41–50 of 265 posts
I have to disagree that it's good for LLMs to do the research, depending on the context.
If by "useful for research" you mean useful for tracking down sources that you, as the writer, digest and consider, then great.
If by "useful for research" you mean that it will fill in your citations for you, that's terrible. That sends a false signal to readers about the credibility of your work. It's critical that the author read and digest the things they are citing to.
To your point, it's entirely a balance. I personally will record a 10-15 minute yap session on a concept I want to share and feed it to an agent to distill it into a series of observations and more compelling concepts. Then you can use this to write your piece.
When I asked it for alternatives/edits, they were not good however.
I've had projects that seemed tedious or obvious in my head only to realize hidden complexity when trying to put their trivial-ness into written words. It really is a sort of meditation on the problem.
In the most important AI assisted project I've shipped so far I wrote the spec myself first entirely. But feeding it through an LLM feedback loop felt just as transformational, it didn't only help me get an easier to parse document, but helped me understand both the problem and my own solution from multiple angles and allowed me to address gaps early on.
So I'll say: Do your own writing, first.
I don't really understand why people will create blogs that are generated by Claude or ChatGPT. You don't have to have a blog, isn't the point of something like a blog to be your writing? If I wanted an opinion from ChatGPT I could just ask ChatGPT for an opinion. The whole point of a blog, in my mind, is that there's a human who has something that they want to say. Even if you have original ideas, if you have ChatGPT write the core article makes it feel kind of inhuman.
I'm more forgiving of stuff like Grammarly, because typos are annoying, though I've stopped using it because I found I didn't agree with a lot of its edits.
I admit that I will use Claude to bullshit ideas back and forth, basically as a more intelligent "rubber duck", but the writing is always me.
I tend to do extensive research (that process in itself would involve LLMs too, sure) in a tech plan, a product spec, etc. and usually end up with a really solid idea in my head and like say, five critical key points about this tech plan or product spec that I absolutely must convey in this document.
Then I basically "brain dump" my critical key points (including everything about it, background/reasoning, why this or that way, what's counterintuitive about it, why is this point important, etc.) in pretty messy writing (but hitting all the important talking points) to a LLM prompt, asking it to produce the document I need (be it tech plan, product spec, whatever) and ask it to write it based on my points.
The resulting document has all the important substance on it this way.
If you use LLM to produce documents like this by a way of a prompt like "Write a tech plan for the product feature XYZ I want to build", you're going to get a lot of fluff. No substance, plenty of mistakes, wrong assumptions, etc.
I agree with most of this, but my one qualm is the notion that LLMs "are particularly good at generating ideas." It's fair enough that you can discard any bad ideas they generate. But by design, the recommendations will be average, bland, mainstream, and mostly devoid of nuance. I wouldn't encourage anyone to use LLMs to generate ideas if you're trying to create interesting or novel ideas.
That being said I don't think LLMs are idea generators either. They're common sense spitters, which many people desperately need.
The title and of this article is Don't Let AI Write For You , when its point seems to be closer to Don't Let AI Think For You (see "Thinking"). This distinction is important, because (1) writing is not the only way to faciliate thinking, and (2) writing is not neccessarily even the best way to facilitate thinking. It's definitely not the best way (a) for everyone, (b) in every situation. Audio can be a great way to c…
I'm finding AI great to have a conversation with to flesh out ideas, with the added benefit it can summarize everything at the end
Then once I feel like I have addressed all the areas, I ask for a "critical" review, which usually pokes holes in something that I need to fix. Finally have the AI draft up a document (Though you have to generally tell it to be as concise and clear as possible).
Sometimes an LLM can shortcut me through a bunch of those misunderstandings. It feels like an easy win.
But ultimately, lacking context for how I got to that point in the debugging litany always slows me down more than the time it saved me. I frequently have to go backwards to uncover some earlier insight the LLM glossed over, in order to "unlock" a later problem in the litany.
If the problem is simple enough the LLM can actually directly tell you the answer, it's great. But using it as a debugging "coach" or "assistant" is actively worse than nothing for me.