OpenAI GPT-3 AI available in Gladys Assistant 🤯

[jparbel]

Well done, Pierre-Gilles!

To the question: « Who is the creator of Gladys Assistant? » here is the glowing answer from Bing GPT4:
![

[pierre-gilles]

Hi everyone!

Good news, I re-ran tests on my side and the ChatGPT API (Model « gpt-3.5-turbo ») seems to have stabilized in terms of response time; I’m getting response times similar to the « text-davinci-003 » model I had been using until now

3 changes:

• Gladys Plus now uses « gpt-3.5-turbo »
• Instead of being limited to 100 messages per month, the limit is now 1000 per month (i.e., 33 per day, comfortable!)
• I reset everyone’s quota for this fresh start

I now need your feedback to confirm that this model meets your expectations as well as « text-davinci-003 » did!

I’m also curious to know how you’re using this integration

[lmilcent]

« Gladys, what do you see in the entrance? »

Answer 1 : Shows me the camera✅
Answer 2 : Describes a « standard » entrance, nothing like my place

[pierre-gilles]

@lmilcent

Answer 2: It describes a « standard » entry to me, nothing like mine

aha, it’s the famous « hallucinations » of ChatGPT ^^ Anyway ChatGPT doesn’t have access to the image so it can’t « see »

[pierre-gilles]

Hi everyone!

New update to the ChatGPT integration today!

I replaced the old model “GPT-3.5-turbo” with the newer, higher-performing model “gpt4o-mini”. In addition to better response quality, you’ll also notice faster response times.

A small performance comparison of the 2 models:

Source: https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

I see there are requests for new features for the AI integrated into Gladys, notably for the camera image analysis part ( Message écrit par une IA qui décrit le mouvement détecté d’une caméra )

If you have other ideas/requests, don’t hesitate, it would be cool to improve the AI part in Gladys

[guim31]

Great, that’s the model I use every day, it’s really cool

[pierre-gilles]

FYI, I’m working on a pretty futuristic feature: AI proactive notifications from scenes!

Message écrit par une IA qui décrit le mouvement détecté d’une caméra

Salut à tous C’est un sujet que je voulais faire depuis longtemps, notamment la partie « Poser une question à l’IA depuis les scènes » (même sans image) pour pouvoir avoir des notifications proactives venant de l’IA (futuriste!) Comme j’avais le nez dans la partie ChatGPT ce matin j’ai développé cette fonctionnalité ce qui n’était pas très compliqué avec la nouvelle API lié à la vision de GPT-4o mini Demander à l’IA de décrire une image de caméra Dans une scène,…

[bertrandda]

It’s great to see this part developing, well done.

@pierre-gilles

If you ever have other ideas/requests, don’t hesitate, it would be cool to improve the AI part in Gladys

Not being a Gladys + user, I’m only looking at the code, how you implemented the AI topics and it seems I have a remark about the TTS part. Tell me if I’m wrong, but I couldn’t see which ElevenLabs model you were using and if you’re using the default one that means it’s the multilingual 1.0 model. You can use their new Turbo 2.5 model, with faster generation (could be ideal in the context of exchanges between the user and Gladys) and cheaper — I use it for a personal project and it works well; you can do about twice as many generations (in time/characters) with the same number of credits. You can even specify the desired language to improve the accent; since we have it in the Gladys user account, that would help to have something even more natural for languages other than English.

For the OpenAI part, I didn’t fully understand how it works on Gladys Plus, but I imagine a large part of the implementation is hidden behind the OpenAI web configuration console.

In any case, great work on what you’ve announced about this part; it will help win the hearts of new users

[pierre-gilles]

@bertrandda

Tell me if I’m wrong, but I couldn’t see which elevenlabs model you were using

I use « eleven_multilingual_v2 » with the « Grace » voice

Good catch — I hadn’t seen the release of their Turbo model!

https://elevenlabs.io/blog/introdu

[jgcb00]

I saw this as well:

The demo is quite convincing in English and the license is permissive, but I’m not sure it will work as well in French, or that there is a model in French

[pierre-gilles]

Well the Turbo model isn’t quite there yet

It’s very English in its pronunciations, it’s borderline ridiculous

Too bad!

[JeuFore]

I was able to test the voice « Guillaume - Narration », I found it not too bad, though

[pierre-gilles]

Ah, you went into custom voices! I was using the official ElevenLabs voices.

I’m looking for a female voice, and if possible without a cost multiplier. Some custom voices cost x2 in price.

I tested « Audrey » and « Jeanne », they’re not bad but in Turbo you can hear the English accent.

Have you found any nice female voices?

[pierre-gilles]

Otherwise there’s « Mademoiselle French - For Conversational » in Turbo :

But it ends up being almost too realistic; I liked the voice sounding just a little robotic (not too much), just so it sounds like an « AI » assistant

[Gnome]

The realist is good too…lol

[pierre-gilles]

While I was at ElevenLabs, I tested one of their new features: the ability to automatically translate a video from one language to another, I did some tests with 2 Gladys YouTube videos:

It seems like I have an Indian accent though

Anyway it’s pretty powerful, maybe one day the videos I post will be translated automatically!