Google denies Bard trained using OpenAI ChatGPT responses
ALSO: Synopsys releases AI tools for faster chip design, and Cerebras open sources family of language models
In brief A Google engineer reportedly quit after warning CEO Sundar Pichai that the company was wrong to train its AI search chatbot Bard on text generated by OpenAI's ChatGPT.
Netizens have posted snippets of their conversations with ChatGPT on a website called ShareGPT. OpenAI prohibits people from using its outputs to train their own models.
AI engineer Jacob Devlin raised concerns that Google would violate OpenAI's terms of service by harvesting data from the website to train its own Bard chatbot, The Information reported last week. Devlin thought the practice was not only wrong, it would make Bard behave too similarly to ChatGPT. After he escalated concerns to Pichai, he reportedly resigned from the company, and joined OpenAI.
Competition to develop and deploy the most attention-grabbing generative AI products amongst Big Tech companies – especially between Google and Microsoft investee OpenAI – is fierce right now. Directly compiling training data from the outputs of a rival's model would be … awkward. It will be difficult to avoid in the future, as text generated by different models spreads on the internet, but doing it on purpose is undeniably naughty.
Google denied it had trained Bard on text produced by ChatGPT. "Bard is not trained on any data from ShareGPT or ChatGPT," a spokesperson from the ad giant told The Verge. The representative, however, declined to comment on whether Google had ever used text generated by ChatGPT to train Bard at all.
Synopsys launches AI tools to design chips
A developer of electronic design automation software, Synopsys, announced a set of AI-powered tools aimed at helping engineers make chips more efficiently.
Microchips are complex systems that can containing billions of transistors, assembled into complex subsystems like CPUs and GPUs. Engineers must ensure those components are carefully arranged.
to do so, engineers use different types of software to refine chip designs . Synopsys has launched a suite of AI tools to tackle system architecture from design to manufacturing.
"AI design tools are enabling chipmakers to push the boundaries of Moore's Law, save time and money, alleviate the talent shortage, and even drag older chip designs into the modern era," the developer said, quoting analysts from Deloitte.
The software reportedly helps engineers produce chip designs more quickly, predicts where bugs could occur, tests for silicon defects, and speeds up the manufacturing process. Increased productivity levels mean chip shops can produce more and more chips to power today's technology.
- Italy bans ChatGPT for 'unlawful collection of personal data'
- FTC urged to freeze OpenAI's 'biased, deceptive' GPT-4
- This US national lab turned to AI to hunt rogue nukes
- UK seeks light-touch AI legislation as industry leaders call for LLM pause
AI hardware startup Cerebras releases seven free open source large language models.
Cerebras, maker of the world's largest chip, trained the GPT-3-based language models using datasets ranging from 111 million to 13 billion parameters in size.
The models were trained under the Chinchilla protocol – a method outlined in a paper published by DeepMind – which figures out how much data a model of a given size should be trained with using limited computational resources.
Cerebras claimed its GPT systems have "faster training times, lower training costs, and consume less energy than any publicly available model to date." They were trained using the company's CS-2 systems – part of its Andromeda AI supercomputer.
"Artificial intelligence has the potential to transform the world economy, but its access is increasingly gated," the AI hardware startup stated in a blog post. "The latest large language model – OpenAI's GPT4 – was released with no information on its model architecture, training data, training hardware, or hyperparameters. Companies are increasingly building large models using closed datasets and offering model outputs only via API access."
"For [large language models] to be an open and accessible technology, we believe it's important to have access to state-of-the-art models that are open, reproducible, and royalty free for both research and commercial applications."
Cerebras's code describes the models' architecture, weights, and training checkpoints, which were made available on Hugging Face and GitHub under the Apache 2.0 license.
US Federal Trade Commission paying close attention to Big Tech and AI
The US Federal Trade Commission's chairwoman Lina Khan warned she would be keeping a close eye on the AI industry to make sure it isn't controlled by Big Tech.
"As you have machine learning that depends on huge amounts of data and also a huge amount of storage, we need to be very vigilant to make sure that this is not just another site for big companies to become bigger," Khan said this week during an event hosted by the Department of Justice, according to Bloomberg.
Under Khan, the FTC has focused on cracking down on the largest technology monopolies and their potential antitrust and anti-competitive issues. The commission has taken an interest in AI, and has urged companies building the technology to ensure their products are safe and trustworthy if they don't want the regulator breathing down their necks.
"Sometimes we see claims that are not fully vetted or not really reflecting how these technologies work," Khan said. "Developers of these tools can potentially be liable if technologies they are creating are effectively designed to deceive." ®