Collaborating Authors

Human reinforced learning could mean 'more truthful and less toxic' AI


AI has been making huge leaps in terms of scientific research, and companies like Nvidia and Meta are continuing to throw more resources towards the technology. But AI learning can have a pretty huge setback when it adopts the prejudices of those who make it. Like all those chatbots that wind up spewing hate speech thanks to their exposure to the criminally online. According to Golem, the OpenAI might have made some headway on that with its new successor to the GPT-3, the autoregressive language model that uses deep learning in an effort to appear human in text. It wrote this article, if you want an example of how that works.

When Do Language Models Need Billion Words In Their Datasets


"What do data-rich models know that models with less pre-training data do not?" The performance of language models is determined mostly by the amount of training data, quality of the training data and choice of modelling technique for estimation. Pretrained language models like BERT use massive datasets on the order of tens or even hundreds of billions of words to learn linguistic features and world knowledge, and they can be fine-tuned to achieve good performance on many downstream tasks. General-purpose pre-trained language models achieve strong performance on NLU tasks through pretraining on billions of words. But what exact knowledge, ask the researchers at NYU, do these models learn from large scale pretraining that they cannot learn from less data? To understand the relation between massiveness of data and learning in language models, the researchers adopted four probing methods -- classifier probing, information-theoretic probing, unsupervised relative acceptability judgment, and fine-tuning on NLU tasks and plotted to learn curves (shown above) for the four probing methods.

OpenAI Releases An Improved Version Of Its Codex AI Model


Today OpenAI is releasing a new and improved version of its Codex AI model to the public. Codex is a descendant of OpenAI's GPT-3, which was released last summer. While Codex shares the same data as its predecessor, it has an added advantage in that it can read and then complete text prompts submitted by a human user. The Codex is like the GPT-3 language engine, but it was only trained on coding. In the latest, OpenAI has made some big changes to Codex by now accepting commands in plain English as well. This allows someone who is building a game or web app without naming any variables whatsoever, and they get live working code back quickly with no hassle.

Microsoft Releases Azure Open AI Service Including Access to Powerful GPT-3 Models


At its recent Ignite conference, Microsoft announced the new Azure OpenAI Service in preview, allowing access to OpenAI's API through the Azure platform. This new Azure Cognitive Service will give customers access to OpenAI's powerful GPT-3 models, along with security, reliability, compliance, data privacy, and other enterprise-grade capabilities available through the Azure platform. Earlier, the company invested in OpenAI, founded initially as a non-profit open-source organization by several investors, including Tesla founder Elon Musk. And the OpenAI API is the first commercial product in the for-profit OpenAI LP entity, allowing developers to leverage the general-purpose model for natural language GPT-3. The model GPT-3 and its fine-tuned derivatives, such as Codex, can be tailored to handle applications requiring a deep understanding of language, such as converting natural language into software code, summarizing large amounts of text, and generating answers to questions.