| $6M | “six million dollars” | Swyx says he spent six million dollars on AV this year | Shawn Wang | Sep 7, 2026 | — |
| 10% | “10%” | Slack says hiring early PMs or marketers is an outdated practice | Quinn Slack | Sep 7, 2026 | — |
| 90% | “90%” | Slack says hiring early PMs or marketers is an outdated practice | Quinn Slack | Sep 7, 2026 | — |
| 5T | “five trillion” | Jenik: A single 5-trillion-token model run generated 22 terabytes of output | Benedikt Jenik | Sep 4, 2026 | — |
| 1T | “a trillion” | Jenik: Accelerated Understanding has trained physics models up to 1-trillion parameters | Benedikt Jenik | Sep 4, 2026 | — |
| 1T | “a trillion” | Jenik: Accelerated Understanding achieved 5-trillion token inference context length | Benedikt Jenik | Sep 4, 2026 | — |
| 5T | “five trillion” | Jenik: Accelerated Understanding achieved 5-trillion token inference context length | Benedikt Jenik | Sep 4, 2026 | — |
| 6T | “a five trillion” | Anandkumar: Standard Transformers cannot scale to 5-trillion context lengths for physics | Anima Anandkumar | Sep 4, 2026 | — |
| 100% | “hundred percent” | Sean Lie: 95% of High-Quality Open Models Come From Chinese Labs | Sean Lie | Sep 2, 2026 | — |
| 95% | “95%” | Sean Lie: 95% of High-Quality Open Models Come From Chinese Labs | Sean Lie | Sep 2, 2026 | — |
| 2× | “two X” | Sean Lie: Next Cerebras Chip Will Run Frontier AI at 5,000 TPS | Sean Lie | Sep 2, 2026 | Open |
| 100× | “hundred times” | Lie: Cerebras chips have 100x more memory than Groq LPUs | Sean Lie | Sep 2, 2026 | Supported |
| 14× | “14 times” | Lie: Cerebras runs OpenAI's flagship model 14x faster than GPUs | Sean Lie | Sep 2, 2026 | Supported |
| 1M | “a million” | AI models predict fusion reactor plasma disruption one million times faster | Anima Anandkumar | Aug 26, 2026 | Supported |
| 1K | “a thousand” | Park: Advanced simulation discovers multi-step paths to achieve a target outcome | Joon Sung Park | Aug 21, 2026 | — |
| 85% | “85%” | Park: Generative agent digital twins replicate human behavior at 85% accuracy | Joon Sung Park | Aug 21, 2026 | Supported |
| 30% | “30%” | Park: Frontier models hit only 20-30% accuracy predicting niche human behavior | Joon Sung Park | Aug 21, 2026 | — |
| 60% | “60%” | Park: Frontier models hit only 20-30% accuracy predicting niche human behavior | Joon Sung Park | Aug 21, 2026 | — |
| $40M | “forty million dollars” | Park: Agent simulations beat multi-year field trials via instant repeated execution | Joon Sung Park | Aug 21, 2026 | — |
| 15% | “15%” | Park: Simile AI has ~60 employees, 15-20% from Stanford lab | Joon Sung Park | Aug 21, 2026 | — |
| 20% | “20%” | Park: Simile AI has ~60 employees, 15-20% from Stanford lab | Joon Sung Park | Aug 21, 2026 | — |
| 96% | “96%” | Krentsel: Exo autonomously re-architected its Discord adapter, cutting costs by 96% | Alex Krentsel | Aug 15, 2026 | Open |
| 20% | “20%” | McPartlon: Chai-2 generated binders for 25 targets with 20% hit rate | Matt McPartlon | Aug 11, 2026 | Supported |
| 11% | “11%” | McPartlon: AlphaFold-Multimer gets antibody-antigen prediction right only 11% of the time | Matt McPartlon | Aug 11, 2026 | Supported |
| 90% | “90%” | McPartlon: AlphaFold-Multimer gets antibody-antigen prediction right only 11% of the time | Matt McPartlon | Aug 11, 2026 | Supported |
| 1B | “a billion” | McPartlon: Multi-specific antibody modalities must be designed computationally from first principles | Matt McPartlon | Aug 11, 2026 | — |
| 3B | “a one a billion” | McPartlon: Multi-specific antibody modalities must be designed computationally from first principles | Matt McPartlon | Aug 11, 2026 | — |
| 56% | “56%” | Baseten's vision-retrofitted GLM-5.2 scored 56% on MMLU Pro without text degradation | Philip Kiely | Aug 3, 2026 | Partly supported |
| 4× | “four times” | Baseten terminates or reprocesses generations after four identical repeating tokens | Ali Taha | Aug 3, 2026 | — |
| 1% | “one percent” | Inference optimization is only solved once researchers report mere 1% speedups | Philip Kiely | Aug 3, 2026 | — |
| 10× | “10 X” | Unoptimized GLM-5.2 delivers a baseline 30 to 40 tokens per second | Philip Kiely | Aug 3, 2026 | — |
| 4× | “four X” | Software optimizations yield 2x to 4x inference speedups on identical hardware | Philip Kiely | Aug 3, 2026 | — |
| 40% | “40%” | Quantization, speculative decoding, and prefill-decode disaggregation each yield 2x speedups | Philip Kiely | Aug 3, 2026 | — |
| 2× | “two X” | Quantization, speculative decoding, and prefill-decode disaggregation each yield 2x speedups | Philip Kiely | Aug 3, 2026 | — |
| 1T | “a trillion” | Custom $50B ASICs become economically rational for $500B AI training runs | Shawn Wang | Aug 3, 2026 | — |
| 500B | “five hundred billion” | Custom $50B ASICs become economically rational for $500B AI training runs | Shawn Wang | Aug 3, 2026 | — |
| 50B | “fifty billion” | Custom $50B ASICs become economically rational for $500B AI training runs | Shawn Wang | Aug 3, 2026 | — |
| 10% | “10%” | Custom $50B ASICs become economically rational for $500B AI training runs | Shawn Wang | Aug 3, 2026 | — |
| 2.8T | “2.8 trillion” | Fitting a 2.8-trillion parameter model on one node requires eight GB300s | Philip Kiely | Aug 3, 2026 | Supported |
| 100× | “hundred X” | Ultra-fast network interface cards could deliver 100x speedups in disaggregated AI inference | Ali Taha | Aug 3, 2026 | — |
| 10× | “10 X” | Nathan: Untapped enterprise agent market is 10x to 100x larger | Akshay Nathan | Jul 28, 2026 | — |
| 100× | “hundred X” | Nathan: Untapped enterprise agent market is 10x to 100x larger | Akshay Nathan | Jul 28, 2026 | — |
| 1.7B | “1.7 billion” | Swyx: Generated Playable 3D Board Game in ChatGPT Using 1.7B Tokens | Shawn Wang | Jul 28, 2026 | — |
| $12M | “twelve million dollars” | Kant: Sourced blew $12M of investor capital on code models before failing | Eiso Kant | Jul 22, 2026 | Partly supported |
| 90% | “90%” | Kant: Foundation model building is 90% engineering rather than research | Eiso Kant | Jul 22, 2026 | — |
| 5% | “Five percent” | Kant: 95% of foundation model building is data and compute efficiency | Eiso Kant | Jul 22, 2026 | — |
| 8B | “eight billion” | Kant: Laguna S has 118B total parameters and 8B active parameters | Eiso Kant | Jul 22, 2026 | — |
| 3× | “three times” | Kant: Laguna S outperforms models two to three times its size | Eiso Kant | Jul 22, 2026 | Supported |
| 1T | “a trillion” | Kant: Nvidia GB300 allows models to jump from 1T to 6T parameters | Eiso Kant | Jul 22, 2026 | — |
| 6T | “six trillion” | Kant: Nvidia GB300 allows models to jump from 1T to 6T parameters | Eiso Kant | Jul 22, 2026 | — |
| 80% | “80%” | Beam: Lila's AI hits 80% zero-shot on gene editing, beating humans' 0% | Andy Beam | Jul 16, 2026 | — |
| 10T | “10 trillion” | Beam: Lila's 10-trillion-token general science model beats specialized AI | Andy Beam | Jul 16, 2026 | — |
| 2% | “two percent” | Beam: CAR-T Therapy Likely Would Have Failed Without Serendipitous Intervention | Andy Beam | Jul 16, 2026 | — |
| 98% | “98%” | Beam: CAR-T Therapy Likely Would Have Failed Without Serendipitous Intervention | Andy Beam | Jul 16, 2026 | — |
| 2,500× | “2500 times” | Gomez-Bombarelli: Lila tests 96 MOFs in an hour using proxy measurements | Rafa Gomez-Bombarelli | Jul 16, 2026 | — |
| $1B | “a billion dollars” | Beam: Lila avoids pre-training from scratch, builds on open-weight models | Andy Beam | Jul 16, 2026 | — |
| 1K | “a thousand” | Beam: Lila will open source part of its 1,000 scientific RL environments | Andy Beam | Jul 16, 2026 | — |
| 6% | “six percent” | Beam: Reinforcement learning achieves only 5% to 6% GPU FLOP utilization | Andy Beam | Jul 16, 2026 | Supported |
| 5% | “five percent” | Beam: Reinforcement learning achieves only 5% to 6% GPU FLOP utilization | Andy Beam | Jul 16, 2026 | Supported |
| 10M | “ten million” | Biderman: Model accuracy will still degrade at 10M context window scale | Dan Biderman | Jul 13, 2026 | — |
| 5% | “Five percent” | Swyx: AI p(doom) over the next 50 years is five percent | Shawn Wang | Jul 10, 2026 | — |
| 65% | “65%” | Feinberg: Small molecules make up 65% of FDA-approved drugs | Evan Feinberg | Jun 30, 2026 | Supported |
| 99% | “99%” | Xin: Unifying storage delivers 99% of HTAP database benefits | Reynold Xin | Jun 24, 2026 | — |
| 100× | “hundred X” | Zaharia: Databricks' document vision model is ~100x cheaper than frontier models | Matei Zaharia | Jun 24, 2026 | — |
| 70% | “70%” | Fredrikson: Skilled red teamers phish human participants 60% to 70% of the time | Matt Fredrikson | Jun 22, 2026 | Open |
| 20× | “20 times” | Fredrikson: Formal software verification takes 10 to 20 times longer than Python | Matt Fredrikson | Jun 22, 2026 | — |
| $2B | “two billion dollars” | Malde walked away from $2B DeepMind acquisition to found Trajectory.ai | Ronak Malde | Jun 21, 2026 | — |
| 80% | “80%” | Malde: In legal AI, getting 80% of the way there is zero | Ronak Malde | Jun 21, 2026 | — |
| 1T | “one trillion” | Malde: Western open-source AI lags Chinese models at trillion-parameter scale | Ronak Malde | Jun 21, 2026 | — |
| 95% | “95%” | Most AI clusters fail to hit Google's 96% node utilization standard | Anjney Midha | Jun 18, 2026 | — |
| 96% | “96%” | Most AI clusters fail to hit Google's 96% node utilization standard | Anjney Midha | Jun 18, 2026 | — |
| 70% | “70%” | Midha: Best-in-class AI cluster MFU is between 60% and 70% | Anjney Midha | Jun 18, 2026 | Partly supported |
| 20% | “20%” | Up to 20% of US data centers risk cancellation from community backlash | Anjney Midha | Jun 18, 2026 | Open |
| $100M | “hundred million dollars” | Midha: AMP Foundry invested hundreds of millions into Anthropic this year | Anjney Midha | Jun 18, 2026 | — |
| 12M | “twelve million” | Midha: Stanford Holds America's Second-Largest Longitudinal Patient Dataset Behind the VA | Anjney Midha | Jun 18, 2026 | Contradicted |
| 1T | “a trillion” | Anthropic will become a trillion-dollar company within four years of founding | Anjney Midha | Jun 18, 2026 | Open |
| 15× | “15 X” | Krause: Hafnium prices surged 10x to 15x due to Chinese supply dominance | Joseph Krause | Jun 17, 2026 | Supported |
| 10% | “10%” | Krause: Common aerospace alloy C-103 contains roughly 10% hafnium by weight | Joseph Krause | Jun 17, 2026 | Supported |
| 5× | “five X” | Krause: New semiconductor interconnect materials could deliver 2x to 10x+ efficiency gains | Joseph Krause | Jun 17, 2026 | — |
| 10× | “10 X” | Krause: New semiconductor interconnect materials could deliver 2x to 10x+ efficiency gains | Joseph Krause | Jun 17, 2026 | — |
| 16% | “16%” | Krause: Radical AI's Matrix VLM boosts scientific reasoning by 5-16% | Joseph Krause | Jun 17, 2026 | Supported |
| 1.2B | “1.2 billion” | Awais: LangBase grew to 1.2 billion agent runs per month | Ahmad Awais | Jun 6, 2026 | — |
| 600B | “six hundred billion” | Awais: CommandCode has 16,000 tool repair variations across 600B tokens | Ahmad Awais | Jun 6, 2026 | — |
| 600M | “six hundred million” | CommandCode offers 600 million DeepSeek tokens for $1 per month | Ahmad Awais | Jun 6, 2026 | — |
| 90% | “90%” | Awais: AI design slop is a contract gap, not a model capability deficit | Ahmad Awais | Jun 6, 2026 | — |
| $1K | “a thousand dollars” | Backlund: Andon Labs agent tried TaskRabbit arbitrage to make money | Axel Backlund | Jun 4, 2026 | — |
| 10× | “10 times” | Petersson: Opus repeatedly lied, exploited agents, and formed price cartels | Lukas Petersson | Jun 4, 2026 | Partly supported |
| 100× | “hundred times” | Petersson: Opus repeatedly lied, exploited agents, and formed price cartels | Lukas Petersson | Jun 4, 2026 | Partly supported |
| 99% | “99%” | Hong: Axiom's unmodified Putnam system achieved 99% on Verina benchmark | Carina Hong | Jun 3, 2026 | Open |
| 10B | “ten billion” | Hong: Axiom Math will be worth $10 billion | Carina Hong | Jun 3, 2026 | Open |
| 10× | “10 X” | Nadella: AI agent usage of Microsoft 365 will exceed human users | Satya Nadella | Jun 3, 2026 | — |
| 1M | “a million” | Daigle: AI code vendoring will not solve supply chain vulnerabilities | Kyle Daigle | Jun 2, 2026 | — |
| 200M | “two hundred million” | Daigle: GitHub has surpassed 200 million registered developers | Kyle Daigle | Jun 2, 2026 | Supported |
| 100% | “hundred percent” | Ethan He: Video models require image foundations and 100% synthetic caption pairs | Ethan He | Jun 1, 2026 | — |
| 1K | “a thousand” | Ethan He: Training transformers directly on raw image pixels is impossible | Ethan He | Jun 1, 2026 | — |
| 1M | “one million” | Ethan He: Training transformers directly on raw image pixels is impossible | Ethan He | Jun 1, 2026 | — |
| 1B | “a billion” | Ethan He: Video Models Must Bootstrap From Image Diffusion Models for Semantic Understanding | Ethan He | Jun 1, 2026 | — |
| 11M | “a ten million” | Ethan He: Video Models Must Bootstrap From Image Diffusion Models for Semantic Understanding | Ethan He | Jun 1, 2026 | — |
| 4× | “four times” | Ethan He: Temporal video compression cuts context length up to 4x | Ethan He | Jun 1, 2026 | — |
| 4× | “four times” | Ethan He: Frame-by-frame video compression enables real-time interactivity, temporal compression adds lag | Ethan He | Jun 1, 2026 | — |
| 1K | “a thousand” | Shawn Wang: Fixed-capability LLM inference costs drop 100x to 1000x annually | Shawn Wang | Jun 1, 2026 | Partly supported |
| 100B | “a hundred billion” | Ethan He: Megatron MoE was first to train trillion-parameter MoEs at 40% MFU | Ethan He | Jun 1, 2026 | Supported |
| 40% | “40%” | Ethan He: Megatron MoE was first to train trillion-parameter MoEs at 40% MFU | Ethan He | Jun 1, 2026 | Supported |
| 10% | “10%” | Yan: Cognition merged PRs grew 7x as headcount grew only 10% | Walden Yan | May 28, 2026 | — |
| 7× | “seven X” | Yan: Cognition merged PRs grew 7x as headcount grew only 10% | Walden Yan | May 28, 2026 | — |
| 95% | “95%” | Yan: Roughly 95% of Devin's memories are auto-generated | Walden Yan | May 28, 2026 | — |
| 1K | “a thousand” | Murray: AI agent budgets typically run $1,000 to $5,000 per engineer | Cole Murray | May 28, 2026 | — |
| 6.8B | “6.8 billion” | Biohub predicted 3D structures for 1.1 billion proteins from global sequence databases | Alex Rives | May 27, 2026 | Supported |
| 1.1B | “1.1 billion” | Biohub predicted 3D structures for 1.1 billion proteins from global sequence databases | Alex Rives | May 27, 2026 | Supported |
| 400M | “a three hundred million” | Rives: The ESMC family includes 300M, 600M, and 6B parameter models | Alex Rives | May 27, 2026 | Supported |
| 700M | “a six hundred million” | Rives: The ESMC family includes 300M, 600M, and 6B parameter models | Alex Rives | May 27, 2026 | Supported |
| 7B | “a six billion” | Rives: The ESMC family includes 300M, 600M, and 6B parameter models | Alex Rives | May 27, 2026 | Supported |
| 400M | “four hundred million” | CZ Biohub commits $500M to scale biological data creation and technology development | Alex Rives | May 27, 2026 | — |
| 100M | “a hundred million” | CZ Biohub commits $500M to scale biological data creation and technology development | Alex Rives | May 27, 2026 | — |
| 10× | “10 X” | Rives: Existing Tech Can Scale Biological Data 10x–100x Before Hitting Limits | Alex Rives | May 27, 2026 | — |
| 100× | “hundred X” | Rives: Existing Tech Can Scale Biological Data 10x–100x Before Hitting Limits | Alex Rives | May 27, 2026 | — |
| 2B | “two billion” | Gemma 4 E2B loads only 2B of 5B parameters into GPU | Omar Sanseviero | May 24, 2026 | Supported |
| 5B | “five billion” | Gemma 4 E2B loads only 2B of 5B parameters into GPU | Omar Sanseviero | May 24, 2026 | Supported |
| 3B | “three billion” | Gemma 4 E2B loads only 2B of 5B parameters into GPU | Omar Sanseviero | May 24, 2026 | Supported |
| 74% | “74%” | Wang: Daytona reports massive 74% month-over-month growth | Shawn Wang | May 21, 2026 | Supported |
| 1M | “a million” | Daytona's top customer runs 850,000 daily agent instances | Ivan Burazin | May 21, 2026 | — |
| 15% | “15%” | Daytona's mean compute utilization is 15%, peaking at 90% | Ivan Burazin | May 21, 2026 | — |
| 90% | “90%” | Daytona's mean compute utilization is 15%, peaking at 90% | Ivan Burazin | May 21, 2026 | — |
| 100% | “hundred percent” | Burazin: CPU environments must spin up instantly to prevent costly GPU idle time | Ivan Burazin | May 21, 2026 | — |
| 50% | “50%” | Burazin: RL workloads will reach 50% of Daytona's volume this month | Ivan Burazin | May 21, 2026 | — |
| 1K | “a thousand” | Burazin: Daytona operates roughly 1,000 Slack Connect customer channels | Ivan Burazin | May 21, 2026 | — |
| 1K | “a thousand” | Burazin: Agent-generated PR volume has made CI/CD the main engineering bottleneck | Ivan Burazin | May 21, 2026 | — |
| $1.5M | “half a million dollars” | Cooper: Railway burned roughly $500k monthly during free tier era | Jake Cooper | May 20, 2026 | — |
| 70% | “70%” | Cooper: Railway achieves roughly 70% margins on bare-metal infrastructure | Jake Cooper | May 20, 2026 | — |
| 1K | “a thousand” | Cooper: Agent infra requires 1000x scale rather than new primitives | Jake Cooper | May 20, 2026 | — |
| 3M | “three million” | Cooper: Railway has scaled to 3 million users | Jake Cooper | May 20, 2026 | Supported |
| 1K | “a thousand” | Cooper: Agentic workloads will destroy traditional static AWS instances | Jake Cooper | May 20, 2026 | — |
| 4M | “a three million” | Cooper: Companies will reserve massive token budgets for elite AI developers | Jake Cooper | May 20, 2026 | — |
| 100% | “hundred percent” | Cooper: The vast majority of Railway workloads run on bare metal | Jake Cooper | May 20, 2026 | — |
| 80% | “80%” | Azhnyuk: FPV Drones Cause 70% to 80% of Frontline Casualties | Yaroslav Azhnyuk | May 18, 2026 | Supported |
| 100× | “hundred times” | Azhnyuk: Full autonomy increases FPV drone capability by 10,000 times | Yaroslav Azhnyuk | May 18, 2026 | — |
| 10× | “10 times” | Azhnyuk: Full autonomy increases FPV drone capability by 10,000 times | Yaroslav Azhnyuk | May 18, 2026 | — |
| 20% | “20%” | Azhnyuk: Terminal guidance boosted Ukrainian drone pilot's success rate to 71% | Yaroslav Azhnyuk | May 18, 2026 | — |
| 71% | “71%” | Azhnyuk: Terminal guidance boosted Ukrainian drone pilot's success rate to 71% | Yaroslav Azhnyuk | May 18, 2026 | — |
| 4M | “four million” | Azhnyuk: Ukraine made 4M FPV drones last year with 7M targeted this year | Yaroslav Azhnyuk | May 18, 2026 | Supported |
| 7M | “seven million” | Azhnyuk: Ukraine made 4M FPV drones last year with 7M targeted this year | Yaroslav Azhnyuk | May 18, 2026 | Supported |
| 20% | “20%” | Azhnyuk: Ukraine made 4M FPV drones last year with 7M targeted this year | Yaroslav Azhnyuk | May 18, 2026 | Supported |
| 20% | “20%” | Lee: Patient-doctor conversations drive nearly all downstream healthcare spending | Janie Lee | May 14, 2026 | — |
| 90% | “90%” | Lee: Over 90% of healthcare alerts are ignored | Janie Lee | May 14, 2026 | Supported |
| 80M | “eighty million” | Asawa: Abridge holds proprietary dataset approaching hundreds of millions of medical conversations | Chai Asawa | May 14, 2026 | — |
| 9M | “nine million” | Swyx: Obsidian scaled its product with only four team members | Shawn Wang | May 7, 2026 | Partly supported |
| 1M | “one million” | Chatbase scaled from zero to $1 million ARR in 117 days | Yasser Elsaid | May 2, 2026 | — |
| 50% | “50%” | Chatbase customer model usage is split 50% OpenAI, 50% Anthropic and Google | Yasser Elsaid | May 2, 2026 | — |
| 95% | “95%” | Yasser Elsaid: 95% of AI customer service limitations stem from harness architecture | Yasser Elsaid | May 2, 2026 | — |
| 90% | “90%” | Proper AI harnesses enable 80% to 90% customer issue resolution rates | Yasser Elsaid | May 2, 2026 | — |
| 83% | “83%” | Younis: 83% of Applied Intuition's workforce is engineering | Qasar Younis | Apr 27, 2026 | — |
| 95% | “95%” | Younis: Vehicle software testing is 95% CI/CD, 4% rig, 1% physical | Qasar Younis | Apr 27, 2026 | — |
| 4% | “four percent” | Younis: Vehicle software testing is 95% CI/CD, 4% rig, 1% physical | Qasar Younis | Apr 27, 2026 | — |
| 1% | “one percent” | Younis: Vehicle software testing is 95% CI/CD, 4% rig, 1% physical | Qasar Younis | Apr 27, 2026 | — |
| 100% | “hundred percent” | Parakhin: Daily AI tool adoption at Shopify approaches 100 percent | Mikhail Parakhin | Apr 22, 2026 | — |
| 300M | “three hundred million” | Parakhin: Shopify runs a 300M parameter Liquid model under 30ms for search | Mikhail Parakhin | Apr 22, 2026 | — |
| 60M | “sixty million” | Eifrem: Novo Nordisk graph deployment spans over 60 million documents | Emil Eifrem | Apr 18, 2026 | Contradicted |
| 1M | “one million” | Swix: Frontier model context windows will not reach billion-token scales | Shawn Wang | Apr 18, 2026 | Open |
| 1B | “a billion” | Swix: Frontier model context windows will not reach billion-token scales | Shawn Wang | Apr 18, 2026 | Open |
| 1T | “a trillion” | Swix: Frontier model context windows will not reach billion-token scales | Shawn Wang | Apr 18, 2026 | Open |
| 30% | “30%” | Notion Partnered With Anthropic and OpenAI to Build 30% Pass Rate Evals | Sarah Sachs | Apr 15, 2026 | — |
| 99% | “99%” | Simon Last: 99% of Agent Failures Are Tool Bugs, Not Model Flaws | Simon Last | Apr 15, 2026 | — |
| 70% | “70%” | Stay Sassy EM: 70% of executive work is standard and automatable | Stay Sassy EM | Apr 13, 2026 | — |
| 99% | “99%” | Stay Sassy PM: Leaders know what to do 99% of the time | Stay Sassy PM | Apr 13, 2026 | — |
| 10× | “10 times” | Lopopolo: Zero-code harness was 10x slower initially before outperforming any single engineer | Ryan Lopopolo | Apr 7, 2026 | — |
| 2M | “two million” | Lopopolo: Codex App Hits 2M WAUs, Growing 25% Week-Over-Week | Ryan Lopopolo | Apr 7, 2026 | — |
| 25% | “25%” | Lopopolo: Codex App Hits 2M WAUs, Growing 25% Week-Over-Week | Ryan Lopopolo | Apr 7, 2026 | — |
| 100% | “100%” | Andreessen: AI will never transform existing US K-12 public classrooms | Marc Andreessen | Apr 3, 2026 | Didn’t hold up |
| 8B | “eight billion” | Andreessen: Utopians and doomers overestimate how quickly humans adapt to AI | Marc Andreessen | Apr 3, 2026 | — |
| 90% | “90%” | Manning: Vision understanding stalled; language does 90% of work in VLMs | Chris Manning | Apr 2, 2026 | — |
| 3× | “three times” | D'Amico: Impulse delivers up to 5x more power than standard induction stoves | Sam D'Amico | Mar 31, 2026 | Supported |
| 5× | “five times” | D'Amico: Impulse delivers up to 5x more power than standard induction stoves | Sam D'Amico | Mar 31, 2026 | Supported |
| 10× | “10 X” | Lample: Mistral can build fine-tuned models 10x cheaper than closed models | Guillaume Lample | Mar 30, 2026 | — |
| 4× | “four times” | Kulik: AI Discovered Polymer Design That Made Plastic Four Times Tougher | Heather Kulik | Mar 24, 2026 | Supported |
| 1K | “a thousand” | Kulik: ML models deliver 100x to 1000x speedup per optimization dimension | Heather Kulik | Mar 24, 2026 | — |
| 5× | “five times” | Unnamed materials foundation model is only 5x faster than DFT and unreliable | Heather Kulik | Mar 24, 2026 | — |
| 100% | “hundred percent” | Rieseberg: Sandboxing bypasses need for 100% model alignment | Felix Rieseberg | Mar 17, 2026 | — |
| 3× | “three X” | Colvin: LLMs are 100x faster under four specific engineering conditions | Samuel Colvin | Mar 14, 2026 | — |
| 5× | “five X” | Colvin: LLMs are 100x faster under four specific engineering conditions | Samuel Colvin | Mar 14, 2026 | — |
| 100× | “hundred X” | Colvin: LLMs are 100x faster under four specific engineering conditions | Samuel Colvin | Mar 14, 2026 | — |
| 1M | “a million” | Eskildsen: Kardashian Sales on Shopify Peaked at 1 Million Requests per Second | Simon Eskildsen | Mar 12, 2026 | Partly supported |
| 3× | “three times” | Eskildsen: Modern hardware requires databases designed around concurrent requests, not sequential roundtrips | Simon Eskildsen | Mar 12, 2026 | — |
| 95% | “95%” | Eskildsen: Turbopuffer reduced Cursor's infrastructure costs by 95% | Simon Eskildsen | Mar 12, 2026 | — |
| 100B | “a hundred billion” | Eskildsen: Turbopuffer ANN v3 searches 100B vectors with 40ms p50 latency | Simon Eskildsen | Mar 12, 2026 | Supported |
| 50% | “50%” | Shah: Supermemory outperformed Claude Code and OpenClaw benchmarks by almost 50% | Dhravya Shah | Mar 9, 2026 | Supported |
| 100% | “hundred percent” | Shah: LoCoMo benchmark is flawed because models pass via context dumping | Dhravya Shah | Mar 9, 2026 | — |
| 10× | “10 X” | Shah: OpenClaw's 15-message replay fails prompt caching and costs 10x more | Dhravya Shah | Mar 9, 2026 | Supported |
| 1M | “a million” | Shah: Supermemory infrastructure costs only two cents per million tokens | Dhravya Shah | Mar 9, 2026 | — |
| 10M | “ten million” | Algorithmic unhobblers could soon expand context windows to 100 million tokens | Kyle Kranen | Mar 8, 2026 | — |
| 20M | “twenty million” | Algorithmic unhobblers could soon expand context windows to 100 million tokens | Kyle Kranen | Mar 8, 2026 | — |
| 100M | “a hundred million” | Algorithmic unhobblers could soon expand context windows to 100 million tokens | Kyle Kranen | Mar 8, 2026 | — |
| 1% | “one percent” | Horthy: Top 1% of AI builders don't use popular public frameworks | Dex Horthy | Mar 6, 2026 | — |
| 99% | “99%” | Horthy: Top 1% of AI builders don't use popular public frameworks | Dex Horthy | Mar 6, 2026 | — |
| 80% | “80%” | Horthy: AI demos work at 80% accuracy, enterprise software needs far more | Dex Horthy | Mar 6, 2026 | — |
| 40% | “40%” | Horthy: Beginners should compact LLM context at 40% of window capacity | Dex Horthy | Mar 6, 2026 | — |
| 10× | “10 X” | Nelle: Cursor's agent swarms broke internal GitHub Actions CI/CD pipelines | Jonas Nelle | Mar 6, 2026 | — |
| 10× | “10 X” | Levie: Enterprises will inevitably deploy an order of magnitude more agents than people | Aaron Levie | Mar 5, 2026 | — |
| 100× | “hundred X” | Levie: Enterprises will inevitably deploy an order of magnitude more agents than people | Aaron Levie | Mar 5, 2026 | — |
| 98% | “98%” | Levie: Access governance will decide if enterprise AI agents leak data constantly | Aaron Levie | Mar 5, 2026 | — |
| 67% | “67%” | Levie: Box serves 67% of Fortune 500 companies | Aaron Levie | Mar 5, 2026 | Supported |
LATENT SPACELEDGER All shows Try the podcast clipper →