Skip to main content

OpenAI DevDay SF

Today at DevDay SF, OpenAI is launching a bunch of new capabilities to the platform.

Realtime API

The new Realtime API, now in public beta, allows paid developers to create low-latency, speech-to-speech experiences in apps, similar to ChatGPT’s Advanced Voice Mode. It supports real-time streaming of audio inputs and outputs, offering more natural and responsive conversations. Alongside this, an update to the Chat Completions API introduces audio input and output, supporting multimodal interactions with text or audio responses. These updates simplify the process for developers by consolidating speech recognition, text processing, and speech synthesis into a single API call, enhancing use cases like customer support and language learning.

Prompt Caching

Prompt Caching, introduced today, allows developers to reuse recently seen input tokens across multiple API calls, reducing both costs and latency. This feature provides a 50% discount and faster prompt processing times. Prompt Caching is automatically applied to the latest versions of GPT-4o, GPT-4o mini, o1-preview, o1-mini, and their fine-tuned counterparts.

Model Distillation

OpenAI is introducing a new Model Distillation offering, providing developers with an integrated workflow to manage the entire distillation pipeline directly within the platform. Model distillation fine-tunes smaller, cost-efficient models using outputs from more capable models, improving performance at a lower cost. This suite simplifies the previously complex, multi-step process with three key features: Stored Completions, which captures input-output pairs to build datasets for fine-tuning; Evals, a tool for custom performance evaluations; and seamless integration with OpenAI’s fine-tuning services. This offering reduces manual effort and streamlines model optimization.

Vision Fine-Tuning

OpenAI has introduced vision fine-tuning on GPT-4o, allowing developers to fine-tune the model using images in addition to text. This enhances the model’s image understanding capabilities, enabling applications such as improved visual search, better object detection for autonomous systems, and more accurate medical image analysis. While many developers have used text-only fine-tuning to improve task-specific performance, the addition of image fine-tuning addresses the limitations of text-based models for more complex, visual tasks.

For more news like this: thenextaitool.com/news


Comments

Post a Comment

Popular posts from this blog

Google Unveils Next-Gen AI Models for Video and Image Generation

 Google Stakes Its Claim in AI Dominance with Veo 2 and Imagen 3 Google has announced the launch of two groundbreaking AI models: Veo 2 and Imagen 3 . These next-generation systems promise to revolutionize video and image generation, delivering unprecedented realism, detail, and creative control. With these releases, Google is solidifying its position as a leader in AI innovation. Veo 2: Redefining Video Generation Veo 2 is Google’s latest video generation model, capable of creating high-resolution 8-second clips at 4K resolution (720p at launch). The model boasts significant improvements in cinematic control, physics simulation, and reduced hallucinations, resulting in more natural and lifelike videos. In head-to-head evaluations against competitors like OpenAI’s Sora, Veo 2 emerged as the clear winner for its superior quality and prompt adherence. The model is being rolled out gradually through the VideoFX waitlist, with plans to integrate it into YouTube Shorts by 2025. Imag...
Post a Comment
Read more

How A Tiny Caribbean Island Hit The Digital Jackpot with AI

Turning The AI Boom Into A Windfall For Anguilla The artificial intelligence boom has transformed industries, fueled innovation, and created fortunes for tech giants like Elon Musk’s xAI and Meta’s AI division. But few expected that a small Caribbean island would also cash in on the frenzy. Anguilla, a British overseas territory, has raked in over $32 million thanks to its unique internet domain: “.ai.” In 1995, Anguilla was assigned the “.ai” country code by ICANN, the organization responsible for managing internet domain names. Fast forward to today, and the island is reaping the benefits as companies scramble to secure “.ai” domains to establish their presence in the AI space. From Google’s google.ai to Elon Musk’s x.ai, businesses are paying between $150 to $200 register these domains, generating millions in revenue for Anguilla. This unexpected windfall has become a lifeline for the island, which relies heavily on tourism and offshore banking. The $32 million earned last yea...
Post a Comment
Read more

Meta AI Releases Llama 3.3 70B Instruct

 Explore Llama 3.3 70B Instruct As It Sets New Standards In AI With Enhanced Reasoning, Multilingual, And Cost-Efficient Features Meta has introduced the Llama 3.3 70B Instruct, an advanced AI model that sets a new benchmark in reasoning, coding, and following instructions. As one of the most adaptable open models available, Llama 3.3 70B brings forth impressive capabilities with a wide array of applications. Enhanced Functionality and Multilingual Support This model shines in producing structured outputs, especially in step-by-step reasoning and JSON formatting, ensuring reliability and accuracy for developers. With support for eight major languages, including English, French, and Hindi, it aims to facilitate global communication. Revolutionizing Software Development The improvements in coding encompass extensive language support, better error handling, and comprehensive feedback, enabling developers to enhance their productivity. Llama 3.3 ’s task-aware tool usage optimizes re...
Post a Comment
Read more