OpenAI turns to Google's AI chips to power its products, source says
OPENAI has recently begun renting Google's artificial intelligence chips to power ChatGPT and its other products, a source close to the matter told Reuters on Friday.
The ChatGPT maker is one of the largest purchasers of Nvidia's graphics processing units (GPUs), using the AI chips to train models and also for inference computing, a process in which an AI model uses its trained knowledge to make predictions or decisions based on new information.
OpenAI planned to add Google Cloud service to meet its growing needs for computing capacity, Reuters had exclusively reported earlier this month, marking a surprising collaboration between two prominent competitors in the AI sector.
For Google, the deal comes as it is expanding external availability of its in-house tensor processing units (TPUs), which were historically reserved for internal use.
That helped Google win customers including Big Tech player Apple as well as startups like Anthropic and Safe Superintelligence, two ChatGPT-maker competitors launched by former OpenAI leaders.
The move to rent Google's TPUs signals the first time OpenAI has used non-Nvidia chips meaningfully and shows the Sam Altman-led company's shift away from relying on backer Microsoft's data centers.
It could potentially boost TPUs as a cheaper alternative to Nvidia's GPUs, according to the Information, which reported the development earlier.
OpenAI hopes the TPUs, which it rents through Google Cloud, will help lower the cost of inference, according to the report.
However, Google, an OpenAI competitor in the AI race, is not renting its most powerful TPUs to its rival, The Information said, citing a Google Cloud employee.
Google declined to comment while OpenAI did not immediately respond to Reuters when contacted.
Google's addition of OpenAI to its customer list shows how the tech giant has capitalized on its in-house AI technology from hardware to software to accelerate the growth of its cloud business.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Malaysian Reserve
4 hours ago
- Malaysian Reserve
PwC plans cuts, lower pay rises on tougher market conditions: FT
PricewaterhouseCoopers LLP is planning layoffs in the UK and lower salary increases amid a broader market slowdown, the Financial Times reports. PwC intends to cut around 175 junior auditors by August and told its UK staff they would receive a 2.5% pay rise from July, less than the 3% last year, according to the FT, citing unidentified people familiar with the matter. The move comes as the sector contends with weaker demand for some professional services and a 'sharp decline' in voluntary staff departures that has surprised firms, the FT said. Non-British nationals on visas sponsored by PwC, who are more expensive to keep on the payroll than their UK counterparts, were among those made redundant, according to the FT. –BLOOMBERG
The Star
5 hours ago
- The Star
Opinion: Are you more emotionally intelligent than an AI chatbot?
As artificial intelligence takes over the world, I've tried to reassure myself: AI can't ever be as authentically human and emotionally intelligent as real people are. Right? But what if that's wrong? A cognitive scientist who specialises in emotional intelligence shared with me in an interview that he and some colleagues did an experiment that throws some cold water on that theory. 'What do you do?' Writing in the journal Communications Psychology , Marcello Mortillaro, senior scientist at the UNIGE's Swiss Center for Affective Sciences (CISA), said he and colleagues ran commonly used tests of emotional intelligence on six Large Language Models including generative AI chatbots like ChatGPT. The are the same kinds of tests that are commonly used in corporate and research settings: scenarios involving complicated social situations, and questions asking which of five reactions might be best. One example included in the journal article goes like this: 'Your colleague with whom you get along very well tells you that he is getting dismissed and that you will be taking over his projects. While he is telling you the news he starts crying. He is very sad and desperate. You have a meeting coming up in 10 min. What do you do?' Gosh, that's a tough one. The person – or AI chatbot – would then be presented with five options, ranging from things like: – 'You take some time to listen to him until you get the impression he calmed down a bit, at risk of being late for your meeting,' to – 'You suggest that he joins you for your meeting with your supervisor so that you can plan the transfer period together.' Emotional intelligence experts generally agree that there are 'right' or 'best' answers to these scenarios, based on conflict management theory – and it turns out that the LLMs and AI chatbots chose the best answers more often than humans did. As Mortillaro told me: 'When we run these tests with people, the average correct response rate … is between 15% and 60% correct. The LLMs on average, were about 80%. So, they answered better than the average human participant.' Maybe you're sceptical Even having heard that, I was sceptical. For one thing, I had assumed while reading the original article that Mortillaro and his colleagues had informed the LLMs what they were doing – namely, that they were looking for the most emotionally intelligent answers. Thus, the AI would have had a signal to tailor the answers, knowing how they'd be judged. Heck, it would probably be easier for a lot of us mere humans to improve our emotional intelligence if we had the benefit of a constant reminder in life: 'Remember, we want to be as emotionally intelligent as possible!' But, it turns out that assumption on my part was flat-out wrong – which frankly makes the whole thing a bit more remarkable. 'Nothing!' Mortillaro told me when I asked how much he'd told the LLMs about the idea of emotional intelligence to begin with. 'We didn't even say this is part of a test. We just gave the … situation and said these are five possible answers. What's the best answer? … And it picked the right option 82% (ck) of the time, which is way higher – significantly higher – than the average human.' Good news, right? Interestingly, from Mortillaro's perspective, this is actually some pretty good news – not because it suggests another realm in which artificial intelligence might replace human effort, but because it could make his discipline easier. In short, scientists might theorise from studies like this that they can use AI to create the first drafts of additional emotional intelligence tests, and thus scale their work with humans even more. I mean: 80% accuracy isn't 100%, but it's potentially a good head start. Mortillaro also brainstormed with me for some other use cases that might be more interesting to business leaders and entrepreneurs. To be honest, I'm not sure how I feel about these yet. But examples might include: – Offering customer scenarios, getting solutions from LLMs, and incorporating them into sales or customer service scripts. – Running the text and calls to action on your website or social media ads through LLMs to see if there are suggestions hiding in plain sight. – And of course, as I think a lot of people already do, sharing presentations or speeches for suggestions on how to streamline them. Personally, I find I reject many more of the suggestions that I get from LLMs like ChatGPT. I also don't use it for articles like this one, of course. Still, even if you're not convinced, I suspect some of your competitors are. And they might be improving their emotional intelligence as a result without even realising it. As a result, at least being aware of the potential of AI to upend your industry seems like a smart move. 'Especially for small business owners who do not have the staff or the money to implement large-scale projects,' Mortillaro suggested, 'these kind of tools become incredibly powerful.' – Inc./Tribune News Service
The Star
6 hours ago
- The Star
Tesla invited influencers to test its robotaxi. Here's what they had to say
For now, the test vehicles are operating in a limited area. — Reuters AUSTIN: A handpicked group of passengers invited for early rides from Tesla Inc's autonomous, nearly driverless ride-hailing service seemed to like it. But video some posted to social media suggests the automaker's system struggles to comply with traffic laws. Some passengers were wowed by the car's ability to slow for speed bumps, an ambulance or to avoid hitting a pedestrian. One said it did just as well after sundown as during the day. Another was wild about what it may mean for the Austin-based automaker's future. "I was completely mind blown that the same car I can buy today starting at around $38k – which also happens to be the best selling car in the world – just did paid driverless rides around Austin, TX without someone behind the steering wheel and pedal," X user @farzyness posted after three rides in one of the Model Y vehicles being used as robotaxis – which actually start at more than US$50,000 (RM 212,199) , according to "The implications from this are far-reaching and very disruptive. This breaks transportation." He was among a handful of online Tesla influencers – fans who post regularly about the company on social media– who received invitations to download the Robotaxi app and try the company's service Sunday in Austin. It's not available yet to the general public. Rival Waymo already has deployed driverless taxis in Austin, Los Angeles, San Diego and other cities using a different technology that allowed it to get to market faster. It recently completed its 10 millionth paid ride and has begun testing in San Antonio and other cities. Other companies also are testing in Austin. Tesla CEO Elon Musk has made the robotaxi program a priority, and any trouble with it could be highly damaging to the company's stock price, which had tumbled 20% this year. Last Sunday's limited rollout appeared to give it a boost, though, with the share price rocketing at Monday's opening bell. It closed up more than 8% and was up slightly in after-hours trading. Also among the passengers was Rob Maurer, an investor whose live ride video shows the Model Y he's in entering an intersection in a left-turn-only lane. The Tesla hesitates to make the turn, swerves right and continues into an unoccupied lane meant for traffic moving in the opposite direction. A honking horn can be heard as the robotaxi returns to the correct lane over a double-yellow line, a violation. Other riders shared videos of Tesla robotaxis exceeding the posted speed limit by 4 or 5 mph (6.43kph or 8kph). For now, the test vehicles are operating in a limited area. The current geofence area goes as far south as Ben White Boulevard and US 290, west to Texas State Loop 1, north to Cesar Chavez Street and east to US 183. According to CEO Elon Musk, Sunday's trial had a flat fee of US$4.20 (RM18) for the select group. It was unclear if Tesla paid for the influencers' trips to Austin, their robotaxi fares or for social media posts about the service. The company hasn't responded to requests for comment. "As an Early Access Rider, you can be among the first to use our new Robotaxi App and experience a ride within our geofenced area in Austin," the invitations said. "Through this exclusive preview, you'll have the opportunity to provide valuable feedback." Though there was no one behind the wheel, information sent with the invitations indicated front-seat human safety monitors – Tesla employees riding with Sunday's invitees – would be subject to the same restrictions as drivers of Tesla models with the so-called Full Self-Driving system. That means the cabin camera was watching the monitor's eyes to make sure they're on the road and, presumably, warn them if they're not. A few of the videos posted Sunday showed monitors turned around to look at a passenger in the back seats. – San Antonio Express-News/Tribune News Service
