---
title: "Blog: AI: too good to be true, too bad to be useful (2026-06-19)"
type: entity
tags: [blog, stub, rlhf, needs-reingestion]
created: 2026-09-17
updated: 2026-09-17
confidence: low
sources:
  - raw/site/blog-too-good-to-be-true.txt
  - raw/site/typesafe-ai-home.txt
jev_version: "jev-1.13.0"
summary: "Stub: the body of this 2026-06-19 TypeSafe post was not captured (JS-rendered); only title, date, category, and the homepage teaser are known."
---

# Blog: AI: too good to be true, too bad to be useful (2026-06-19)

> **TL;DR** Stub page. The body of this post was **not captured** — `raw/site/blog-too-good-to-be-true.txt` contains only the page chrome, because the article body is rendered client-side. Do not cite anything from this post beyond the metadata and teaser below until it is re-ingested.

## Facts

| Field | Value |
|---|---|
| Title | "AI: too good to be true, too bad to be useful \| TypeSafe AI" |
| URL | https://typesafe.ai/blog/ai-too-good-to-be-true-too-bad-to-be-useful-typesafe-ai |
| Category | Thoughts |
| Date | 2026-06-19 (page shows "Jun 19, 2026") |
| Author | Not captured |
| Body | **Not captured** — JS-rendered; the raw file holds only nav, "Back", category, date, title, and footer |
| Confidence | low |

## Homepage teaser

The only substantive sentence available, from the blog carousel on https://typesafe.ai:

> "RLHF-trained language models please humans and assist rather than make reliable autonomous decisions. What comes next?"

That framing matches the company's other writing — RLHF optimizes for human preference, which yields assistants rather than components software can depend on — but the post's actual argument, examples, and any numbers are unknown. Anything beyond the teaser would be speculation.

## Re-ingestion note

**This page should be rewritten once the body is captured.** The raw capture is the pre-hydration HTML/text of a client-rendered page. To re-ingest, fetch the post with a renderer that executes JavaScript (headless browser) or find a server-rendered variant, replace `raw/site/blog-too-good-to-be-true.txt`, then rewrite this page with: thesis, structure, claims, and links, and raise `confidence`. It is the oldest of the three captured TypeSafe posts, so it likely contains the earliest public statement of the System One thesis (inferred).

## Related

- [[entities/blog-bitterest-lesson]] — 2026-09-10, the right-task argument
- [[entities/blog-antibenchmaxxing]] — 2026-09-11, the measurement argument
- [[entities/blog-introducing-system-one]] — 2026-09-15, the launch
- [[entities/manifesto]] — the same critique of RLHF-shaped AI, in full
- [[entities/typesafe-ai]] — the company

## Sources

- raw/site/blog-too-good-to-be-true.txt (https://typesafe.ai/blog/ai-too-good-to-be-true-too-bad-to-be-useful-typesafe-ai) — body not captured
- raw/site/typesafe-ai-home.txt (https://typesafe.ai/) — teaser sentence