GPT-4

Product · Tech

262

Mentions

Episodes

192

Claims

GPT-4, mentioned 262 times across podcast episodes and expert conversations analyzed by Sonic.

What GPT-4 has said

AI scaling laws have held true over approximately 15 orders of magnitude of compute, from early models documented by Kaplan to GPT-4.

bullish·Leopold Aschenbrenner

The per-token inference cost of GPT-3 at its launch was higher than the current per-token cost of the more capable GPT-4.

bullish·Leopold Aschenbrenner

The performance gap between the initial and current versions of GPT-4 is larger than the gap between Claude 3 Opus and Claude 3 Haiku, which have a 60x price difference.

bullish·Leopold Aschenbrenner

AI scaling laws have held true over approximately 15 orders of magnitude of compute, from early models documented by Kaplan to GPT-4.

Expert perspectiveLeopold AschenbrennerJul 28

The per-token inference cost of GPT-3 at its launch was higher than the current per-token cost of the more capable GPT-4.

Expert perspectiveLeopold AschenbrennerJul 28

The performance gap between the initial and current versions of GPT-4 is larger than the gap between Claude 3 Opus and Claude 3 Haiku, which have a 60x price difference.

Expert perspectiveLeopold AschenbrennerJul 28

Post-training improvements to GPT-4 since its launch have resulted in a performance gain of about 100 ELO points on the LMSys leaderboard.

Expert perspectiveLeopold AschenbrennerJul 28

The training cluster for GPT-4 was rumored to consist of 25,000 NVIDIA A100 GPUs, cost approximately $500 million to build, and consumed roughly 10 megawatts of power.

Expert perspectiveLeopold AschenbrennerJul 28

AI scaling laws have held true over approximately 15 orders of magnitude of compute, up to the scale of GPT-4.

Expert perspectiveLeopold AschenbrennerJul 28

The per-token inference cost of GPT-3 at its launch was higher than the current inference cost of GPT-4.

Expert perspectiveLeopold AschenbrennerJul 28

The performance gap between Claude 3 Opus and Claude 3 Haiku is smaller than the post-launch improvement of GPT-4, despite a 60x price difference between the Claude models.

Expert perspectiveLeopold AschenbrennerJul 28

Post-launch improvements to GPT-4 resulted in a performance gain of approximately 100 ELO points on the LMSys leaderboard.

Expert perspectiveLeopold AschenbrennerJul 28

The training cluster for GPT-4 was rumored to consist of 25,000 A100 GPUs, cost approximately $500 million, and consume around 10 megawatts of power.

Expert perspectiveLeopold AschenbrennerJul 28

The performance of OpenAI's o3 model on ARC-AGI demonstrates that architectural changes, not just increased compute on existing architectures like GPT-4, are necessary for breakthroughs in AI capabili...

Expert perspectiveFrançois CholletJul 28

Performance on the ARC-AGI-1 benchmark improved from 0% with GPT-3 in 2020 to 5% with GPT-4o in 2024.

Expert perspectiveFrançois CholletJul 28

OpenAI has made significant progress on reducing hallucinations in GPT-4, but the model is not yet at the required level of reliability.

Expert perspectiveMira MuratiJul 28

The definition of AGI is a moving target, and systems with the capabilities of GPT-4 or GPT-5 would have been considered AGI by many people a decade ago.

Expert perspectiveSam AltmanJul 28

AI models like DALL·E 3 and GPT-4 have demonstrated that creativity was an easier capability for AI to develop than was previously expected.

Expert perspectiveSam AltmanJul 28

The most persuasive AI models in a research study were Opus 4.1 and Opus 4.6, followed by models from OpenAI (GPT-4o, GPT-5.4), Google (Gemini 2.5 Pro), and xAI (Grok 4.20).

Expert perspectiveJack ClarkJul 28

A key breakthrough for AGI would be combining the knowledge acquisition and linguistic skills of GPT-4 with the planning ability of AlphaGo.

Expert perspectiveYoshua BengioJul 28

An EPFL study demonstrated that GPT-4 has superior persuasion abilities compared to ordinary humans when given access to a target's Facebook page.

Expert perspectiveYoshua BengioJul 28

The best fully open language models are currently approaching the performance levels of the original GPT-4.

Expert perspectiveNathan LambertJul 28

The Chinese model Yi-Large is considered a GPT-4-class model on the LMSys leaderboard, but it was released over a year after OpenAI's GPT-4.

Expert perspectiveLeopold AschenbrennerJul 28

Create a free account to see GPT-4's full intelligence report - every claim, the relationship network, and AI Q&A across all sources. No card needed.

Get started free

Back to Entities Entity Detail