justin․searls․co

Why is OpenAI so stingy with ChatGPT web search?

For however expensive LLM inference supposedly is, OpenAI continues to be stupidly stingy with respect to web searches—even though any GPT 5.2 Auto request (the default) is extremely likely to be wrong unless the user intervenes by enabling web search.

Meanwhile, ChatGPT's user interface offers:

  • No way to enable search by default
  • No keyboard shortcut to enable search
  • No app (@) or slash (/) command to trigger search
  • Ignores personalization instructions like "ALWAYS USE WEB SEARCH"
  • Frequently hides web search behind multiple clicks and taps, and aggressively A/B tests interface changes that clearly will result in fewer searches being executed

All of this raises the question: how does ChatGPT implement search? What is the cost of the search itself and the extent of chain-of-thought reasoning needed to aggregate and discern the extraordinary number of tokens that need to be ingested by those search results?

It's interesting that OpenAI is so eager to goose usage by lighting dumpsters full of venture capital on fire, but is so stingy when it comes to ensuring their flagship product knows basic facts like "iPhone Air is a product that exists."

Early in my career, I met a few COBOL developers who came out of retirement in the run-up to January 2000, getting paid $300+ per hour to remediate Y2K bugs when nobody else was left who knew COBOL.

Suspect a similar trajectory for highly-skilled, well-rounded "pre-AI" engineers

Copied!
3.1k likes ·

Coding agents are a TON of fun if you are (1) extremely ambitious and (2) have middling standards.

They are still fun if you are (1) extremely ambitious and (2) ruthlessly exacting standards, but markedly less so.

Copied!
2.7k likes ·

Now that I've spent ten hours with Claude Code after a few months with Codex CLI, I can say with confidence:

  • Claude is much much faster
  • Claude makes much stupider mistakes much more often, even with Opus 4.5
  • With either agent, I end each session frustrated and exhausted
Copied!
187 likes ·

LOL, also apparently Adobe Premiere on iPad will frequently silently fail when generating captions. And there's no way to export them as text, subtitle files, etc.

This is starting to seem like bad software that nevertheless gets recommended to people ceaselessly.

Copied!
248 likes ·

Trying Adobe Premiere for the first time since version 6.0 in 2002. Paid for Creative Cloud Pro. First thing I tried: start a project on iPad, sync via cloud, finish on my Mac.

LOL, nope. Their "cloud" can't sync projects. It's just a one-way, manual upload and import. And slow.

Copied!
260 likes ·

Over the last 20 years, my time in Japan felt like Season 1 of Pluribus. A nation acting in harmony, going from happy to see me everywhere I went to constantly signaling they "need some space"

In cities, it's thanks to over-tourism. In the country, it's anti-immigrant populism.

Copied!
1.2k likes ·

Tooling for coding agents is overly focused on scaling numerous parallel workers instead of ensuring correctness. That continues to be where all my time goes and is the real barrier to scaling up. (e.g., Why am I exploratory testing this UI when vision models could be doing it?)

Copied!
3.2k likes ·

This Opus 4.5 failure mode rarely happens with Codex: Claude copy-pasted a source listing by mistake, edited that instead of the correct file, declared success. When I pointed out its error, it confidently asserted: "The code is correct - this is a browser caching issue." 🖕

Copied!
2.2k likes ·

After hearing lots of Opus 4.5 hype, I decided to switch to Claude Code this month from Codex CLI (with which I have a separate set of frustrations), and I've learned two things:

  1. Claude Code is significantly more productive (parallel by default) and generally smarter (smarter coding) than it was in August

  2. When literally anything goes wrong, it routinely takes the path of least resistance (making excuses, declaring victory when things don't work, reaching for workarounds when basic reason points elsewhere)

Disappointed overall. Both GPT-5.2 and Opus 4.5 (and their CLI's surrounding chrome) feel barely improved since last summer.

Copied!
2.5k likes ·
1
2
3
4
5
6
77