Image: Getty Images/Eugène Mymrin.
OpenAI is still leading the artificial intelligence race. This does not prevent it from continuing to release modifications, updates and even new technologies – the pace of development of the company has not weakened.
The company announced last week a highly anticipated improvement in the performance of GPT-4 Turbo, as well as a reduction in the costs of its API.
A less “lazy” model
Last Thursday, OpenAI unveiled a series of updates. Among these new features, the company presented a new version of GPT-4 Turbo, its most advanced AI model. Since its release, the latter has been the subject of numerous complaints regarding its performance. In particular, he was accused of being “lazy”.
To resolve this issue, OpenAI has released a preview update: gpt-4-0125-preview. The latter allows the GPT-4 model to better accomplish certain tasks – such as generating code for example. According to the company, this preview would also “reduce cases where the model shows “laziness” by not completing a task”.
OpenAI also provided an update on the highly anticipated GPT-4 Turbo with Vision, which promises to become its most advanced AI model. It should be available to everyone in the coming months.
A less expensive API
Next week, OpenAI will also present a new GPT-3.5 Turbo model, gpt-3.5-turbo-0125, which will bring “various improvements”, including more precise responses, respecting the requested formats, but also a correction of bugs in Text encoding for requests in languages other than English.
Prices for GPT-3.5 Turbo will also drop, for the third time since the beginning of last year. The entry price for the new model should therefore increase to $0.0005 per 1,000 tokens (i.e. a 50% cost reduction), and the exit price to $0.0015 per 1,000 tokens (-25%). ).
The tokens are used as currency for each task carried out on the API offered by OpenAI. The number of tokens needed to complete a task depends on the length of the input and output and can often add up quickly if using ChatGPT for large amounts of text. The price drop should help offset the cost of using ChatGPT for these applications.
An improved free API
The latest update concerns the free moderation API, which developers can use to identify potentially dangerous text. OpenAI has released its most robust moderation model, text-moderation-007.
As for the new models, OpenAI has released two new embedding models: the text-embedding-3-small model and the text-embedding-3-large model. Embedding models represent text, natural language, or code as vectors, which helps machine learning models understand relationships between content.
Source: ZDNet.com