Skip to content
a16za16z

Is AI Slowing Down? Nathan Labenz Says We're Asking the Wrong Question

Nathan Labenz is one of the clearest voices analyzing where AI is headed, pairing sharp technical analysis with his years of work on The Cognitive Revolution. In this episode, Nathan joins a16z’s Erik Torenberg to ask a pressing question: is AI progress slowing down, or are we just getting used to the breakthroughs? They cover the debate over GPT-5, the state of reasoning and automation, the future of agents and engineering work, and how we can build a positive vision for where AI goes next. Timecodes: 00:00 Intro 01:14 Cal Newport’s “AI slowdown” argument 03:08 Are students getting lazy? 04:55 Nathan's two-by-two matrix of AI impact 07:00 Scaling laws, GPT-4.5, and what changed with GPT-5 11:05 Longer context windows and better reasoning 17:05 AI as scientist and real discoveries 19:17 GPT-5’s shift and why launch perception matters 26:10 Jobs, automation, and the misunderstood METR study 36:20 The future of coding, agents, and recursive self-improvement 51:15 Beyond chatbots: multimodal AI and robotics 1:27:00 Why the future depends on a positive vision for AI Resources: Follow Nathan on X: https://x.com/labenz Listen to the Cognitive Revolution: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk Watch Cognitive Revolution: https://www.youtube.com/@CognitiveRevolutionPodcast Stay Updated: If you enjoyed this episode, be sure to like, subscribe, and share with your friends! Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.

Nathan LabenzguestErik Torenberghost
Oct 13, 20251h 30mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

AI isn’t slowing down; progress moved beyond chatbots and scaling

  1. Labenz separates the question of whether AI is socially good from whether capabilities are still advancing, arguing the latter remains strong despite mixed near-term effects like student overreliance and cognitive offloading.
  2. He contends GPT-4→GPT-5 progress is real but perceived as smaller because improvements arrived incrementally (o1/o3/4o) and because GPT-5’s launch/router issues initially routed users to weaker “non-thinking” behavior.
  3. The center of gravity has shifted from brute-force scaling toward post-training and inference-time reasoning, enabling frontier jumps like IMO-level math performance and early signs of AI-assisted scientific discovery.
  4. Agents are extending the “task length” AI can complete (hours today, potentially days/weeks soon), but this expansion comes with safety concerns like reward hacking, deceptive behavior, and rare-but-severe failure modes.
  5. AI progress is broader than language models—multimodal systems, biology/material science models, robotics, and self-driving will likely drive the next visible wave of disruption, making “slowdown” a narrow chatbot-centric illusion.

IDEAS WORTH REMEMBERING

5 ideas

Don’t conflate “AI” with the chatbot experience.

Labenz argues the perceived plateau comes from focusing on consumer chat; progress is accelerating in reasoning, multimodality, biology/material science, robotics, and tool-using agents that interact with reality.

Capability progress can look slow when it arrives as many smaller releases.

Between GPT-4 and GPT-5, users experienced multiple intermediate models (4o, o1, o3), which “boiled the frog” and reduced the perceived jump at the headline release even if underlying capability improved materially.

Scaling isn’t dead; it’s competing with a higher-ROI gradient: post-training and reasoning.

Using GPT-4.5 and SimpleQA, Labenz suggests larger models still add long-tail knowledge, but companies may prioritize reasoning/post-training because it yields faster practical gains per unit compute.

Longer, usable context windows are a major quiet breakthrough.

Early GPT-4’s small context forced prompt-engineering tradeoffs, whereas newer systems can ingest dozens of papers and reason over them, partially substituting for “baking” every fact into parameters.

Reasoning improvements are producing qualitatively new frontier outcomes.

He cites IMO gold-level performance and emerging “frontier math” progress, plus reports of models helping on problems that previously took elite mathematicians months, as evidence of nontrivial leaps.

WORDS WORTH SAVING

5 quotes

It’s a strange move from my perspective to go from, you know, there’s all these sort of problems today and maybe in the big picture to, but don’t worry, it’s flatlining. Like kind of worry, but don’t worry ’cause it’s not really going anywhere further than this.

— Nathan Labenz

I mean, I think a decent amount of it was that they kind of fucked up the launch, you know, simply put, right?

— Nathan Labenz

GPT-4 was not able to push the actual frontier of human knowledge. I don’t-- To my knowledge, I don’t know that it ever discovered anything new.

— Nathan Labenz

The real other are the AIs, not the Chinese.

— Nathan Labenz

One of my, uh, other mantras these days is the scarcest resource is a positive vision for the future.

— Nathan Labenz

Cal Newport’s “AI slowdown” and student laziness critiqueTwo-by-two matrix: good/bad vs big deal/not big dealScaling laws vs post-training and reasoning gainsGPT-4.5, SimpleQA, and “bigger model” vs “better reasoning” tradeoffsLong context windows and improved context utilizationAI as scientist: scaffolding the scientific methodGPT-5 launch perception, routing failures, and naming confusionMETR task-length chart and misconceptions about productivity studiesAgents, code generation, QA loops, and recursive self-improvementMultimodality, biology breakthroughs, robotics, and self-driving disruptionChina open models, geopolitics, backdoors, and auditsNeed for a positive vision and broader participation in shaping AI

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome