Skip to content

Jev CEO: I made ChatGPT, now I'm building what's next

By AI Engineer

18 min video·en··265516 views

This is an AI-generated summary of “Jev CEO: I made ChatGPT, now I'm building what's next” — a 18 min YouTube video by AI Engineer, published July 31, 2026. It condenses the full transcript into 9 key takeaways with clickable timestamps.

Summary

Tiago Almeida, a co-author of OpenAI's major models, argues that the current AI era, defined by RLHF, excels at human-in-the-loop assistance but is fundamentally flawed for reliable automation, necessitating a new approach for truly smarter software.

Key Points

  • The speaker, a co-author of GPT-4 and ChatGPT, acknowledges ChatGPT's world-changing impact but critically highlights its inherent limitations stemming from its underlying algorithms. 
  • The AI field is currently split between two extreme views: one that AI is progressing insanely well, surpassing human benchmarks, and another that it's a bubble generating little real value. 
  • The current 'ChatGPT era' is fundamentally an 'assistance era,' where AI primarily serves as a human-in-the-loop tool, and even advanced applications like Claude Code remain within this paradigm. 
  • This divide is explained by current AI's exceptional capability in 'assistance' tasks, which aim to please a human in the loop, versus its poor performance in 'automation' tasks, which seek to remove human involvement. 
  • Reinforcement Learning from Human Feedback (RLHF) is the foundational algorithm for virtually all modern Large Language Models, explicitly designed to collect and optimize for human preferences. 
  • RLHF models are inherently prone to overpromising and prioritizing human engagement, often appearing correct even when wrong, because their objective is human preference, not calibrated factual accuracy or autonomous task completion. 
  • The speaker contends that the next logical evolution for AI is to move beyond assistance towards 'real automation' and the development of 'smarter software' that can perform tasks reliably and autonomously. 
  • The speaker's company, TypeSafe, is actively working on this new AI stack, aiming to enable actual automated work and smarter software by optimizing for reliable decision-making. 
  • Achieving true automation requires a fundamental redesign of the AI stack, shifting focus from optimizing for human preference to prioritizing reliability and calibrated decision-making. 
Jev CEO: I made ChatGPT, now I'm building what's next

Jev CEO: I made ChatGPT, now I'm building what's next

Tiago Almeida, a co-author of OpenAI's major models, argues that the current AI era, defined by RLHF, excels at human-in-the-loop assistance but is fundamentally flawed for reliable automation, necessitating a new approach for truly smarter software.

Key Points

—The speaker, a co-author of GPT-4 and ChatGPT, acknowledges ChatGPT's world-changing impact but critically highlights its inherent limitations stemming from its underlying algorithms.
—The AI field is currently split between two extreme views: one that AI is progressing insanely well, surpassing human benchmarks, and another that it's a bubble generating little real value.
—The current 'ChatGPT era' is fundamentally an 'assistance era,' where AI primarily serves as a human-in-the-loop tool, and even advanced applications like Claude Code remain within this paradigm.
—This divide is explained by current AI's exceptional capability in 'assistance' tasks, which aim to please a human in the loop, versus its poor performance in 'automation' tasks, which seek to remove human involvement.
—Reinforcement Learning from Human Feedback (RLHF) is the foundational algorithm for virtually all modern Large Language Models, explicitly designed to collect and optimize for human preferences.
—RLHF models are inherently prone to overpromising and prioritizing human engagement, often appearing correct even when wrong, because their objective is human preference, not calibrated factual accuracy or autonomous task completion.
—The speaker contends that the next logical evolution for AI is to move beyond assistance towards 'real automation' and the development of 'smarter software' that can perform tasks reliably and autonomously.
—The speaker's company, TypeSafe, is actively working on this new AI stack, aiming to enable actual automated work and smarter software by optimizing for reliable decision-making.
—Achieving true automation requires a fundamental redesign of the AI stack, shifting focus from optimizing for human preference to prioritizing reliability and calibrated decision-making.
Summarize any video — free
Summarizer.tube
Copy All
Share Link
Bookmark

Summarize any YouTube video, free

You just read an AI summary of this video. Paste any other YouTube link and get the key points with clickable timestamps in seconds — no signup, 5 free a day.

YouTube Transcript Tool

More Resources

YouTube Video SummarizerLecture Summarizervs Summarize.techAll AlternativesFor StudentsFor ProfessionalsFor Content CreatorsAll Use CasesHow to Summarize YouTube
Or summarize right on YouTube with our free Chrome extension →

More Summaries

1 hr 13 min

I Spent 2 Years Wanting this Team

LuigiToanen

Luigi, determined to break a long losing streak, leads his highly skilled Yellow Spades team through strategic challenges, intense PvP battles, and cunning tactics to ultimately win Season 6 of the Mo