|
Dec 20, 2024
|
|
|
|
|
|
TGIF! Google releases its own reasoning AI model. A senior OpenAI researcher who helped develop some of the company's most important AI models is leaving. Elon Musk unexpectedly joins Donald Trump's dinner with Jeff Bezos.
|
|
|
|
Google on Tuesday released a new artificial intelligence model that aims to automate more complex tasks by spending a longer time processing a person’s instructions, known as reasoning. The launch comes several months after OpenAI pioneered a reasoning model of its own, o1, for which it charges $200 a month to customers that use it the most. Reasoning AI models have been most useful in applications such as coding and deep scientific research. Google’s move comes as it is preparing to make chatbot-like answers prominently available in its search engine, The Information reported
Thursday, a recognition of the progress OpenAI’s ChatGPT has made in attracting people who use it to search for information across the web. Google’s new Gemini “Thinking” model provides answers that include showing the steps it uses to arrive at the answers, known as chain of thought. That’s somewhat different from OpenAI’s approach to obscure its reasoning model’s chain of thought by only providing a summary of it. OpenAI said it does so for security and competitive reasons. Google’s decision to post the raw chain of thought could make it easier for other AI developers to find ways to develop their own models that emulate Google’s. In a post on X, Noam Shazeer, the famed Google researcher who is co-leading Google’s reasoning team, showed how the model could solve a probability problem. However, the model still struggles on a well-known question that has tripped up other AI models: how many R’s are in the word “strawberry.” Google’s AI incorrectly said that there are only two R’s.
|
|
|
|
Alec Radford, an OpenAI researcher who helped develop some of its most important artificial intelligence, on Thursday told colleagues he was leaving to pursue research independently. Radford said he planned to collaborate with OpenAI as well as other AI developers, according to a person who saw Radford’s message. “We have deep respect and appreciation for Alec and his contributions, and look forward to continuing projects with him as he explores independent research,” Mark Chen, OpenAI’s head of resarch, said in a statement. Radford had gained
legendary status for being the lead author on the company’s seminal research paper on generative pretrained models, GPT, a type of AI that now powers ChatGPT. He worked on future versions of GPT, a speech recognition model called Whisper, and the company’s image-generating model, Dall-E. Radford joined OpenAI around 2016, making him among the most senior employees at a company that has exploded in size over the past two years. Employees at the company have sold more than $800 million worth of shares to new investors in the past year or so. Other senior OpenAI research leaders including Chief Technology Officer Mira Murati, research chief Bob McGrew and co-founder and researcher John Schulman also left in recent months for various reasons.
|
|
|
|
President-elect Donald Trump had dinner with Elon Musk and Jeff Bezos at Mar-a-Lago on Wednesday, the New York Times reported. Musk wasn’t expected to be part of the dinner but joined after it had already begun, according to the report. The Times did not report what the trio discussed. Earlier in December, Bezos said he was “very optimistic” about Trump and hopes to help him cut government regulation. It’s not the first time Musk has unexpectedly joined one of Trump’s meetings with a competitor. When Google CEO Sundar Pichai called Trump in November to congratulate him on his election victory, Musk unexpectedly joined the call, The Information has reported.
|
|
|
|
Fintech banking startup Chime Financial has filed confidential paperwork for an initial public offering, Bloomberg reported. Chime declined to comment. The company, which has long been planning to go public next year, offers fee-free digital banking sources. The company was valued at $25 billion during the 2021 venture funding boom, although it more recently was valued at $9.5 billion in the secondary market for private tech stocks, according to Caplight. Chime has been honing its products ahead of a
planned IPO, including by launching a new cash advance product in May that allows customers to access up to $500 of their wages before they are paid. Earlier this year, Chime CEO Chris Britt said that the company had 7 million active customers, and that it was profitable in the first quarter of 2024.
|
|
|
|
OpenAI is currently prepping the next generation of its o1 reasoning model, which takes more time to “think” about questions users give it before responding, according to two people with knowledge of the effort. However, due to a potential copyright or trademark conflict with O2, a British telecommunications service provider, OpenAI has considered calling the next update “o3” and skipping “o2,” these people said. Some leaders have referred to the model as o3 internally. The startup has poured resources into its reasoning AI research following a slowdown in the
improvements it’s gotten from using more compute and data during pretraining, the process of initially training models on tons of data to help them make sense of the world and the relationships between different concepts. Still, OpenAI intended to use a new pretrained model, Orion, to develop what became o3. (More on that here.) OpenAI launched a preview of o1 in September and has found paying customers for the model in coding, math and science fields, including fusion energy researchers. The company recently started charging $200 per month per person to use ChatGPT that’s powered by an upgraded version of o1, or 10 times the regular subscription price for ChatGPT. Rivals have been racing to catch up; a Chinese firm released a comparable model last month, and Google on Thursday released its first reasoning model publicly.
|
|
|
|
Apple has criticized Meta Platforms for making a large number of requests to access Apple’s “sensitive technologies” within its mobile operating system iOS, warning that such requests could allow Meta to read iPhone owners’ emails and messages. Meta made the requests under a rule introduced by Europe’s Digital Markets Act, according to a report this week by Apple. The rule requires Apple to open access to its technologies to other companies and developers, so that their services can work with Apple’s. According to Apple, Meta has made more interoperability requests than any
company, and, in many cases, “Meta is seeking to alter functionality in a way that raises concerns about the privacy and security of users.” In response to the report, Meta spokesperson Andy Stone stated that “What Apple is actually saying is they don’t believe in interoperability.” This week the European Commission also shared preliminary findings on how Apple should handle the interoperability issue, and set a deadline of January 9 for interested parties to provide feedback.
|
|
|
|
| |
