Post by account_disabled on Mar 11, 2024 5:55:59 GMT
OpenAI news presented at DevDay: some reflections There are so many new features presented during the event. All innovations, however, which we could define as " horizontal ", ranging in the dimensions of speed , price and integration . There was no talk of substantial new abilities from a cognitive point of view . Question of time waiting for GPT-5 or the signal of a plateau? OpenAI DevDay A summary of the news Let's go step by step, summarizing the new features presented by key points. Increased speed and context window , with the new GPT-4 Turbo model that can handle 128k tokens ( impressive! ). This means that, as an estimate, we will be able to handle documents of over 300 pages in a single prompt.
GPT-4 Turbo will have training data updated in April 2023 , and will have India Mobile Number Data fine-tuning functionality , to create refined models. ChatGPT will also benefit from the new knowledge cutoff. API Assistants , to create specialized agents capable of also using functions , using Code Interpreter ( therefore developing Python code to respond to user requests ) and using a customized Knowledge Base for training. In the following image you can see an example of use. Vision and DALLE-3 available via API. This means that we will be able to process and generate images within our applications in total multimodality ( also considering TTS and ASR which we will see shortly ). New speech synthesis APIs ( TTS - Text To Speech ) that will allow you to generate high quality voices. Cost Reduction : GPT-4 Turbo APIs will have 3x lower costs than GPT-4 for input tokens, and 2x lower costs for output tokens.
Copyright Shield relating to the use of the API and ChatGPT Enterprise. OpenAI will cover customers' legal fees if they are accused of copyright infringement. Whisper V3 , the new model dedicated to speech recognition (ASR). You will be able to convert audio into text in a more advanced and faster way. ChatGPT customization , through " GPTs ". As I said from the beginning, custom instructions were practically useless, because they were too restrictive for those who use ChatGPT for different tasks. The new functionality allows you to create different agents with customized characteristics. Small specialized assistants who respond to our needs in chat mode. GPTs will be able to be connected to the " real world " through API calls , connection to databases and other data sources. For example, personalized GPTs can be created to assist different figures in the company.
GPT-4 Turbo will have training data updated in April 2023 , and will have India Mobile Number Data fine-tuning functionality , to create refined models. ChatGPT will also benefit from the new knowledge cutoff. API Assistants , to create specialized agents capable of also using functions , using Code Interpreter ( therefore developing Python code to respond to user requests ) and using a customized Knowledge Base for training. In the following image you can see an example of use. Vision and DALLE-3 available via API. This means that we will be able to process and generate images within our applications in total multimodality ( also considering TTS and ASR which we will see shortly ). New speech synthesis APIs ( TTS - Text To Speech ) that will allow you to generate high quality voices. Cost Reduction : GPT-4 Turbo APIs will have 3x lower costs than GPT-4 for input tokens, and 2x lower costs for output tokens.
Copyright Shield relating to the use of the API and ChatGPT Enterprise. OpenAI will cover customers' legal fees if they are accused of copyright infringement. Whisper V3 , the new model dedicated to speech recognition (ASR). You will be able to convert audio into text in a more advanced and faster way. ChatGPT customization , through " GPTs ". As I said from the beginning, custom instructions were practically useless, because they were too restrictive for those who use ChatGPT for different tasks. The new functionality allows you to create different agents with customized characteristics. Small specialized assistants who respond to our needs in chat mode. GPTs will be able to be connected to the " real world " through API calls , connection to databases and other data sources. For example, personalized GPTs can be created to assist different figures in the company.