My 2 cents on the state of plugins

Community
21

I chatted with the dev who ported my Vectra project to python today and I think a port of AlphaWave is coming. Also have someone looking into a .NET port of AlphaWave. It’s ability to guarantee that you only ever get validated responses (like schema validated JSON) back from the model is pretty compelling so I think we’ll see the ideas in AlphaWave spread to other frameworks as well.

I’ve actually gotten derailed from my Agent work for AlphaWave because there’s a team in Microsoft that’s wanting to use AlphaWave for a project so I’m having to wire it up to Microsoft Teams. My goal is to have Auto-GPT style Agents working by the weekend though. I’m almost there.

1 Like
22

I agree. There are now multple versions offering the same functionality and others that are utterly useless. Its quickly become a pile of mostly trash that has to be sifted through to find meaningful features. Not every possible entity needs to offer there own plugin. Get ahold of this and lets limit plugins (if required at all) to those that actually enhance and bring value.

23

Honestly, I love the idea of plugins, but the current state of things is a complete shit-show.

They absolutely need to build a marketplace application where each plugin can have it’s own information page.

The idea of using a plugin but having zero idea of the identity of the developer is insane to me.

The fact there is no way (at least none I’m aware of) to access documentation about the individual plugins is insane to me.

The fact they don’t appear to have any quality standards for plugins other than they be functional is insane to me.

I mean just look at the first eight plugins down to me under All.

Screenshot_20230617_150611_Chrome
Screenshot_20230617_150611_Chrome2014×1559 220 KB

Screenshot_20230617_150505_Chrome
Screenshot_20230617_150505_Chrome2052×1492 252 KB

Why do I need to talk to an AI about planning a trip involving Greek ferries‽ I mean, I guess maybe this would be useful in some context. But, it’s just too narrow.

Now, a plugin that would give ChatGPT access to detailed trip/transportation information for all modes of transportation everywhere… That would be useful.

Or, why would I need to chat with an AI about animal transportation companies? Could someone need that? Sure, probably. But, it’s so narrow! And it’s one company’s data.

Now, if you gave ChatGPT the ability to access information from dozens of sources, e.g. Yelp, BBB, Trustpilot, specialized sites like this, and maybe pull business registration information from government sites, any legal proceedings the company has been involved with, identify the company principals and employees, find any political donations they’ve made, pull in posts and comments from things like Twitter, Reddit, Facebook etc, then had ChatGPT synthesize a comprehensive report on each company tailored specifically to an individual’s hierarchy of concerns? Interesting.

Giving it the ability to aggregate and combine disparate services to provide a cohesive platform agnostic experience? Interesting.

Giving it the ability to essentially run a search on one very narrow website? Hot, boring, garbage.

1 Like
24

Good points. I pay for “plus” with the intention that it will result in efficiency improvements in the work and research that i conduct. What is inefficient is playing trial and error to determine which plugin actually has value. I get that not everyone is utilizing the capabilities the same as i am, but the direction this is heading feels like it is going to be more work than necessary for a paid service. Some vetting, quality control, and removal of essentially duplicate plugins is overdue.

25

Perhaps the next evolution for plugins could be a plugin of plugins. The functions within it might be constituted by other plugins. For instance, a user could simply add a travel plugin, which in turn utilizes various other plugins to book flights, rent cars, and secure hotel accommodations. The best solution is then presented to the user.

26

I think the only future is one which doesn’t involve users needing to curate their own plugins.

It should be incumbent on OpenAI to gatekeep plugins for popular consumption. Plugins should only be approved if they ameliorate some deficiency in the model or provide some substantial, widely applicable, new and novel capability.

We don’t need 200 extremely niche plugins which give access to 200 extremely niche data APIs.

We need one data plugin that lets the model find any relevant data in any and all available services, aggregates it, synthesizes as complete of a snapshot as possible of everything known about whatever information is needed by the model at the time.

We don’t need 30 plugins for querying a PDF. We need one that actually works—that is it can read more than just the text. It can read mathematical expressions, tables, algorithms, understands references, can interpret plots and pictures and their captions.

That’s it, just give me one jibber-jabber-with-your-pdf plugin that works.

Honestly, the state of plugins is depressing.

1 Like