| 1K | “a thousand” | Dean: 1,000x energy penalty for moving data forces machine learning batching | Jeff Dean | Jul 30, 2026 | — |
| 1% | “one percent” | Dean: AI startups should target tasks with 1% model success, not 20% | Jeff Dean | Jul 30, 2026 | — |
| 20% | “20%” | Dean: AI startups should target tasks with 1% model success, not 20% | Jeff Dean | Jul 30, 2026 | — |
| 1K | “a thousand” | Jeff Dean: Frontier AI models consume 1,000x more data than 18-year-old humans | Jeff Dean | Jul 30, 2026 | — |
| 50% | “50%” | Stuart: GPU networking can consume up to 50% of LLaMA pre-fill runtime | Stuart (Stu) | Jul 29, 2026 | Supported |
| 80% | “80%” | Stuart: Bypassing NCCL intermediate buffers speeds up all-reduce by 80% | Stuart (Stu) | Jul 29, 2026 | Supported |
| 200M | “two hundred million” | Wang: AI tools will increase businesses on Meta to billions | Alexandr Wang | Jul 29, 2026 | Open |
| 8× | “eight X” | Tan: Meta's Muse Spark matches Claude Opus at one-eighth the cost | Garry Tan | Jul 29, 2026 | — |
| 1M | “one million” | Wang: Agentic loops using 1,000,000x more tokens are AI's biggest opportunity | Alexandr Wang | Jul 29, 2026 | — |
| 1,000× | “1000 X” | Wang: Agentic loops using 1,000,000x more tokens are AI's biggest opportunity | Alexandr Wang | Jul 29, 2026 | — |
| 6× | “six times” | Scholl: XB-1 Broke Sound Barrier Six Times With No Ground Boom | Blake Scholl | Jul 28, 2026 | Supported |
| 10% | “10%” | Tan: Hard tech startups at YC have surged to 15–25% | Garry Tan | Jul 28, 2026 | Supported |
| 25% | “25%” | Tan: Hard tech startups at YC have surged to 15–25% | Garry Tan | Jul 28, 2026 | Supported |
| 500B | “five hundred billion” | Altman: In 6.5 years, average person will use 500B tokens monthly | Sam Altman | Jul 28, 2026 | — |
| 30% | “30%” | Hu: Anthropic Opus 5 achieved 30% on ARC-AGI | Diana Hu | Jul 27, 2026 | Supported |
| 80% | “80%” | Anthropic Deleted 80% of Claude Code System Prompt for Opus 5 | Boris Cherny | Jul 27, 2026 | Supported |
| $300M | “three hundred million dollars” | Huang: NVIDIA went public in 1999 at a $300M valuation | Jensen Huang | Jul 26, 2026 | Supported |
| 5T | “five trillion” | Huang: 'Founder mode' can scale a company to five trillion dollars | Jensen Huang | Jul 26, 2026 | — |
| 1K | “a thousand” | Huang: NVIDIA engineers use autonomous coding agents like Claude Code and Cursor | Jensen Huang | Jul 26, 2026 | — |
| 20% | “20%” | Huang: Radiology jobs grew 20% over recent years despite AI scan automation | Jensen Huang | Jul 26, 2026 | Supported |
| $10B | “ten billion dollars” | Huang: NVIDIA physical AI and automotive business is near $10 billion | Jensen Huang | Jul 26, 2026 | — |
| 100B | “hundred billion” | Huang: Physical AI will be NVIDIA's next $100B business within 10 years | Jensen Huang | Jul 26, 2026 | Open |
| 65% | “65%” | Pomel: Major Datadog Investor Dumped All Shares When COVID Lockdowns Began | Olivier Pomel | Jul 24, 2026 | Partly supported |
| 100B | “a hundred billion” | Garry Tan: Context-aware consumer AI will become a $10B-$100B company | Garry Tan | Jun 25, 2026 | — |
| $1M | “a million dollars” | Garry Tan discloses spending $1M annually on AI tokens | Garry Tan | Jun 25, 2026 | — |
| 1M | “a million” | Tan: Peter Steinberger spends up to $1.1M monthly on AI tokens | Garry Tan | Jun 25, 2026 | — |
| $1.1M | “1.1 million dollars” | Tan: Peter Steinberger spends up to $1.1M monthly on AI tokens | Garry Tan | Jun 25, 2026 | — |
| 20× | “20 times” | Tan: Write markdown teaching LLMs to code rather than calling LLMs | Garry Tan | Jun 25, 2026 | — |
| 1K | “a thousand” | Tan: AI token compute costs will drop drastically within two years | Garry Tan | Jun 25, 2026 | Open |
| 1% | “one percent” | Tan: Webflow powers 1% of all websites on the internet | Garry Tan | Jun 19, 2026 | Partly supported |
| 12% | “12%” | Chou: About 12% of the current YC batch uses Ploy | Bryant Chou | Jun 19, 2026 | — |
| 5M | “five million” | Tan: AI puts me on track to write 5M lines of code | Garry Tan | Jun 19, 2026 | — |
| 98% | “98%” | Brex's AI Recruiting Agent Auto-Passes 98% of Security Checks | Pedro Franceschi | Jun 10, 2026 | — |
| 2% | “two percent” | Brex's AI Recruiting Agent Auto-Passes 98% of Security Checks | Pedro Franceschi | Jun 10, 2026 | — |
| 1% | “1%” | Franceschi: True AI Adoption Is Low Because Few Hit Max Plan Limits | Pedro Franceschi | Jun 10, 2026 | — |
| 10× | “10 X” | Franceschi: CEOs Break Corporate Process 10x Easier Than Executives | Pedro Franceschi | Jun 10, 2026 | — |
| 80% | “80%” | Franceschi: Founders Should Build Custom AI Solutions for Unsolved Daily Problems | Pedro Franceschi | Jun 10, 2026 | — |
| 20% | “20%” | Franceschi: Founders Should Build Custom AI Solutions for Unsolved Daily Problems | Pedro Franceschi | Jun 10, 2026 | — |
| $1M | “a million dollars” | Tan: $100K AI token workloads will drop to $200 in two years | Garry Tan | May 27, 2026 | Open |
| $4B | “four billion dollars” | Tan: Jeff Lawson was ousted from Twilio 199 days after voting protections expired | Garry Tan | May 22, 2026 | Supported |
| 390% | “390%” | Tan: Jeff Lawson was ousted from Twilio 199 days after voting protections expired | Garry Tan | May 22, 2026 | Supported |
| 1% | “a percent” | Tan: Jeff Lawson was ousted from Twilio 199 days after voting protections expired | Garry Tan | May 22, 2026 | Supported |
| 80% | “80%” | Ries: Twilio fired Jeff Lawson despite revenue growth since IPO | Eric Ries | May 22, 2026 | Supported |
| $8B | “eight billion dollars” | Ries: Philip Morris generates $600B in annual US external costs | Eric Ries | May 22, 2026 | Supported |
| $600B | “six hundred billion dollars” | Ries: Philip Morris generates $600B in annual US external costs | Eric Ries | May 22, 2026 | Supported |
| 300B | “three hundred billion” | Ries: Philip Morris generates $600B in annual US external costs | Eric Ries | May 22, 2026 | Supported |
| $300B | “three hundred billion dollars” | Ries: Philip Morris generates $600B in annual US external costs | Eric Ries | May 22, 2026 | Supported |
| 10% | “10%” | Ries: Foundation-owned companies are 6x more likely to reach year 50 | Eric Ries | May 22, 2026 | Partly supported |
| 60% | “60%” | Ries: Foundation-owned companies are 6x more likely to reach year 50 | Eric Ries | May 22, 2026 | Partly supported |
| 6× | “six times” | Ries: Foundation-owned companies are 6x more likely to reach year 50 | Eric Ries | May 22, 2026 | Partly supported |
| 400× | “400 X” | Tan: Returning to Coding with AI Yields 400x Output | Garry Tan | May 8, 2026 | — |
| $20M | “twenty million dollars” | Garry Tan: Twitter acquired Posterous for approximately $20 million | Garry Tan | May 8, 2026 | Supported |
| $4M | “four million dollars” | Tan: Building Garry's List with Claude Code took $200 and five days | Garry Tan | May 8, 2026 | — |
| 100% | “hundred percent” | Garry Tan: 80% to 90% test coverage is optimal for AI coding | Garry Tan | May 8, 2026 | — |
| 90% | “90%” | Garry Tan: 80% to 90% test coverage is optimal for AI coding | Garry Tan | May 8, 2026 | — |
| 10× | “10 X” | Tan: Asking for 10x value at 2x effort unlocks ambitious products | Garry Tan | May 8, 2026 | — |
| 2× | “two X” | Tan: Asking for 10x value at 2x effort unlocks ambitious products | Garry Tan | May 8, 2026 | — |
| 10× | “10 X” | Tan: Untested AI-generated code is 10x worse than human-written code | Garry Tan | May 8, 2026 | — |
| 70% | “70%” | Chaubard: HRM scored 70% on ARC Prize 1 without pre-training | Francois Chaubard | May 1, 2026 | Partly supported |
| 1K | “a thousand” | Gupta: Recursive architectures achieve compute depth without parameter depth | Ankit Gupta | May 1, 2026 | — |
| 1M | “a million” | Gupta: Recursive architectures achieve compute depth without parameter depth | Ankit Gupta | May 1, 2026 | — |
| 29M | “a twenty-eight million” | Chaubard: 7M parameter TRM scored 87% on ARC Prize 1 | Francois Chaubard | May 1, 2026 | Contradicted |
| 8M | “a seven million” | Chaubard: 7M parameter TRM scored 87% on ARC Prize 1 | Francois Chaubard | May 1, 2026 | Contradicted |
| 70% | “70%” | Chaubard: 7M parameter TRM scored 87% on ARC Prize 1 | Francois Chaubard | May 1, 2026 | Contradicted |
| 87% | “87%” | Chaubard: 7M parameter TRM scored 87% on ARC Prize 1 | Francois Chaubard | May 1, 2026 | Contradicted |
| 1M | “a million” | Hassabis: A million-token context window holds only 20 minutes of live video | Demis Hassabis | Apr 29, 2026 | Supported |
| 70% | “70%” | Massad: Replit-native agencies are 60% to 70% cheaper than traditional firms | Amjad Masad | Apr 25, 2026 | — |
| 7% | “seven percent” | Massad: Seven percent week-over-week growth is ideal for bootstrapping product lines | Amjad Masad | Apr 25, 2026 | — |
| 4× | “four times” | Massad: Replit was rejected from YC multiple times before Graham and Altman invited them | Amjad Masad | Apr 25, 2026 | — |
| $10M | “ten million dollars” | Tan: Rebuilt Posterous with AI, Replacing 10 Engineers and $10M | Garry Tan | Apr 23, 2026 | — |
| 78% | “78%” | Dill: Over 78% of the Forbes AI 50 use Stripe | Katie Dill | Apr 22, 2026 | Supported |
| 50% | “50%” | Vuong: OpenX generalist robot model outperformed embodiment-specific specialists by 50% | Quan Vuong | Apr 16, 2026 | Supported |
| 10% | “10%” | Vuong: Solving general robotics could add 10% to the US GDP | Quan Vuong | Apr 16, 2026 | — |
| 1K | “A thousand” | Vuong: Absorbing diverse data is easier than scaling proprietary hardware | Quan Vuong | Apr 16, 2026 | — |
| 10× | “10 times” | Vuong: Robotics evaluation difficulty scales superlinearly with task duration | Quan Vuong | Apr 16, 2026 | — |
| 50% | “50%” | Vuong: Claude-based on-call agent boosted compute utilization by 50% | Quan Vuong | Apr 16, 2026 | — |
| $4B | “four billion dollars” | Friedman: BillionToOne Went Public at Over $4 Billion Valuation | Jared Friedman | Apr 6, 2026 | Partly supported |
| 20% | “20%” | Bilgen: BillionToOne Processes Over 600,000 Tests Annually With ~20% Market Share | Ozan Bilgen | Apr 6, 2026 | Supported |
| 2M | “two million” | Bilgen: Current facility can scale to nearly two million tests annually | Ozan Bilgen | Apr 6, 2026 | Open |
| 20% | “20%” | Bilgen: 20% of early-stage cancer surgery patients retain undetectable microscopic residue | Ozan Bilgen | Apr 6, 2026 | Supported |
| $1B | “a billion dollars” | Bilgen: Direct early cancer detection requires raising over $1B pre-revenue | Ozan Bilgen | Apr 6, 2026 | — |
| 21M | “twenty-one million” | Friedman: Variance emerges from stealth with $21M Series A | Jared Friedman | Mar 31, 2026 | Open |
| 99% | “99%” | Mellata: AI agent triage leaves a complex 1% requiring investigative dashboards | Karine Malata | Mar 31, 2026 | — |
| 1% | “one percent” | Mellata: AI agent triage leaves a complex 1% requiring investigative dashboards | Karine Malata | Mar 31, 2026 | — |
| 10× | “10 x” | Mellata: OpenAI model upgrades cut Variance's pilot costs tenfold | Karine Malata | Mar 31, 2026 | — |
| 15% | “15%” | Chollet: NDEA Has Only a 10% to 15% Chance of Success | François Chollet | Mar 27, 2026 | — |
| 10% | “ten percent” | Chollet: Base LLMs Score Under 10% on ARC-AGI-1 | François Chollet | Mar 27, 2026 | Supported |
| 97% | “97%” | Hu: Confluence Labs Saturated the ARC-2 Benchmark at 97% Accuracy | Diana Hu | Mar 27, 2026 | Supported |
| 80% | “80%” | Jha: 80% of Emergent users have zero programming knowledge | Mukund Jha | Mar 16, 2026 | — |
| 20% | “20%” | Mukund Jha: Coding is only 20% of shipping software to production | Mukund Jha | Mar 16, 2026 | — |
| 30% | “30%” | Emergent extracts 20% to 30% more performance from models using custom harness | Mukund Jha | Mar 16, 2026 | — |
| $3M | “3004 thousand dollars” | Madhav Jha: Emergent saves $3,000–$4,000 monthly replacing Asana internally | Madhav Jha | Mar 16, 2026 | — |
| 20% | “20%” | Mukund Jha: Roughly 20% of apps built on Emergent are agentic | Mukund Jha | Mar 16, 2026 | — |
| $4M | “four million dollars” | Mukund Jha: A company raised $4M for an app built on Emergent | Mukund Jha | Mar 16, 2026 | — |
| 1K | “a thousand” | Hodak: Humans who will live to 1,000 are likely alive today | Max Hodak | Mar 9, 2026 | — |
| 50% | “50%” | Hodak: Personal probability of AI doom is well below 50 percent | Max Hodak | Mar 9, 2026 | — |
| 10× | “10 x” | Epstein: Blanket '10x' startup landing page metrics feel fabricated | Aaron Epstein | Mar 6, 2026 | — |
| 45% | “45%” | Poetiq outperformed Gemini 3 Deep Think on ARC-AGI-2 at half the cost | Ian Fisher | Feb 27, 2026 | Supported |
| 54% | “54%” | Poetiq outperformed Gemini 3 Deep Think on ARC-AGI-2 at half the cost | Ian Fisher | Feb 27, 2026 | Supported |
| 55% | “55%” | Poetiq scored 55% on Humanity's Last Exam, outperforming Claude Opus 4.6 | Ian Fisher | Feb 27, 2026 | Supported |
| 53.1% | “53.1%” | Poetiq scored 55% on Humanity's Last Exam, outperforming Claude Opus 4.6 | Ian Fisher | Feb 27, 2026 | Supported |
| 5% | “five percent” | Code-based reasoning strategies boosted Gemini 1.5 Flash performance from 5% to 95% | Ian Fisher | Feb 27, 2026 | Not publicly verifiable |
| 95% | “95%” | Code-based reasoning strategies boosted Gemini 1.5 Flash performance from 5% to 95% | Ian Fisher | Feb 27, 2026 | Not publicly verifiable |
| 200× | “200 times” | Tan: Groq runs transcription 200 times faster than Whisper v1 | Garry Tan | Feb 21, 2026 | Contradicted |
| 10× | “10 X” | Tan: Groq is 10x cheaper than legacy Whisper for transcription | Garry Tan | Feb 21, 2026 | Partly supported |
| 10% | “10%” | Cherny: Claude Code wrote only 10 percent of his code at launch | Boris Cherny | Feb 17, 2026 | — |
| 20% | “20%” | Scaffolding gains get wiped out by next-generation AI models | Boris Cherny | Feb 17, 2026 | — |
| 150% | “150%” | Claude Code increased Anthropic engineer productivity by 150 percent | Boris Cherny | Feb 17, 2026 | — |
| 100% | “hundred percent” | Cherny uninstalled his IDE and lands 20 AI-written PRs daily | Boris Cherny | Feb 17, 2026 | — |
| 90% | “90%” | Claude generates between 70 and 90 percent of Anthropic's codebase | Boris Cherny | Feb 17, 2026 | — |
| 100% | “hundred percent” | Claude generates between 70 and 90 percent of Anthropic's codebase | Boris Cherny | Feb 17, 2026 | — |
| 70% | “70%” | Mercury data shows 70 percent of startups choose Claude | Boris Cherny | Feb 17, 2026 | Contradicted |
| 4% | “four percent” | SemiAnalysis reports Claude Code generates four percent of all public commits | Boris Cherny | Feb 17, 2026 | Supported |
| 100% | “hundred percent” | Anthropic built the Claude desktop app in 10 days using AI | Boris Cherny | Feb 17, 2026 | — |
| 80% | “80%” | Steinberger: 80% of consumer apps will disappear due to AI agents | Peter Steinberger | Feb 7, 2026 | — |
| 5× | “five times” | Garry Tan: Claude Code is an unlock that enables 5x faster coding | Garry Tan | Feb 6, 2026 | — |
| 10× | “10 times” | Taggar: Top 18-to-22-Year-Olds in 5 Years Will Have 'Off the Charts' Taste | Harj Taggar | Feb 6, 2026 | — |
| 100% | “hundred percent” | Garry Tan: Reaching 100% Test Coverage Supercharged Coding Velocity with Claude Code | Garry Tan | Feb 6, 2026 | — |
| $990M | “nine hundred ninety million dollars” | Lapsa: Stoke Space has raised about $990M to date | Andy Lapsa | Jan 8, 2026 | — |
| 52% | “52%” | Hu: Anthropic saw over 52% growth among YC applicant tech stacks | Diana Hu | Dec 22, 2025 | — |
| 3% | “three percent” | Hu: Google Gemini adoption in YC batch rose to around 23% | Diana Hu | Dec 22, 2025 | — |
| 23% | “23%” | Hu: Google Gemini adoption in YC batch rose to around 23% | Diana Hu | Dec 22, 2025 | — |
| 8B | “eight billion” | Hu: A YC startup's 8B model beat OpenAI on healthcare benchmarks | Diana Hu | Dec 22, 2025 | — |
| 100% | “hundred percent” | Tan: Vibe coding cannot yet ship 100% solid production code in 2025 | Garry Tan | Dec 22, 2025 | — |
| 90% | “90%” | Tan: Enterprise incompetence slows down dangerous AI fast takeoff | Garry Tan | Dec 22, 2025 | — |
| $100M | “a hundred million dollars” | Tan: Gamma reached $100M ARR with only 50 employees | Garry Tan | Dec 22, 2025 | Supported |
| 76M | “a seventy-five million” | PostHog raises a $75M Series E at a $1.4B valuation | Brad Flora | Dec 10, 2025 | Supported |
| 2.4B | “a 1.4 billion” | PostHog raises a $75M Series E at a $1.4B valuation | Brad Flora | Dec 10, 2025 | Supported |
| 1K | “a thousand” | Funny content generates 1,000x more reach than useful technical posts | James Hawkins | Dec 10, 2025 | — |
| 80% | “80%” | Skates: The hype around AI killing SaaS is way overblown | Spencer Skates | Dec 3, 2025 | — |
| 50M | “fifty million” | Skates: Amplitude generates $350M in annual revenue and spends nearly that much | Spencer Skates | Dec 3, 2025 | Supported |
| 1K | “a thousand” | Tan: Micro-creators with 1,000 to 10,000 followers are mispriced distribution assets | Garry Tan | Nov 28, 2025 | — |
| 40% | “40%” | Paffenholz: A 40% response rate requires five minutes of personalization per candidate | David Paffenholz | Nov 7, 2025 | — |
| 30% | “30%” | Paffenholz: Job descriptions should be at least 30% selling the company | David Paffenholz | Nov 7, 2025 | — |
| $20M | “twenty million dollars” | Heller: Casetext pivoted with $20M revenue and 100 employees for GPT-4 | Jake Heller | Oct 28, 2025 | — |
| 1K | “a thousand” | Heller: AI application TAM is 1,000x larger when capturing labor budgets | Jake Heller | Oct 28, 2025 | — |
| 40% | “40%” | Heller: 30% to 40% of Casetext Employees, Including Coders, Were Lawyers | Jake Heller | Oct 28, 2025 | — |
| 90% | “90%” | Heller: Most AI app developers fail to run evaluations or study workflows | Jake Heller | Oct 28, 2025 | — |
| $1T | “a trillion dollars” | Heller: Targeting legal tech pre-LLMs was a mistake due to low software spend | Jake Heller | Oct 28, 2025 | — |
| 30% | “30%” | Airbnb maintained a minimum technical staff ratio to prevent stalled productivity | Gustaf Alströmer | Oct 21, 2025 | — |
| 1B | “a billion” | Multiple Y Combinator startups aim for billion-dollar valuations with ten employees | Gustaf Alströmer | Oct 21, 2025 | — |
| $5M | “five million dollars” | Joseph: Public estimates placed GPT-3 training cost at $5 million | Nick Joseph | Sep 30, 2025 | Supported |
| 5M | “five million” | Joseph: Public estimates placed GPT-3 training cost at $5 million | Nick Joseph | Sep 30, 2025 | Supported |
| 5% | “five percent” | Joseph: LLM training requires collaborative infrastructure work over publishable research papers | Nick Joseph | Sep 30, 2025 | — |
| 10% | “10%” | Joseph: Sparsely linked long-tail data may be most valuable for frontier AI | Nick Joseph | Sep 30, 2025 | — |
| 1B | “a billion” | Joseph: Sci-fi underestimates AGI because AI software can be copied a billion times | Nick Joseph | Sep 30, 2025 | — |
| 1K | “a thousand” | Pollak: Base Compresses Transactions to Ethereum to Slash Costs 1,000% | Jesse Pollak | Sep 23, 2025 | Supported |
| 10× | “10 X” | Pollak: Four maturing unlocks have created crypto's golden building era | Jesse Pollak | Sep 23, 2025 | — |
| $1B | “a billion dollars” | Pollak: USDC took several years to reach $1 billion in issuance | Jesse Pollak | Sep 23, 2025 | — |
| $200B | “two hundred billion dollars” | Pollak: Global stablecoin supply is near $200 billion | Jesse Pollak | Sep 23, 2025 | Contradicted |
| 1K | “a thousand” | Pollak: Coinbase and Shopify condensed payment systems into 1,000-line smart contract | Jesse Pollak | Sep 23, 2025 | — |
| 20B | “twenty billion” | Levie: 100 Startups Created Between 2024-2028 Will Reach $5B to $20B Valuations | Aaron Levie | Sep 16, 2025 | Open |
| 10× | “10 times” | Levie: B2B founders must master three essential strategy books | Aaron Levie | Sep 16, 2025 | — |
| 1K | “a thousand” | Masad: Developers will reliably run 1,000 concurrent AI agents within two years | Amjad Masad | Sep 12, 2025 | Open |
| 95% | “95%” | Masad: Developers will reliably run 1,000 concurrent AI agents within two years | Amjad Masad | Sep 12, 2025 | Open |
| 1M | “one million” | Truell confirms Cursor reached roughly $1M in revenue around 2023 | Michael Truell | Sep 3, 2025 | Supported |
| 80% | “80%” | Hu: Cursor usage in YC batches jumped from single digits to 80% | Diana Hu | Sep 3, 2025 | — |
| $80M | “eighty million dollars” | Legora raises $80M Series B led by Iconiq and General Catalyst | Max Junestrand | Aug 26, 2025 | — |
| 99% | “99%” | Legal due diligence commoditized as clients refuse paying for manual review | Max Junestrand | Aug 26, 2025 | — |
| 100× | “hundred times” | Legora's 30 engineers outship legacy legal tech incumbents with thousands of developers | Max Junestrand | Aug 26, 2025 | — |
| 90% | “90%” | Blomfield: 90% of B2B Founders Get Stuck in Unpaid Design Partnerships | Tom Blomfield | Aug 15, 2025 | — |
| 5% | “five percent” | Field: 5% of Figma Design files were slides before dedicated product launched | Dylan Field | Aug 8, 2025 | — |
| 10% | “10%” | Englander: Ten percent of top global financial firms use Model ML | Chas Englander | Jul 31, 2025 | — |
| 95% | “95%” | Englander: Model ML auto-generates near-complete earnings summaries from new releases | Chas Englander | Jul 31, 2025 | — |
| 3× | “three X” | Kaplan: AI training and inference will see 3x-10x yearly efficiency gains | Jared Kaplan | Jul 29, 2025 | Held up |
| 10× | “10 X” | Kaplan: AI training and inference will see 3x-10x yearly efficiency gains | Jared Kaplan | Jul 29, 2025 | Held up |
| 4B | “a three billion” | Finn: Physical Intelligence uses PaliGemma 3B with flow matching diffusion head | Chelsea Finn | Jul 22, 2025 | — |
| 300M | “three hundred million” | Finn: Physical Intelligence uses PaliGemma 3B with flow matching diffusion head | Chelsea Finn | Jul 22, 2025 | — |
| 2.4% | “2.4%” | Finn: Bedroom and kitchen tidying data was only 2.4% of pre-training mix | Chelsea Finn | Jul 22, 2025 | — |
| 80% | “80%” | Finn: Architectural fix boosted robot language following rate from 20% to 80% | Chelsea Finn | Jul 22, 2025 | Supported |
| 20% | “20%” | Finn: Architectural fix boosted robot language following rate from 20% to 80% | Chelsea Finn | Jul 22, 2025 | Supported |
| 60% | “60%” | Finn: Full Pre-Training Mixture Boosts Robot Performance Over 20% in Novel Homes | Chelsea Finn | Jul 22, 2025 | Supported |
| 20% | “20%” | Finn: Full Pre-Training Mixture Boosts Robot Performance Over 20% in Novel Homes | Chelsea Finn | Jul 22, 2025 | Supported |
| 3,000× | “3000 times” | Jumper: Protein sequence discovery outpaces structure determination by 3,000 times | John Jumper | Jul 15, 2025 | Supported |
| 1% | “one percent” | Jumper: AlphaFold 2 architecture on 1% of data matched AlphaFold 1 | John Jumper | Jul 15, 2025 | Supported |
| 200M | “two hundred million” | Jumper: AlphaFold database scaled from 300,000 to 200 million protein predictions | John Jumper | Jul 15, 2025 | Supported |
| 10% | “10%” | Jumper: AlphaFold made structural biology 5% to 10% faster | John Jumper | Jul 15, 2025 | — |
| 50B | “fifty billion” | Srinivas: Foundation model labs will copy any high-revenue AI startup idea | Aravind Srinivas | Jul 11, 2025 | — |
| 50% | “50%” | Ng: AI coding makes prototyping 10x faster, production only 50% faster | Andrew Ng | Jul 10, 2025 | — |
| 10× | “10 times” | Ng: AI coding makes prototyping 10x faster, production only 50% faster | Andrew Ng | Jul 10, 2025 | — |
| 3× | “three times” | Ng's teams rebuilt a codebase three times in one month using AI | Andrew Ng | Jul 10, 2025 | — |
| 10× | “10 times” | Ng: Bad AI architecture decisions slow startups down by 10x, not 2x | Andrew Ng | Jul 10, 2025 | — |
| 6.1% | “6.1%” | Garry Tan: CS majors face higher unemployment than art history majors | Garry Tan | Jul 8, 2025 | Supported |
| 12M | “twelve million” | Hu: Cursor scaled from $1M to $100M revenue in one year | Diana Hu | Jul 8, 2025 | Supported |
| 10% | “10%” | Chollet: Scaling LLMs 50,000x Only Lifted ARC Accuracy to 10% | François Chollet | Jul 3, 2025 | Supported |
| 100% | “100%” | Chollet: Ten random people with majority voting score 100% on ARC-2 | François Chollet | Jul 3, 2025 | Open |
| 2% | “two percent” | Chollet: Base LLMs score 0% and static reasoning scores 1-2% on ARC-2 | François Chollet | Jul 3, 2025 | Supported |
| 100% | “hundred percent” | Fei-Fei Li: AI capabilities have completely outrun theoretical understanding | Dr. Fei-Fei Li | Jul 1, 2025 | — |
| 4% | “four percent” | Green: Major CPG firm saw word-of-mouth sales jump from 4% to 48% | Kirsten Green | Jun 27, 2025 | — |
| 48% | “48%” | Green: Major CPG firm saw word-of-mouth sales jump from 4% to 48% | Kirsten Green | Jun 27, 2025 | — |
| 5× | “five times” | Altman: OpenAI's o3 model cost dropped by 5x in one week | Sam Altman | Jun 21, 2025 | Supported |
| 10× | “10 times” | Musk: AI could expand the economy thousands or millions of times | Elon Musk | Jun 19, 2025 | — |
| 1K | “a thousand” | Musk: Government efficiency work is a side quest compared to AI | Elon Musk | Jun 19, 2025 | — |
| 2% | “two percent” | Elon Musk: Raw materials are only 1% to 2% of rocket costs | Elon Musk | Jun 19, 2025 | Supported |
| 5× | “five times” | Musk: Humanoid robots will outnumber humans five to tenfold | Elon Musk | Jun 19, 2025 | Open |
| 10× | “10 times” | Musk: Humanoid robots will outnumber humans five to tenfold | Elon Musk | Jun 19, 2025 | Open |
| 20% | “20%” | Musk: There is a 10% to 20% chance AI causes annihilation | Elon Musk | Jun 19, 2025 | — |
| 90% | “90%” | Musk: There is a 10% to 20% chance AI causes annihilation | Elon Musk | Jun 19, 2025 | — |
| 1K | “a thousand” | Musk: Augmented Human Intelligence Will Probably Be One Billionth of Digital Intelligence | Elon Musk | Jun 19, 2025 | — |
| 10B | “a nine billion” | Tan: Cursor Reached $100M ARR in 20 Months at $9B Valuation | Garry Tan | Jun 11, 2025 | Supported |
Y COMBINATORLEDGER All shows Try the podcast clipper →