organizzazionecomunitasolidali / SEM

Shared Economy Marketplace featuring alternative currencies
GNU General Public License v3.0
0 stars 0 forks source link

Passare a GPT-4o #6

Closed lbattt closed 3 months ago

lbattt commented 3 months ago

image

 

image

Hi there,

We launched GPT-4o in the API—our new flagship model that’s as smart as GPT-4 Turbo and much more efficient. We’re passing on the benefits of the model’s efficiencies to developers, including:

GPT-4o in the API currently supports text and vision capabilities. It has better vision capabilities and improved support for non-English languages compared to GPT-4 Turbo. It has a 128k context window and has a knowledge cut-off date of October 2023. We plan to launch support for GPT-4o’s new audio and video capabilities in the API to a small group of trusted partners in the coming weeks.

We recommend that developers using GPT-4 or GPT-4 Turbo consider switching to GPT-4o. You can access GPT-4o in the Chat Completions API and Assistants API, or in the Batch API where you get a 50% discount on batch jobs completed asynchronously within 24 hours.

To get started, test the model in Playground, which now supports vision capabilities, and check out our API documentation. To learn how to use vision to input video content with GPT-4o today, check out the Introduction to GPT-4o cookbook. If you have questions, please reach out in the OpenAI developer forum. —The OpenAI team

mazzmarr commented 3 months ago

Ragazzi, mi sembra che GPT-4 si ciuccia un sacco di dollaroni...

image

Avevo configurato l'autoricarica confidando che 10 dollari durassero mesi e invece...

Per cui ho tolto l'autoricarica e ora quando arriva a 0 si ferma...

@lbattt @Neggia Bisogna capire un attimo cosa succede.

Neggia commented 3 months ago

Si non mettere l'autoricarica perché potrebbero esserci dei bug, e il motore deve funzionare comunque anche senza ChatGPT. In mesi di test io avevo speso meno di 10$, ma magari avevo trovato i siti meno onerosi.

lbattt commented 3 months ago

@Neggia Riguardo a versione 4o , non vale la pena. Sembra essere non superiore al 4-turbo come qualita' delle risposte. Ora usiamo il 4-turbo , che poi serve per via del contesto molto largo (128k).

@mazzmarr In effetti c'erano dei bachi e ho dovuto farlo rigirare parecchie volte. Ora penso si sia stabilizzato , ma comunque sto pensando a un modo per ridurre l'utilizzo di chatGPT a meno di un decimo di quello attuale. Ne parliamo in call penso oggi

lbattt commented 3 months ago

@neggia possiamo chiudere questa issue. Stiamo usando il 4o https://github.com/organizzazionecomunitasolidali/SEM/commit/9bbb9f2040cc1f2b52c2e02c9bc9f5fc826fb8ac

lbattt commented 3 months ago

@mazzmarr Non tenere mai piu' di 5$ di bilancio. E mai autoricarica.