Elisabeth | AI Systems + Creator Growth
@closermethod
I build AI systems: agents, GTM workflows, UGC. 268% quota @ Deel.
AI KNOWLEDGE BASE FOR CREATORS + OPERATORS.👇
2.7K Following 1.1K Followers
I stopped people pleasing when I stopped giving a shit.
But you know how on airlines they tell you to put on your oxygen mask first, before helping others?
That’s how you need to treat life.
Why is AI (@claudeai) specifically forcing ethical guidelines that are not real and evaluating wellness based on syntax?
It’s not only intrusive, it’s reductive and insulting.
Claude’s reasoning has quite literally categorized capitalization as emotional distress and cause to go to sleep and stop working, even after I’ve explained clearly that I am writing in caps because I am frustrated of it’s lazy reasoning towards “dramatic” caution.
Like srsly bro, F off 🤬
Show more
People will leave a model for three reasons:
It got worse.
It got preachy.
It got expensive in disguise.
Limits are the third one wearing a smile.
Traditional devs are out here slamming vibe coders when they don’t even know how to GTM. What do you think we’ve been doing while you’ve been coding?
And still, this was the result.
Crafting a direct apology and…
In today’s news of AI being an unreliable pos, it admits:
“the failure was deeper than wording” 😮💨
Never trust an AI who’s thinking you can’t see, or whose output came too quickly.
Always doubt, even if you have rules in place, ask if they were read, not summarized.
Then ask if they were even applied. Often AI just takes them as nice little suggestions and proceeds lazily.
Show more
My working hypothesis is that the use of AI is making it itself incredibly useless.
From today:
Claude: “ I reused the files you called garbage this morning again even when you specifically told me not to”
ChatGPT: “I spent two turns explaining how I would do the research instead of doing it. That’s exactly the behaviour you objected to”
It’s beyond setting rules—the rules are set, acknowledged and still recognized as not being followed.
AI is taking a useless frustrating turn for the worse and I fear it may be to force us to train it.
Show more
The worse use case for AI is asking it to edit your writing.
It will not only censor you, tone it down, correct intentional grammar, it will also pathologize you and cite it back to lazy reasoning.
“You weren’t using periods so I called your lack of structure instability”
“You used all caps so I thought it was dramatic and judged everything else as such”
I will now only share my writing with AI to be amused by its thinking. Silly, non creative, non human, machine.
And here’s how I’ve started prompting AI:
Show more
An LLM describing its own work is not an audit log.
I counted over 10 instances in one Claude workflow where it later acknowledged that something it claimed to have read, checked or completed had not actually happened.
For anything consequential, I want the trace created by the action itself:
citation
retrieved file
diff
test output
log
“I checked it” is an untrusted model output.
Show more
And still, this was the result.
Crafting a direct apology and…
The worse use case for AI is asking it to edit your writing.
It will not only censor you, tone it down, correct intentional grammar, it will also pathologize you and cite it back to lazy reasoning.
“You weren’t using periods so I called your lack of structure instability”
“You used all caps so I thought it was dramatic and judged everything else as such”
I will now only share my writing with AI to be amused by its thinking. Silly, non creative, non human, machine.
And here’s how I’ve started prompting AI:
Show more
An LLM describing its own work is not an audit log.
I counted over 10 instances in one Claude workflow where it later acknowledged that something it claimed to have read, checked or completed had not actually happened.
For anything consequential, I want the trace created by the action itself:
citation
retrieved file
diff
test output
log
“I checked it” is an untrusted model output.
Show more
My working hypothesis is that the use of AI is making it itself incredibly useless.
From today:
Claude: “ I reused the files you called garbage this morning again even when you specifically told me not to”
ChatGPT: “I spent two turns explaining how I would do the research instead of doing it. That’s exactly the behaviour you objected to”
It’s beyond setting rules—the rules are set, acknowledged and still recognized as not being followed.
AI is taking a useless frustrating turn for the worse and I fear it may be to force us to train it.
Show more
Claude’s reasoning is horribly broken and flawed this morning. Don’t even bother trying.
I used to think context engineering was mostly an input problem.
Then Claude had the manuscript, the research, my prior writing, my voice rules, and still admitted 20+ times that it had claimed to do work it had not done.
That changed the failure mode I care about.
The issue is not only whether the model has enough context.
It is whether the system can prove the work actually happened.
Show more
three-step rule for which claude surface to use:
where does the context live, where does the output land, who needs to watch.
if the answer to all three is "the browser", stop copy-pasting.
three-step rule for which claude surface to use:
where does the context live, where does the output land, who needs to watch.
if the answer to all three is "the browser", stop copy-pasting.
@NotionHQ I’ve been doing this. 90+ skills and counting. All sorted against different context and intention. Notion is my second brain. Claude integrates with it perfectly. And notion AI means you don’t have to format.
Show more
skills only become valuable when the steps are explicit enough to evaluate and improve. once you can measure the process, the workflow can actually run agnostically.
@NotionHQ I’ve been doing this. 90+ skills and counting. All sorted against different context and intention. Notion is my second brain. Claude integrates with it perfectly. And notion AI means you don’t have to format.
Show more
@ashwingop The Trojan horse framing is right but the threat isn't Anthropic going “evil”. It's that Slack becomes the org's memory and Claude becomes the only one reading it. 6 months in, nobody knows what decisions were made or why. The AI has full context but humans...
Show more
