valar Atomics Will Scale Fission to Make Energy 10x Cheaper	Isaiah Taylor	Aug 15, 2026	—
10×	“10 X”	Tan: Setting goal of 10x shareholder return at Intel within a decade	Lip-Bu Tan	Jun 18, 2026	—
10×	“10 X”	Nadella: AI agent usage of Microsoft 365 could exceed human end users	Satya Nadella	Jun 4, 2026	—
10×	“10 times”	McDermott: Replicating a ServiceNow app with LLMs costs 10x more	Bill McDermott	Apr 17, 2026	—
10×	“10 X”	Baszucki: Software CEOs must maintain both 3X and 10X targets	David Baszucki	Feb 5, 2026	—
10×	“10 times”	Huang: OpenAI's revenue would scale 10x with 10x compute capacity	Jensen Huang	Jan 8, 2026	—
10×	“10 X”	Pereyra: Law firms will 10x in size again due to AI	Gabe Pereyra	Dec 5, 2025	—
10×	“10 times”	Liemandt: Learning science enables 2x–10x speedups, but fails in traditional classrooms	Joe Liemandt	Sep 25, 2025	—
10×	“10 times”	Liemandt: Alpha School yields top 1% academic performance in two hours daily	Joe Liemandt	Sep 25, 2025	Partly supported
10×	“10 times”	Prince: Google Clicks Are 10x Harder to Get Than 10 Years Ago	Matthew Prince	Aug 7, 2025	Supported
10×	“10 X”	Chen: Early Startups Should Not Hire Data Scientists as First Employees	Edwin Chen	Jul 24, 2025	—
10×	“10 times”	Goldman: 10x ore grade reduces rock processing and capital intensity tenfold	Josh Goldman	Apr 17, 2025	—
10×	“10 times”	Vogt: Robotaxis have no MVP without superhuman safety performance	Kyle Vogt	Feb 20, 2025	—
10×	“10 times”	Bernhardsson: Developer productivity gains historically increase the number of engineers	Erik Bernhardsson	Jan 9, 2025	—
10×	“10 X”	Davis: Democratizing AI Compute Will Increase AlphaFold-Level Breakthroughs 10x to 100x	Jared Quincy Davis	Aug 22, 2024	—
10×	“10 x”	Davis: Compound AI approach increased prime factorization accuracy from 3.7% to 36.6%	Jared Quincy Davis	Aug 22, 2024	Supported
10×	“ten-x”	Garman estimates 85% of enterprise workloads still run on-premise	Matt Garman	Aug 29, 2024	—
10×	“10 X”	Masad: Professional programmers will see 2x to 10x productivity gains from AI	Amjad Masad	Sep 21, 2023	—
10×	“10 X”	Gil: Existing models yield 10x to 100x gains without base model scaling	Elad Gil	Oct 5, 2023	—
10×	“10 X”	Kelly: Ginkgo gives 10x super-voting shares to all active employees	Jason Kelly	Sep 28, 2023	—
10×	“10 X”	A $500 million AI model will likely be trained by 2025	Noam Brown	Apr 25, 2023	Didn’t hold up
9×	“nine times”	Kushner: Affinity saw nearly 9x return on Phoenix Holdings in a year	Jared Kushner	Sep 11, 2025	Partly supported
7×	“seven times”	Liemandt: Skipping Multiplication Table Memorization Dooms Students in Advanced Math	Joe Liemandt	Sep 25, 2025	—
6×	“six times”	Clerico: Permitting conservation logging on 170-acre parcel took three years	Bill Clerico	Jan 23, 2025	—
6×	“six x”	Mansour: Kalshi grew sign-ups and volume 6x day-over-day after election launch	Tarek Mansour	Oct 31, 2024	—
5×	“five times”	Brown: Scaffolding Easily Inflates AI Benchmark Scores Without Real Gains	Noam Brown	Jun 26, 2026	—
5×	“five X”	Gil: Major engineering cost shifts from AI coding agents remain years away	Elad Gil	Jul 18, 2024	—
5×	“five x”	Scott Wu: AI will enable software engineers to do 5x more	Scott Wu	May 2, 2024	—
5×	“five X”	Masad: Professional programmers will see 2x to 10x productivity gains from AI	Amjad Masad	Sep 21, 2023	—
5×	“five times”	Parikh: SOTA text-to-audio AI currently works reasonably well one in five times	Devi Parikh	Jul 20, 2023	—
5×	“five times”	Schimpf: Hypersonic startups operate at 5x pace and fraction of Lockheed's cost	Brian Schimpf	Jun 1, 2023	—
5×	“five X”	Guo: Retraining the same size model is 5x cheaper on the second run	Sarah Guo	Apr 27, 2023	—
4×	“four X”	Dolgov: Swiss Re study showed 100% bodily injury claim reduction for Waymo	Dmitri Dolgov	Oct 24, 2024	Supported
4×	“four X”	Ma: Voyage AI Produces Embeddings 3-4x Smaller Than Competitors	Tengyu Ma	Jun 6, 2024	Supported
4×	“four times”	Mensch: Chinchilla scaling yields models four times cheaper to serve	Arthur Mensch	Nov 9, 2023	Supported
4×	“four times”	Gil: SARS leaked from laboratories four times in first two years	Elad Gil	Sep 28, 2023	Partly supported
4×	“four times”	Khan: Khan World School students are learning three to four times faster	Sal Khan	Aug 17, 2023	Supported
3×	“three X”	Roblox targets capturing 10% of global gaming content and 300M DAUs	David Baszucki	Feb 5, 2026	—
3×	“three X”	Baszucki: Software CEOs must maintain both 3X and 10X targets	David Baszucki	Feb 5, 2026	—
3×	“three times”	Mitchell: Power, water, and heat will halt 2x-3x compute scaling cycles	Ty Mitchell	Dec 12, 2024	—
3×	“three X”	Ma: Voyage AI Produces Embeddings 3-4x Smaller Than Competitors	Tengyu Ma	Jun 6, 2024	Supported
3×	“three times”	Khan: Failing to teach automatic arithmetic fluency is educational malpractice	Sal Khan	Aug 17, 2023	—
3×	“three times”	Khan: Internalized knowledge and mathematical fluency matter more in the AI era	Sal Khan	Aug 17, 2023	—
2×	“two x”	Huang: AI compute will follow hyper Moore's Law over the next decade	Jensen Huang	Nov 7, 2024	Open
2×	“two X”	Ma: Sophia Optimizer Improves LLM Pre-Training Efficiency by 2x	Tengyu Ma	Jun 6, 2024	Supported
2×	“two X”	Doshi: Curated data in fine-tuning matters more than vision algorithmic tricks	Suhail Doshi	Apr 18, 2024	—
2×	“two X”	Masad: Professional programmers will see 2x to 10x productivity gains from AI	Amjad Masad	Sep 21, 2023	—
2×	“two times”	Zipline is projected to increase its flight volume by 400% in 2023	Keller Rinaudo Cliffton	May 10, 2023	Didn’t hold up

The Numbers Museum

Turn any episode into a week of clips.