In brief Joe Biden and vice president Kamala Harris met with leaders from some of the top players in AI – Google, Microsoft, OpenAI, and Anthropic – this week to discuss AI safety risks as the government steps up efforts to regulate the technology.

The meeting coincided with other announcements from the White House, including spending $140 million to launch seven new AI research institutes under the National Science Foundation, and releasing draft policies outlining how the government will use AI that will be open for public comment in the future. 

It’s not clear exactly what was discussed. The Biden administration said the conversation was “to underscore this responsibility and emphasize the importance of driving responsible, trustworthy, and ethical innovation with safeguards that mitigate risks and potential harms to individuals and our society.” 

The White House was criticized for excluding academic institutions and other non-profit organizations from the meeting. But the administration hinted that AI experts outside of industry will get a chance to talk to officials at some point too. 

“The meeting is part of a broader, ongoing effort to engage with advocates, companies, researchers, civil rights organizations, not-for-profit organizations, communities, international partners, and others on critical AI issues,” according to a White House statement.

How much did it cost OpenAI to build ChatGPT?

OpenAI reportedly lost $540 million as it developed its viral chatbot, ChatGPT, last year.

Training large language models is expensive. It requires thousands of GPUs, and experts to build the systems and datasets. OpenAI losses reportedly doubled, according to The Information. It’s not clear whether that figure reflects the value of the cloud computing resources donated by Microsoft.

Microsoft is the biggest shareholder in the startup, and has invested $10 billion for 75 per cent of OpenAI’s profits until it breaks even. ChatGPT helped spark the current craze in language models, and is free to use. 

OpenAI, however, makes money by selling API access to companies, and has many different generative AI products – including its GPT-based language models, text-to-image DALL-E system, and audio-generating Whisper model.

The company reportedly expects its revenues to hit $200 million this year – a big jump from the $28 million it made last year. Reuters previously reported that OpenAI expects revenues to jump to a whopping $1 billion in 2024. It bagged a new round of funding worth $300 million from other top VC firms such as Sequoia Capital, Andreessen Horowitz, and others to reach a valuation of $27 billion to $29 billion, according to TechCrunch.

Microsoft drops waitlist for Bing AI

Netizens can now chat to Microsoft’s AI-enabled web search chatbot Bing without having to sign up for a waitlist and be approved.

The GPT-4-powered agent is designed to help people surf the internet in a more conversational manner. Instead of having to sift through webpages to find relevant information, Bing can provide answers to queries by summarizing text. But be warned: these new internet search chatbots don’t always provide accurate or up-to-date information.

Microsoft launched the new Bing three months ago, and has been tinkering with the software to prevent it going off the rails. Bing was known to generate unhinged responses, especially in the early days. Still, people seem to enjoy interacting with the new toy and continue to flock to the site, according to Yusuf Mehdi, Microsoft’s corporate vice president and consumer chief marketing officer.

“Bing has grown to exceed 100 million daily active users and daily installs of the Bing mobile app have increased 4X since launch. As a result we are seeing growth of Bing share and it follows the eight straight quarters of growth in our Microsoft Edge browser share,” he wrote in a blog post.

Bing has also expanded its capabilities, and is able to process and generate data across different formats like images and video on top of text. Microsoft is working to make the chatbot accessible to Windows users as a taskbar app to increase usage. Mehdi said the move could reach over half a billion customers every month.

TV and film writers in the US strike over AI

The Writers Guild of America, a union group representing writers working in TV and film, is concerned AI will cost jobs.

Large language models like ChatGPT were on members’ minds as they announced they were striking this week after failing to reach agreement with top production studios over working conditions.

On top of pay and guaranteed gigs, the union wants [PDF] to regulate the use of AI in the industry. Writers want to make sure ChatGPT cannot be used to produce literary material, and that their work won’t be used to train models. But officials from production companies like Netflix and HBO have not agreed with their proposals.

“The challenge is we want to make sure that these technologies are tools used by writers and not tools used to replace writers,” John August, a screenwriter and author on WGA’s negotiating committee, told The Hollywood Reporter. “The worry is that down the road you can see some producer or executive trying to use one of these tools to do a job that a writer really needs to be doing.” ®

Source: https://go.theregister.com/feed/www.theregister.com/2023/05/08/in_brief_ai/