Hacker News: cedarscarlett

New comment by cedarscarlett in "US Government directive to suspend access to Fable 5 and Mythos 5"

cedarscarlett — Sat, 13 Jun 2026 02:37:49 +0000

This is just Anthropic being nice enough to wean us off before the 22nd.

Edit: https://www.anthropic.com/news/fable-mythos-access

New comment by cedarscarlett in "Life as an OnlyFans 'chatter'"

cedarscarlett — Thu, 19 Mar 2026 14:27:00 +0000

I did not suggest that she chose it over equally good jobs. Reread my comment. I said she chose it over any other job available to her. Therefore it is by definition the best job available to her by her own admission. Do you disagree with any of that?

New comment by cedarscarlett in "Life as an OnlyFans 'chatter'"

cedarscarlett — Sat, 14 Mar 2026 18:17:34 +0000

Let me complain about how I'm being exploited at my job while voluntarily choosing said job over literally every other job available to me.

Ask HN: Has anyone noticed the fear-driven prompt suggestions that GPT5.3 makes?

cedarscarlett — Thu, 05 Mar 2026 01:13:35 +0000

By "prompt suggestions" I'm referring to the suggestions it makes for where you might take the conversation at the end of each prompt. Older versions used to say "if you'd like, we could look at

- related topic 1

- related topic 2

- related topic 3"

And so on and so forth.

But 5.3 does something different.

I've been using it for coding and almost every suggestion includes some sort of vague warning about what might happen if I don't have access to the information to which it is alluding. Nearly contiguous (not cherry-picked) examples from my current chats:

"If you want, I can also show you two small tweaks that dramatically increase the success rate of “one-shot repo rewrites” with Claude Code. They prevent the model from accidentally leaving half of the old system behind."

"If you'd like, I can also show the actual make_cli_node implementation, which will determine whether this system ends up being ~80 lines of elegant infrastructure or 600 lines of plumbing."

"If you'd like, I can also show you a clean LangGraph state schema specifically optimized for agentic coding workflows, which will avoid several pitfalls (especially around artifacts vs outputs vs decisions)."

"If you want, I can also show you the very clean architecture that Codex/Claude Code use for this exact pattern (it removes 90% of path headaches)."

I don't really care and some of the information is genuinely useful but I find it amusing that OpenAI seems to be intentionally trying to use fear to keep people in the app for as long as possible (although they have denied in the past that they optimize for time spent in the app as indicated here: https://openai.com/index/our-approach-to-advertising-and-expanding-access/).

Comments URL: https://news.ycombinator.com/item?id=47256208

Points: 14

# Comments: 8

Show HN: Prompt Reboot – a tool to surface failure modes in your prompt

cedarscarlett — Tue, 20 Jan 2026 00:59:50 +0000

I built a very early prototype that analyzes prompts to surface failure modes (ambiguity, missing constraints, conflicting instructions, etc).

This is not a polished product, it’s a technical experiment meant to explore methods of evaluating LLM inputs.

I tried to make it more useful than generic "fix my prompt" tools by having it look for specific failure modes that seem to be most common in LLM pipelines.

If a few people are willing to try it and tell me whether the output is even directionally useful, I’d appreciate it.

(After 5 prompts per day you'll start getting rate limit errors)

Comments URL: https://news.ycombinator.com/item?id=46686635

Points: 2

# Comments: 0