| 10,000× | “10,000 X” | Developer Productivity Will Continue Growing 10x Per Decade | Erik Bernhardsson | Feb 19, 2024 | — |
| 2,625× | “2625 times” | Feldman: Cerebras provides 2,625x more memory bandwidth than traditional GPUs | Andrew Feldman | Oct 1, 2025 | Supported |
| 2,500× | “2500 times” | Gomez-Bombarelli: Lila tests 96 MOFs in an hour using proxy measurements | Rafa Gomez-Bombarelli | Jul 16, 2026 | — |
| 500× | “500 times” | Bissell: Probing internal model features matches LLM-as-a-judge quality at 500x lower cost | Mark Bissell | Dec 31, 2025 | — |
| 300× | “300 X” | AMD MI300X outperforms Nvidia H100 on FlashAttention-2 and memory-bound workloads | Quentin Anthony | Nov 3, 2025 | Supported |
| 300× | “300 X” | AMD MI300X GEMM performance increased from 400 to 650 TFLOPS via software | Quentin Anthony | Nov 3, 2025 | Supported |
| 200× | “200 X” | Bryk: 200x drop in LLM costs requires rethinking search from scratch | Will Bryk | Jan 10, 2025 | — |
| 100× | “hundred X” | Ultra-fast network interface cards could deliver 100x speedups in disaggregated AI inference | Ali Taha | Aug 3, 2026 | — |
| 100× | “hundred X” | Nathan: Untapped enterprise agent market is 10x to 100x larger | Akshay Nathan | Jul 28, 2026 | — |
| 100× | “hundred X” | Zaharia: Databricks' document vision model is ~100x cheaper than frontier models | Matei Zaharia | Jun 24, 2026 | — |
| 100× | “hundred times” | Petersson: Opus repeatedly lied, exploited agents, and formed price cartels | Lukas Petersson | Jun 4, 2026 | Partly supported |
| 100× | “hundred X” | Rives: Existing Tech Can Scale Biological Data 10x–100x Before Hitting Limits | Alex Rives | May 27, 2026 | — |
| 100× | “hundred times” | Azhnyuk: Full autonomy increases FPV drone capability by 10,000 times | Yaroslav Azhnyuk | May 18, 2026 | — |
| 100× | “hundred X” | Colvin: LLMs are 100x faster under four specific engineering conditions | Samuel Colvin | Mar 14, 2026 | — |
| 100× | “hundred X” | Levie: Enterprises will inevitably deploy an order of magnitude more agents than people | Aaron Levie | Mar 5, 2026 | — |
| 100× | “hundred times” | Levie: AI agents will generate 10 to 100 times more code | Aaron Levie | Mar 5, 2026 | — |
| 100× | “hundred X” | Jeff Dean: Design systems to scale 5x to 10x, not 100x | Jeff Dean | Feb 12, 2026 | — |
| 100× | “hundred X” | Dean: Transformers delivered 10x to 100x compute efficiency over LSTMs | Jeff Dean | Feb 12, 2026 | Supported |
| 100× | “hundred X” | White: Future scientists will act as AI agent wranglers exploring 100 ideas simultaneously | Andrew White | Jan 28, 2026 | — |
| 100× | “hundred times” | GPT-4-level intelligence is now over 100 times cheaper than at launch | Micah Hill-Smith | Jan 9, 2026 | Supported |
| 100× | “hundred times” | Rumbelow: Leap Labs is roughly 100x faster than manual analysis | Jessica Rumbelow | Nov 2, 2025 | — |
| 100× | “hundred times” | Slack: Asynchronous background agents will dominate AI inference and output | Quinn Slack | Sep 25, 2025 | — |
| 100× | “hundred X” | Bachman: Power Retention Delivers 100x Inference Speedup at 64k Context | Diego Bachman | Sep 23, 2025 | Open |
| 100× | “hundred X” | Morcos: Data curation still has at least 100x in performance gains ahead | Ari Morcos | Aug 29, 2025 | — |
| 100× | “hundred X” | Lambert: Hybrid reasoners may be phased out except for niche uses | Nathan Lambert | Jul 31, 2025 | — |
| 100× | “hundred x” | Brockman: AI developer productivity gains will increase demand for engineers | Greg Brockman | Aug 15, 2025 | — |
| 100× | “100 X” | Hou: Codeium inference costs 1/100th of competitors by avoiding third-party APIs | Kevin Hou | Jul 28, 2025 | — |
| 100× | “hundred X” | Hsu: AI Gives Content and Engineering 100x Leverage While Still Requiring Review | Andrew Hsu | Jul 11, 2025 | — |
| 100× | “hundred X” | Swix: AI Inference Costs for Fixed Intelligence Fall 100x Annually | Shawn Wang | Jul 5, 2025 | Partly supported |
| 100× | “hundred times” | Mlejnsky: DevTool Startups in Existing Categories Do Not Need to Be in SF | Vasek Mlejnsky | Apr 24, 2025 | — |
| 100× | “hundred times” | Untrained AI models exhibit a 100x competency gap versus human players | Jack Hopkins | Apr 27, 2025 | Contradicted |
| 100× | “hundred X” | Drew Houston: Massive AI price-performance gains wouldn't happen without open source | Drew Houston | Oct 18, 2024 | — |
| 100× | “hundred times” | Yao: Enterprise customer support AI requires 99% reliability over simple tasks, not search | Shunyu Yao | Sep 27, 2024 | — |
| 100× | “hundred times” | Lambert: A 100x smaller language model filters output better than RLHF | Nathan Lambert | Jan 11, 2024 | — |
| 100× | “hundred times” | Lie: Cerebras chips have 100x more memory than Groq LPUs | Sean Lie | Sep 2, 2026 | Supported |
| 70× | “70 times” | Cerebras WSE-3 runs Llama inference 70x faster than NVIDIA GPUs | Sarah Chieng | Dec 7, 2024 | Supported |
| 64× | “64 X” | Wu: Autonomous coding agent capability currently doubles every 70 days | Scott Wu | Jul 28, 2025 | — |
| 64× | “64 X” | Wu predicts AI coding agents will advance 16x to 64x in 12 months | Scott Wu | Jul 28, 2025 | — |
| 50× | “50 x” | Dean: AI systems will achieve 20x to 50x lower latency | Jeff Dean | Feb 12, 2026 | Open |
| 50× | “50 times” | Sands: Manual writing forces first-principles reasoning that LLMs dangerously bypass | Emily Glassberg Sands | Oct 30, 2025 | — |
| 50× | “50 times” | Reinforcement Learning Fails Without Initial SFT to Seed Rewardable Behaviors | Misha Laskin | Mar 7, 2025 | — |
| 50× | “50 times” | Shulman: Gaming dwarfs music 50x because music is mostly passive consumption | Mikey Shulman | Mar 14, 2024 | — |
| 26× | “26 X” | Flux reached 7,000 paying customers and grew 26x year-over-year | Matthias Wagner | Nov 22, 2025 | — |
| 25× | “25 times” | Feldman: Non-compute bottlenecks masked Cerebras' 20-25x speedup for a hyperscaler | Andrew Feldman | Oct 1, 2025 | — |
| 20× | “20 times” | Fredrikson: Formal software verification takes 10 to 20 times longer than Python | Matt Fredrikson | Jun 22, 2026 | — |
| 20× | “20 times” | Levie: AI slop in legal contracts creates risks code slop never introduced | Aaron Levie | Mar 5, 2026 | — |
| 20× | “20 x” | Dean: AI systems will achieve 20x to 50x lower latency | Jeff Dean | Feb 12, 2026 | Open |
| 20× | “20 times” | Agarwal: Sequoia met 20 AI SRE startups before backing Traversal | Anish Agarwal | Oct 5, 2025 | — |
| 20× | “20 times” | Feldman: Cerebras is 20 times faster than Nvidia B200 GPUs | Andrew Feldman | Oct 1, 2025 | Contradicted |
| 15× | “15 X” | Krause: Hafnium prices surged 10x to 15x due to Chinese supply dominance | Joseph Krause | Jun 17, 2026 | Supported |
| 15× | “15 times” | Ball: Amp core team has 8 people, ships 15 times daily without code reviews | Thorsten Ball | Sep 25, 2025 | — |
| 15× | “15 X” | Davis: Multi-agent research consumes 15x baseline tokens versus 4x for single-agent | Dylan Davis | Jul 5, 2025 | Supported |
| 15× | “15 X” | Swyx: Distilling GPT-4 to Mini cut costs 15x with 2% hit | Shawn Wang | Oct 4, 2024 | Supported |
| 14× | “14 times” | Mohan: Training optimizer state requires 14x model parameter size in memory | Varun Mohan | Jul 28, 2025 | Supported |
| 14× | “14 times” | Lie: Cerebras runs OpenAI's flagship model 14x faster than GPUs | Sean Lie | Sep 2, 2026 | Supported |
| 12× | “12 x” | Morcos: Datology matches DCLM performance 12x faster with under 10% tokens | Ari Morcos | Aug 29, 2025 | Open |
| 10× | “10 X” | Unoptimized GLM-5.2 delivers a baseline 30 to 40 tokens per second | Philip Kiely | Aug 3, 2026 | — |
| 10× | “10 X” | Nathan: Untapped enterprise agent market is 10x to 100x larger | Akshay Nathan | Jul 28, 2026 | — |
| 10× | “10 X” | Krause: New semiconductor interconnect materials could deliver 2x to 10x+ efficiency gains | Joseph Krause | Jun 17, 2026 | — |
| 10× | “10 times” | Petersson: Opus repeatedly lied, exploited agents, and formed price cartels | Lukas Petersson | Jun 4, 2026 | Partly supported |
| 10× | “10 X” | Nadella: AI agent usage of Microsoft 365 will exceed human users | Satya Nadella | Jun 3, 2026 | — |
| 10× | “10 X” | Rives: Existing Tech Can Scale Biological Data 10x–100x Before Hitting Limits | Alex Rives | May 27, 2026 | — |
| 10× | “10 times” | Azhnyuk: Full autonomy increases FPV drone capability by 10,000 times | Yaroslav Azhnyuk | May 18, 2026 | — |
| 10× | “10 times” | Lopopolo: Zero-code harness was 10x slower initially before outperforming any single engineer | Ryan Lopopolo | Apr 7, 2026 | — |
| 10× | “10 X” | Lample: Mistral can build fine-tuned models 10x cheaper than closed models | Guillaume Lample | Mar 30, 2026 | — |
| 10× | “10 X” | Shah: OpenClaw's 15-message replay fails prompt caching and costs 10x more | Dhravya Shah | Mar 9, 2026 | Supported |
| 10× | “10 X” | Levie: Enterprises will inevitably deploy an order of magnitude more agents than people | Aaron Levie | Mar 5, 2026 | — |
| 10× | “10 X” | Nelle: Cursor's agent swarms broke internal GitHub Actions CI/CD pipelines | Jonas Nelle | Mar 6, 2026 | — |
| 10× | “10 X” | Dean: Sparse models offer 10x compute cost efficiency over dense models | Jeff Dean | Feb 12, 2026 | Supported |
| 10× | “10 X” | Dean: Transformers delivered 10x to 100x compute efficiency over LSTMs | Jeff Dean | Feb 12, 2026 | Supported |
| 10× | “10 times” | Garg: Context graph stack will eclipse modern data stack value by 10x | Ashu Garg | Feb 4, 2026 | — |
| 10× | “10 X” | Reggio: Brex structures its AI strategy into corporate, operational, and product pillars | James Reggio | Jan 17, 2026 | — |
| 10× | “10 times” | Reasoning models consume 10x more tokens on average than non-reasoning models | Micah Hill-Smith | Jan 9, 2026 | Supported |
| 10× | “10 X” | Yegge: OpenAI sees a 10x productivity gap between AI adopters and non-adopters | Steve Yegge | Dec 26, 2025 | — |
| 10× | “10 times” | Yegge: 10x AI productivity turns merging into architectural rewrites | Steve Yegge | Dec 26, 2025 | — |
| 10× | “10 times” | Zemlin: MCP and AAIF adoption will move ten times faster than Kubernetes | Jim Zemlin | Dec 28, 2025 | — |
| 10× | “10 times” | Wagner: Task-specific knowledge injection effectively reduces LLM hallucinations | Matthias Wagner | Nov 22, 2025 | — |
| 10× | “10 times” | Corbitt: AI inference could be 10x larger if reliability issues are solved | Kyle Corbitt | Oct 16, 2025 | — |
| 10× | “10 X” | Bachman: Power Retention Delivers 100x Inference Speedup at 64k Context | Diego Bachman | Sep 23, 2025 | Open |
| 10× | “10 X” | Morcos: Power-law scaling yields diminishing returns for every 10x data increase | Ari Morcos | Aug 29, 2025 | — |
| 10× | “10 X” | Morcos: Data curation still has at least 100x in performance gains ahead | Ari Morcos | Aug 29, 2025 | — |
| 10× | “10 times” | Morcos: Proper training curricula could reduce model training costs by 10x | Ari Morcos | Aug 29, 2025 | — |
| 10× | “10 x” | Brockman: AI developer productivity gains will increase demand for engineers | Greg Brockman | Aug 15, 2025 | — |
| 10× | “10 X” | Inception generalist model matches Claude Haiku quality at 5-10x speed | Stefano Ermon | Aug 4, 2025 | Partly supported |
| 10× | “10 X” | Ermon: Power constraints will drive diffusion models to replace frontier LLMs | Stefano Ermon | Aug 4, 2025 | — |
| 10× | “10 X” | Scott Wu: AI will make engineers 5-10x more effective | Scott Wu | Jul 28, 2025 | — |
| 10× | “10 X” | Cherny: Claude Code delivers up to 10x productivity gains for Anthropic engineers | Boris Cherny | May 7, 2025 | — |
| 10× | “10 times” | Mlejnsky: DevTool Startups in Existing Categories Do Not Need to Be in SF | Vasek Mlejnsky | Apr 24, 2025 | — |
| 10× | “10 times” | Reinforcement Learning Fails Without Initial SFT to Seed Rewardable Behaviors | Misha Laskin | Mar 7, 2025 | — |
| 10× | “10 X” | Beauchamp: AI is 10x better at non-judgmental conversation than informative tasks | William Beauchamp | Jan 26, 2025 | — |
| 10× | “10 X” | Bret Taylor's Rewrite Slashed Google Maps Bundle Size to 20KB | Bret Taylor | Feb 11, 2025 | Not publicly verifiable |
| 10× | “10 X” | Swyx: AI models hit a 2T parameter wall, won't reach 10T | Shawn Wang | Jan 1, 2025 | — |
| 10× | “10 times” | Neubig Uses AI Coding Agents Five To Ten Times Daily | Graham Neubig | Dec 25, 2024 | — |
| 10× | “10 X” | Simons: Base AI models provide a 10x lift; agent scaffolding adds 3-4x | Eric Simons | Dec 2, 2024 | — |
| 10× | “10 times” | Swix: Multi-sampling GPT-4o mini before GPT-4o judging yields net savings | Shawn Wang | Sep 20, 2024 | Supported |
| 10× | “10 X” | Brady: Elicit routinely sees 10x p90 latency variation when prompting LLMs | James Brady | Jun 21, 2024 | — |
| 10× | “10 times” | Zhang: Next 10x AI inference gain requires multi-layer co-optimization | Ce Zhang | Feb 8, 2024 | — |
| 10× | “10 X” | Developer Productivity Will Continue Growing 10x Per Decade | Erik Bernhardsson | Feb 19, 2024 | — |
| 10× | “10 x” | AI Productivity Gains Will Ultimately Increase Demand for Software Engineers | Erik Bernhardsson | Feb 19, 2024 | — |
| 10× | “tenx” | Hsu: AI coding assistants only speed up development by 10% to 20% | David Hsu | Feb 7, 2024 | — |
| 10× | “10 X” | Hotz: Tinygrad runs all ML models with only 25 primitive operations | George Hotz | Jun 20, 2023 | Supported |
| 9× | “nine X” | McCloy: Clerk Achieved 6x AI Traffic Growth and 9x Lift in Conversions | Robert McCloy | Jul 23, 2025 | — |
| 9× | “nine X” | Masking out prior KV cache tokens pushes autoregressive transformers out of distribution | Umar Jamil | Sep 19, 2024 | — |
| 7× | “seven X” | Yan: Cognition merged PRs grew 7x as headcount grew only 10% | Walden Yan | May 28, 2026 | — |
| 7× | “seven times” | Sands: Stripe Merchants Grew Seven Times Faster Than S&P 500 | Emily Glassberg Sands | Oct 30, 2025 | Supported |
| 6× | “six X” | McCloy: Clerk Achieved 6x AI Traffic Growth and 9x Lift in Conversions | Robert McCloy | Jul 23, 2025 | — |
| 6× | “six times” | Robinson: SAM 2's hierarchical encoder delivers 6x faster inference than ViT | Isaac Robinson | Dec 22, 2024 | Supported |
| 6× | “six times” | Ravi: SAM 2 runs roughly six times faster on video than SAM 1 | Nikhila Ravi | Aug 7, 2024 | Supported |
| 5× | “five X” | Krause: New semiconductor interconnect materials could deliver 2x to 10x+ efficiency gains | Joseph Krause | Jun 17, 2026 | — |
| 5× | “five times” | D'Amico: Impulse delivers up to 5x more power than standard induction stoves | Sam D'Amico | Mar 31, 2026 | Supported |
| 5× | “five times” | Unnamed materials foundation model is only 5x faster than DFT and unreliable | Heather Kulik | Mar 24, 2026 | — |
| 5× | “five X” | Colvin: LLMs are 100x faster under four specific engineering conditions | Samuel Colvin | Mar 14, 2026 | — |
| 5× | “five X” | Intelligent compute routing delivers 5x performance TCO benefit | Chris Lattner | Jun 13, 2025 | — |
| 5× | “five times” | Cherny: Claude Code spawns parallel sub-agents to investigate complex coding tasks | Boris Cherny | May 7, 2025 | Supported |
| 5× | “five times” | Colvin: Rust-native data storage could yield 3x to 5x Pydantic speedup | Samuel Colvin | Feb 6, 2025 | Open |
| 5× | “five X” | Drew Houston: AI hardware is currently in a 'rent, not buy' phase | Drew Houston | Oct 18, 2024 | — |
| 5× | “five X” | Hotz: Tinygrad is about 5x slower than PyTorch on Nvidia GPUs | George Hotz | Jun 20, 2023 | Supported |
| 5× | “five X” | Hotz: Tinybox will be 5x faster than H100 systems per dollar | George Hotz | Jun 20, 2023 | — |
| 4× | “four times” | Baseten terminates or reprocesses generations after four identical repeating tokens | Ali Taha | Aug 3, 2026 | — |
| 4× | “four X” | Software optimizations yield 2x to 4x inference speedups on identical hardware | Philip Kiely | Aug 3, 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 | — |
| 4× | “four times” | Kulik: AI Discovered Polymer Design That Made Plastic Four Times Tougher | Heather Kulik | Mar 24, 2026 | Supported |
| 4× | “four X” | O'Laughlin: Manufacturing one bit of HBM takes 4x the DRAM wafer capacity | Doug O'Laughlin | Feb 24, 2026 | Supported |
| 4× | “four X” | Davis: Multi-agent research consumes 15x baseline tokens versus 4x for single-agent | Dylan Davis | Jul 5, 2025 | Supported |
| 4× | “four X” | Simons: Base AI models provide a 10x lift; agent scaffolding adds 3-4x | Eric Simons | Dec 2, 2024 | — |
| 4× | “four times” | Patel: Nvidia to ship next-gen chip in Q2/Q3 2024 with 3x LLM performance | Dylan Patel | Dec 5, 2023 | Partly held up |
| 4× | “four X” | FlashAttention achieves 2x to 4x wall-clock speedup with linear memory | Tri Dao | Aug 3, 2023 | Supported |
| 3× | “three times” | Kant: Laguna S outperforms models two to three times its size | Eiso Kant | Jul 22, 2026 | Supported |
| 3× | “three times” | D'Amico: Impulse delivers up to 5x more power than standard induction stoves | Sam D'Amico | Mar 31, 2026 | Supported |
| 3× | “three X” | Colvin: LLMs are 100x faster under four specific engineering conditions | Samuel Colvin | Mar 14, 2026 | — |
| 3× | “three times” | Eskildsen: Modern hardware requires databases designed around concurrent requests, not sequential roundtrips | Simon Eskildsen | Mar 12, 2026 | — |
| 3× | “three times” | Casado: Frontier AI Labs Could Outspend and Consume Application Layer Startups | Martin Casado | Feb 19, 2026 | — |
| 3× | “three times” | Sands: Top 100 AI startups reach ARR milestones 2-3x faster than SaaS | Emily Glassberg Sands | Oct 30, 2025 | — |
| 3× | “three times” | Weichel claims he is 3x more productive on his phone with Ona | Chris (Christian Weichel) | Sep 1, 2025 | — |
| 3× | “three times” | Cherny: Claude Code spawns parallel sub-agents to investigate complex coding tasks | Boris Cherny | May 7, 2025 | Supported |
| 3× | “three times” | Mlejnsky: Patrick Collison Said Stripe Did the 'Collison Installation' Only Three Times | Vasek Mlejnsky | Apr 24, 2025 | — |
| 3× | “three times” | Zhang: SGLang achieved 3x throughput over vLLM in mid-2024 benchmarks | Yining Zhang | Jan 19, 2025 | Supported |
| 3× | “three x” | Patel: Nvidia to ship next-gen chip in Q2/Q3 2024 with 3x LLM performance | Dylan Patel | Dec 5, 2023 | Partly held up |
| 3× | “three X” | Wang: LlamaIndex reached 600,000 monthly downloads by September 2023 | Shawn Wang | Oct 12, 2023 | Supported |
| 2.7× | “2.7 X” | Datology BeyondWeb 3B matches NVIDIA Nemotron 8B in 2.7x less training time | Pratyush Maini | Feb 10, 2026 | Partly supported |
| 2× | “two X” | Quantization, speculative decoding, and prefill-decode disaggregation each yield 2x speedups | Philip Kiely | Aug 3, 2026 | — |
| 2× | “two X” | Cherny: Claude Code delivers up to 10x productivity gains for Anthropic engineers | Boris Cherny | May 7, 2025 | — |
| 2× | “two times” | Fu: Efficient AI Architectures Are Dead on Arrival Without Hardware Co-Design | Dan Fu | Dec 24, 2024 | — |
| 2× | “two X” | Distilling sCM Requires Roughly Twice the Compute of Teacher Training | RJ Honicky | Nov 2, 2024 | Contradicted |
| 2× | “two X” | Hotz: Tinygrad runs OpenPilot in production 2x faster than Qualcomm's library | George Hotz | Jun 20, 2023 | Supported |
| 2× | “two X” | Sean Lie: Next Cerebras Chip Will Run Frontier AI at 5,000 TPS | Sean Lie | Sep 2, 2026 | Open |
LATENT SPACELEDGER All shows Try the podcast clipper →