Le créateur de ChatGPT vient de sortir une IA qui ne parle pas (ça change tout)

Le créateur de ChatGPT vient de sortir une IA qui ne parle pas (ça change tout)

The creator of ChatGPT just released an AI that doesn't talk (it changes everything)

Title history
  • Le créateur de ChatGPT vient de sortir une IA qui ne parle pas (ça change tout) — observed on 20261112-0057 UTC (YouTube)

by Thomas Berton · View the channel on PSI

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Thomas Berton 👥 48K 📅 September 20, 2026 ⏱ 15 min 👁 14K 📄 expert opinion 🧭 2026-11-12
Available in: English (current) Français

Keywords

JevTypeSafesystem 1decision-makingOpenRouter

Summary

The video introduces Jev, a new AI model from TypeSafe, founded by a former OpenAI researcher. Unlike typical LLMs (ChatGPT, Claude, Gemini) that generate text, Jev is a ‘system 1’ model that outputs only decisions (choice, score, null). It is extremely fast (e.g., classifies 50 emails in 19 seconds vs 131 for Claude Sonnet and 81 for GPT-5.6) and much cheaper (about 81x cheaper than Claude Sonnet and 52x cheaper than GPT-5.6). The presenter demonstrates how to access Jev via OpenRouter, bypassing the waitlist. He runs a comparative test, showing Jev achieves 186/200 correct classifications with a median latency of 350ms. He discusses limitations (slightly more errors, but acceptable given cost/speed trade-off) and numerous business use cases: email triage, lead filtering, customer support routing, live moderation. He also mentions combining Jev with his ‘Jarvis’ assistant. The video concludes that Jev is not a ChatGPT replacement but a complementary tool for high-volume decision tasks.

153 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers substantial value by showcasing a novel AI paradigm with concrete performance metrics from a hands-on test. The argument is well-structured: it clearly explains Jev’s unique characteristics, compares it with leading LLMs on a real task, and provides transparent cost and speed data. The presenter honestly acknowledges Jev’s limitations and does not overclaim its capabilities, correctly positioning it as a specialized decision-maker rather than a general conversational AI. The argument for its usefulness in high-throughput classification scenarios is convincing, supported by the measured results.

Scientific Rigor, Source Quality, Title Accuracy

The presenter references official sources (TypeSafe documentation, OpenRouter) and provides links in the description. However, the test methodology is not peer-reviewed, and the sample size is limited (50 emails, 200 classifications). The accuracy of the claimed cost reductions is based on a single run and may vary in different contexts. The title is accurate in describing the AI’s non-verbal nature, but the ‘changes everything’ claim is promotional. Overall, the information is presented transparently with clear disclaimers about limitations.

179 words

Title / Content Match

The title accurately highlights the unique aspect of Jev (no text output) and its potential impact, though 'changes everything' is somewhat hyperbolic. Overall, the content matches the promise.

Quality & Reliability

7/10

The video provides concrete test results and honest discussion of limitations, but the assessment is based on a single informal test and the presenter has commercial interests (affiliate links).

Chapters

Cited Sources

External References

Contribution & Novelties

The video presents an original hands-on evaluation of a newly released AI model (Jev) that operates differently from mainstream LLMs. It provides specific performance and cost data from a real-world task (email classification), which is original content not readily available elsewhere. The presenter also suggests novel applications such as live moderation and real-time decision-making. The discussion of Jev as a ‘system 1’ model adds conceptual clarity.

Pour aller plus loin :

  • Thinking, Fast and Slow — Background on the system 1/system 2 framework used to describe Jev.
  • Jevons paradox — The economic principle that inspired Jev’s name, relevant to the idea that efficiency leads to more consumption.
  • Large language model — Useful for understanding the contrast between Jev and traditional LLMs.

121 words

Radar Profile

The radar scores indicate a well-rounded video with strong information quantity and decent quality, but slightly lower technical depth and overall reliability. The high quantity reflects the extensive coverage, while the moderate quality and reliability scores suggest room for more rigorous testing and scientific validation.

Reliability 7/10