scripod.com
Jev: System One models for Prod, not God — with Diogo Almeida, CEO, TypeSafe AI

Overview

Shownote

Highlights

Transcript

Chapters

Pins

Jev: System One models for Prod, not God — with Diogo Almeida, CEO, TypeSafe AI

Latent Space: The AI Engineer Podcast

1 DAYS AGO
Latent Space: The AI Engineer Podcast

Latent Space: The AI Engineer Podcast

1 DAYS AGO
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

Tickets for AIE NYC now open, and apply for the invite-only AIE CODE. Join us! We have an unusual relationship with today’s guest: for years since coauthoring the InstructGPT paper, Diogo Almeida had been saying that API-available frontier models have bee...

Highlights

Diogo Almeida explains why TypeSafe built Jev as a programmable System 1 model for software rather than another chatbot, and why reliability, calibration, composability, and intelligence per dollar matter more than benchmark scores.
01:05
The AI economic revolution is back on the table
03:23
A new class of models
06:11
RLHF can erase minority modes
12:02
Refusals are a type error
20:53
The right task beats more compute
27:33
RLCD defines a new direction for AI
30:16
AI is smart, but not yet connected to the economy
42:23
Robustness matters more than determinism
51:00
Jev optimizes for intelligence per dollar
56:00
Jev introduces a new programming vocabulary
1:01:17
System messages are global variables
1:18:43
AI should disappear into the background
1:34:11
Jev Turns Voice into Computer Control
1:40:10
Jev could make coding agents weird
1:43:52
Pacing debates assume too much
1:49:33
I wouldn't pre-train with a billion dollars
1:58:24
Code, not people, would consume AI APIs
2:02:35
Research needs the right task
2:06:32
Don’t mislead people for the greater good
2:09:25
Free coding agents from the KV cache
2:12:36
The future of agent swarms
2:19:50
More shapes of intelligence

Chapters

Jev Launch Week and the AI Economic Revolution
00:00
What Is Jev? System One Models and Programmable AI
02:50
RLHF, Mode Collapse, Calibration, and Yann LeCun
05:54
Programmatic AI, Refusals, and Safety Alignment
10:29
Why TypeSafe Rejects Public Benchmarks
17:21
The Bitterest Lesson: Data, Compute, and the Right Task
20:43
RLCD vs. RLHF and RLVR
24:59
Why Powerful AI Still Hasn’t Automated the Economy
28:42
Reliability, Robustness, and Determinism
39:55
Model Versioning, LTS, Speed, and Intelligence per Dollar
48:11
Inside Jev’s API and Programming Primitives
54:04
How to Build with Jev: Structure, Decomposition, and Small Decisions
58:28
The Inverse SaaS-pocalypse and AI Disappearing into Software
1:18:28
Computer Use, Dark Data, and Jev’s Biggest Use Cases
1:33:21
How Jev Could Reshape Coding Agents
1:38:48
AI Safety, Frontier Pacing, and the Limits of RLVR
1:41:00
Why Diogo Wouldn’t Pre-Train with $1 Billion
1:48:03
The OpenAI Story Behind TypeSafe
1:55:19
Why Diogo Thinks Most Neo-Labs Are Getting AI Wrong
2:01:41
Coding Agents Beyond the KV Cache and the Multi-Agent Future
2:08:00

Transcript

Speaker 2: Okay, we're in the studio, a special occasion because this week, Diogo, my good buddy, launched Jev and it's been taking over the complete timeline. How do you feel? What's it like to be you right now? Diogo Almeida: Emotionally, never been wor...