Please do not use an LLM to create a conference talk.
Please.
We want to hear your actual words in your actual voice We do not want diluted words created for you by sparkling auto-complete. We really don't.
The vibe around LLM slop may seem better to you. But it's generally easily seen as slop because it's tasteless, flavorless, and generally content-free.
Some Observations
LLM's Write Crap Outlines
Happy to report that folks who want to present things on building AI agents are often incapable of providing a detailed outline.
Please check your LLM output.
When the LLM says "show 3-4 examples" please actually figure out if you're going to show 3 or 4 examples, and then, actually include something about the examples.
We don't want an idea of what you might present.
We need to know what you will present.
I know, I know. You're using an LLM to spam us with talk proposals and you're waaaay too busy to check the output and make sure it's coherent.
LLM's Can't Fill In Forms
If you’re too lazy to submit conference talk proposals, and are using an LLM to prepare your submissions, be aware of this:
LLMs fail at filling in conference proposal forms.
Please check what slop your LLM generated in what field. The semantics are obvious to people and opaque to sparkling auto complete.
You don’t want your proposal to stand out because the LLM didn’t fill in the submission form properly.
LLM's Foster Bad Design
Last year it was GAN's and RAG's. This year it's MCP. Can folks not pitch talks on complicated workarounds to the known problems with LLM's and FIX THE LLM's?
Larding on more components to make a complicated process more complicated and -- somehow -- mitigate a number of known problems is just plain bad engineering.
I gotta say, the number of talks that can be summarized as "fix these well-known LLM problems by accreting yet more complexity" is astounding. (They seem to use the same LLM, so the submissions all have the same look with slightly different autocomplete text strings.)
I'm feeling very old. Do any of these LLM vibers know about "root cause"? Why can't someone propose a talk that addresses the root cause of these well-known LLM problems?
LLM's Don't Read the Room
Before you sit down for an in-depth chat with an LLM to use sparkling autocomplete to write your conference talk proposal, do this.
Have your LLM examine the previous five years of the conference agenda. Have the LLM summarize this to see what's been popular in the past. Have the LLM bias your fresh, new idea to blend with previous content and fit the conference focus.
LLM's Can't Paraphrase to Clarify
Hint for folks using LLM's to pitch your ideas. Make sure all of your key buzzword-bingo phrases have some brief explanation. Those clunky phrases you felt needed to be included tend to stand out. You know what it means and the LLM seems to use it in grammatically correct-looking output. It checks a box for you. What about your audience?
We may not know what it means. And the LLM may be using it in weird, stilted ways. And the LLM can never paraphrase to clarify.