AI Newsletter #013 (2023/08/21 – 2023/08/27)

The New York Times Blocks OpenAI’s Web Crawler
The New York Times has blocked OpenAI’s web crawler, preventing OpenAI from using the publication’s content to train its AI models. The block was implemented on August 17th and can be seen on the NYT’s robots.txt page.

OpenAI Opens GPT-3.5 Turbo Up for Custom Tuning
OpenAI has announced that businesses can now fine-tune their own data using GPT-3.5 Turbo, which can match or exceed the abilities of GPT-4 for certain tasks. OpenAI will also be opening up GPT-4 for the same purpose later this fall.

Meta Releases an AI Model That Can Transcribe and Translate Close to 100 Languages
Meta has developed an AI model called SeamlessM4T that can translate and transcribe nearly 100 languages across text and speech. The model represents a significant breakthrough in AI-powered speech-to-speech and speech-to-text translation. Meta scraped publicly available text and speech data to create the training dataset for SeamlessM4T. However, concerns about biases in the model have been raised, including gender biases and toxic translations.

Google, Amazon, Nvidia, and Others Put $235 Million into Hugging Face
Open-source AI model repository Hugging Face has raised $235 million in a Series D fundraising round, with investments from Google, Amazon, Nvidia, Salesforce, and other major companies. Hugging Face, which serves as a GitHub for machine learning and AI models, plans to use the funding to expand its team and invest in more open-source AI and platform building. The investment highlights the growing importance of the open-source community in the development of generative AI.

OpenAI Partners with Scale AI to Allow Companies to Fine-Tune GPT-3.5
OpenAI has announced a partnership with Scale AI to enable companies to customize and fine-tune its GPT-3.5 text-generating model. This collaboration will allow Scale AI customers to use their own data to enhance and customize the AI model, tailoring it to specific tasks and requirements. The fine-tuned models will be reviewed by human experts to ensure performance and safety.

Meta Releases Code Llama, a Code-Generating AI Model
Meta has open sourced Code Llama, a machine learning system that can generate and explain code in natural language. Code Llama is capable of completing code and debugging existing code across various programming languages. The models were trained using a large dataset of code-related data and tokens. While Code Llama has potential benefits for programmers, there are also risks associated with generative AI tools, such as security vulnerabilities and potential copyright infringement.

Want to receive TechNavi Newsletter in email?