Another Airtable AI extension released

Just found this:

The publisher contact email
points to:

https://alioth.io/

@bfrench I can’t believe somebody beat you on this one.

1 Like

Haha! Who said I didn’t already build this? There are many things I build and use but cannot say a word about. :wink:

This is yet another of the millions of copilots I predicted would emerge. This looks awesome, and it will help many people do some extraordinary stuff in Airtable. However cool this is, the creator has to know that it could (and probably will) be made obsolete as soon as Airtable integrates this functionality. I place this project in the “aftermarket DMZ”; a zone where you tread at great risk.

As to pricing, it costs $0.10 per record to use AI with this extension. I predict there will be few takers at any significant volume because this - on face value - seems to have a 10x markup, possibly more. I don’t have time to do a proper evaluation, but it looks pricey with what I know about OpenAI’s pricing.

Business models in AI are critical. Ideally, you want fast and furious adoption of the extension, and the way to do that is to set the hook with a free trial. Each user should bind the extension to their own OpenAI API key. This would lower inferencing costs while making it possible to use the existing rate-limiting systems in place at OpenAI.

Then, as users get hooked (and they will), extract revenues based on the value the tool creates, not on the volume of inferences consumed.

I agree on pricing, it’s quite costly, given not every try will provide useable results. Wish somebody who has written several extensions for Airtable already and who reads this would pick on the idea. One time purchase and let us use OpenAI awesomeness. Who could it be :thinking:

This extension looks cool! However, as @bfrench mentioned, it is extraordinarily pricey. And extensions can’t be automated.

All of this can already be completely automated for free on Make’s free plan:

2 Likes

Two things about this are equally troubling:

  1. Does Make guarantee that your prompt data is not going to OpenAI for training and other analytics?
  2. Does Make provide limiters to ensure a runaway automation doesn’t accidentally bankrupt you?

To point #1 - presumably, this process is using your OpenAI API key which when used directly by you is opted out of data sharing. However, API keys used by proxy through a third party may ignore this default and share your data through other serquitous processes. Imagine a deal cut with Make to trade your data for lower inferencing costs or some other benefit.

My clients use Make heavily for these purposes, and they love it!

I will be showing off a brief demo this upcoming Tuesday on the BuiltOnAir podcast.

I’ll post the exact link to the exact episode after the podcast airs, but the generic link to the channel is here:

1 Like

According to OpenAI they promised no longer use new confidential data. But that is of course only as good as their promise is worth. I’ll trust them a bit more than Facebook but without a clear policy on this it’s a problem waiting to happen.

For first-party API calls. Is that true when you entrust another party to make calls on your behalf? Is it possible that promise may not apply under certain cases? Is it possible big aggregators make deals to share their data (I.e., your data)?

1 Like

Just for the record scripts supplied by Airtable users.

I am a bit leery about having Make/Zapier glue to interact with ChatGPT given that there is cost to that and I may make a mistake I will rack up costs. So I somehow prefer the interaction to run within the confines of Airtable.

Here is extension script:

And here is Automation script:

I predict this will not be the pathway that causes the most significant financial loss or risk. Rather, it will be unskilled prompt “engineers” who use Airtable automations with LLMs.

Building unattended processes from app-to-LLM requires very special considerations that include prompt design and logic that cannot easily be crafted [presently] in Airtable without code.

This seems to be pretty neat GPT implementation because it can interact directly in the table because the field supports scripting.

2 Likes