OpenAI held its first DevDay, the place quite a few wonderful new options had been introduced.
The OpenAI DevDay convention was hosted by the corporate’s CEO Sam Altman. He highlighted that ChatGPT has change into one of many quickest rising client merchandise and reached 100 million customers monthly only a few months after its launch. At this time, greater than 100 million individuals use ChatGPT each week. And greater than 2 million builders create their options utilizing the ChatGPT and Whisper APIs by OpenAI.
Here is a abstract of OpenAI’s main bulletins in case you missed the occasion.
Launching GPT-4 Turbo
Customers will quickly achieve entry to a brand new mannequin – GPT-4 Turbo. This mannequin offers a 128K context window, has a decrease value, and incorporates the equal of greater than 300 pages of textual content in a single immediate. GPT-4 Turbo encompasses data of world occasions till April 2023. The pricing for enter and output tokens has been lowered by 3 and a couple of instances respectively in comparison with GPT-4.
Assistants API
Builders can now create specialised AI assistants that implement particular directions, use extra data from the consumer base, and may name upon fashions and instruments to carry out their duties (for instance, a Python code interpreter).
Actually, this is a chance to personalize the OpenAI mannequin to your duties and your data base.
Multimodal capabilities within the API
In GPT-4 Turbo, it is possible for you to to transmit not solely textual content but in addition pictures by way of the API and create prompts about them. It can even be potential to generate pictures utilizing an API request and convert textual content to human-like speech utilizing the speech synthesis API.
Reproducible Mannequin Calls
API calls now have a seed parameter – a beta perform that means that you can get reproducible outcomes, which is helpful for circumstances of reproducing requests for debugging, writing extra numerous unit assessments, and usually monitoring mannequin habits.
Log possibilities
In a number of weeks, OpenAI plans to launch a perform that returns the log possibilities of every subsequent token in GPT-4 Turbo and GPT-3.5 Turbo. This will likely be helpful for creating options equivalent to autocomplete in search engines like google, in addition to for detecting mannequin hallucinations.
Extra tokens per minute
Token limits per minute for paid GPT-4 customers have been doubled. The brand new limits can be found in response to the utilization tier.
Customized Fashions
A brand new program is obtainable for organizations that require extra customization than the same old mannequin setup. This system means that you can work with OpenAI researchers to create customized GPT-4 fashions for particular domains.