Skip to content
Y CombinatorY Combinator

The 10 Trillion Parameter AI Model With 300 IQ

Earlier this month, OpenAI raised the largest venture round ever at $6.6 billion. The company’s CFO says AI is now at the point where orders of magnitude matter and the next generation of models will be capital intensive. In this episode of the Lightcone, the hosts consider what a world with ultra intelligent models would look like and what potential unlocks could be made possible. Apply to Y Combinator: https://ycombinator.com/apply Chapters (Powered by https://bit.ly/chapterme-yc) - 0:00 Coming Up 0:54 What models get unlocked with the biggest venture round ever? 5:35 Some discoveries take a long time to actually be felt by regular people 9:53 Distillation may be how most of us benefit 14:26 o1 making previously impossible things possible 21:17 The new Googles 23:47 o1 makes the GPU needs even bigger 25:44 Voice apps are fast growing 27:05 Incumbents aren’t taking these innovations seriously 31:52 Ten trillion parameters 33:15 Outro

Garry TanhostDiana HuhostHarj Taggarhost
Oct 31, 202433mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Ten-Trillion-Parameter AI: From 300 IQ Models To Startup Goldrush

  1. The episode explores what ultra-large models (imagined at 10 trillion parameters and ~300 IQ) and OpenAI’s new o1 model mean for founders, enterprises, and the broader economy.
  2. Hosts argue current frontier models already rival typical knowledge workers and that o1-like breakthroughs may unlock previously impossible use cases while making AI behavior more deterministic and reliable.
  3. They discuss the likely role of huge “teacher” models and distillation into cheaper models, the rapidly shifting LLM and coding-tool market (e.g., Cursor vs GitHub Copilot), and early signs of real business impact from automation and AI voice agents.
  4. The conversation closes on a bullish vision where massively superhuman AI can digest humanity’s scientific output, accelerating discovery toward radical technologies, while competitive dynamics keep UX-focused startups central to capturing value.

IDEAS WORTH REMEMBERING

5 ideas

Ultra-large models may act as expensive “teachers,” not everyday workhorses.

10T-parameter models will likely be too slow and costly for routine use, but can train smaller distilled models that deliver most of the capability at a fraction of the price, similar to Meta’s large 405B teacher improving its 70B model.

Developer market share is fragmenting; OpenAI no longer has automatic dominance.

Within recent YC batches, usage has diversified significantly—Claude moved from ~5% to ~25% of companies, LLaMA from 0% to 8%, showing founders will freely switch to better or cheaper models and tools.

o1-level reasoning can make previously non-viable AI products suddenly viable.

Startups report big jumps in accuracy (e.g., ~80% to ~99%) simply by swapping models, unlocking mission-critical and regulated use cases that were too risky when LLMs were less reliable.

As AI becomes more deterministic, competitive advantage shifts back to classic software execution.

If prompt wrangling and fragile workflows matter less, winners will be those who excel at UX, domain depth, sales, and integration—AI becomes infrastructure, and moats look more like traditional SaaS moats.

Voice-based AI is reaching a ‘works-in-practice’ inflection and threatens call centers.

With real-time voice APIs priced around $9/hour and much lower latency, AI can now handle debt collection, logistics coordination, and support calls at human-like quality and similar or lower cost.

WORDS WORTH SAVING

5 quotes

You mean they're going to capture a light cone of all future value?

— Host (on OpenAI’s potential dominance with o1)

You could make a strong case that AGI is basically already here.

— Host (on current state-of-the-art models and knowledge work)

It took 150 years until the average Joe could feel the Fourier transform.

— Diana (on long lags between foundational discoveries and everyday impact)

At this point, AI has passed Turing tests and is solving all of these very menial problems over the phone.

— Host (on the new generation of AI voice agents)

What this might be is not merely a bicycle for the mind. It might actually be a self-driving car, or, even crazier, maybe a rocket to Mars.

— Host (on how powerful AI could transform human capability)

OpenAI’s massive fundraise, scaling laws, and the path to 10T-parameter modelsCapabilities and implications of OpenAI’s o1 model for buildersTeacher–student model dynamics and distillation for cheaper, faster inferenceShifting developer preferences across LLMs (OpenAI, Claude, LLaMA, Gemini) and coding IDEs (Cursor vs Copilot)Real-world productivity gains from AI in startups and mature companiesVoice agents, real-time AI, and the disruption of call-center and service workLong-term societal and scientific impacts of potential artificial superintelligence

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