ChatGPT API Fallback Skill Integration

Hi everyone!

I saw few topics about GPT-3 skills here, but as soon as direct access to ChatGPT API were announced, is anyone is about to create new skill?

I guess it’s a great idea to give ChatGPT a voice, might be very useful skill.

I found a link recently to this fallback skill for OpenVoiceOS:

I’m planning to give it a try this weekend, but I’m not sure what route I’ll try (I’m not sure if OVOS plugins are compatible with Neon, which is what my Mark II is currently running, or what the state of OVOS compatibility with the Mark II in general is).

1 Like

ovos plugins and skills work in classic core and Neon

neon-core is built on top of ovos-core, everything made for OVOS works in Neon, and the reverse is also usually True

in general OVOS + Neon stuff also works in classic core, with some specific components having bugs and limitations under classic core due to imperfect compatibility layers (eg. OCP framework)

the linked skill is under active development and will soon migrate to the new ChatGPT Api

Good to know on the compatibility, thank you.

I could be mistaken, but from my limited reading I was under the impression that the existing OpenAI chat API has been updated in place to support GPT-3, so the process of moving to the “new” API just means swapping the model name in the request to the new gpt-3.5-turbo one.

I’m working on integrating OpenAI’s Whisper API for STT and then the ChatGPT API as the primary engine. I’m wanting to avoid the built-in datacollection conducted by Mycroft with their primary engine. OpenAI claims to not do training on their API input, so I feel more secure utilizing that than the built in engine.

Still sorting through it all. I’ve followed Mycroft for several years, but I haven’t done anything with Mycroft until now. So I’ve got a bit of a learning curve just getting Mycroft stood up, let alone working in those new APIs.

Hi, I am currently working on the gpt-skill that uses the openai api. I will provide the link when I’ve pushed to git

1 Like

Hi and thanks for the new skill, used a fresh secret token from my OpenAi account, but think it has not been used so far. Do I need a different token for the skill settings? Have AIML and Wolfram Alpha installed, do they may be interfering? when can I go sure to trigger the use of GPT3? Did a shortcut on an iPhone with another secret key and can use it by voice and text, so I am sure this part works.
Thank you and a good day, Michael

Yeah, I’m looking into Wolfram Alpha interfering. It seems it’s due to wolfram being a Queryskill while gpt is a Fallback Skill. Meaning GPT will only trigger when wolfram doesn’t have an answer. For the token, you’ll need to setup a payment method and before creating a token. You could test if the GPT skill works by running it from __init__.py. as a program in your skills directory and logging the response. Let me know if that helps.

Hi Victor, Thaks a lot for your answer. Not sure what you mean with payment method, for the “Siri pro” implementation (using an iPhone -shorcut, following a YouTube tutorial) it works with the generated key.
For the problem with fallback, would it be possible to give Mycroft a key phrase like “ask GPT to do xyzzy”, and so Mycroft would forward the request to the OpenAi API and gives back. More or less Trigger skill with key phrase, Speech to text; input to GPT, get answer; output Text-to-speech. Think this would be a great ad-on and would make Mycroft a control station for additional services. In Principe this is what they do with streaming on tunein, right?
Have a good day, Michael

Yeah for the “problem with fallback”, It’s a brilliant idea. I’ll look into it. What I meant for the payment method is, try generating a new key

leaving this here, as it is our replacement to fallback-unknown skill

1 Like