Latest / AI World Today: Daily News / Google Unveils Gemini 4 Carbon, ChatGPT Adds API Credits | AI World Today: Daily News #10
Transcript
- Hi, and welcome. This is AI World Today: Daily News. Today, Google's next big model is going to cyber defenders first. Some ChatGPT subscribers can claim credits to build their own AI tools. And the latest earnings show how quickly cloud revenue, and the infrastructure bill, are growing. We also look at the platforms developers rely on, and a museum partnership that brings the conversation back to what people might actually do with all this technology. Let's get into it. Google announced Gemini four Carbon yesterday. The company says it is built for demanding engineering work and for finding, checking and fixing software vulnerabilities. In internal use, Google says Carbon has worked through complex tasks for days, including tracking down subtle bugs in large codebases. That's an ambitious claim. The useful test will be whether teams can reproduce those results on their own software, with changes they can understand and verify. Access is starting this week with a subset of cyber defenders. Google plans to expand through its Fairwind program, but has not announced a date for wider availability to developers, businesses and consumers. Fairwind gives selected organizations early access to advanced capabilities for defensive work. Its priorities include governments, healthcare providers and telecommunications services. The reasoning is straightforward. Tools that help uncover a weakness can also help someone exploit it. Giving defenders an early opportunity to find and repair problems could make a difference. For developers, the interesting shift is how much of a task a model can carry. Imagine asking it to investigate a stubborn bug, trace the cause across several files, propose a repair and explain how to test it. That is the kind of workflow to watch. A convincing demonstration is one step. A patch that survives review and keeps working is the result that matters. Next, a practical update from OpenAI. Eligible ChatGPT Pro and Business subscribers can now claim A P I credits. These pay for supported uses of OpenAI's models through software you build or connect. For the eligible monthly Pro tiers, the full-period allowance is one hundred, two hundred or five hundred dollars, depending on the plan. Business credits are pooled across eligible paid seats, with a maximum of five hundred dollars per workspace per month. There are a few details worth checking before you plan a project around them. Your eligible paid subscription must have been active for at least seven days. The owner needs to activate the benefit and link an A P I organization they own. Availability also depends on country and rollout. The first grant is prorated. Unused credits expire at the end of the applicable billing cycle. ChatGPT Plus is not included in this offer. And this is a promotional benefit, so future availability and allowances can change. These credits do not increase your ChatGPT or signed-in Codex limits. Codex use through A P I keys is excluded. If you already qualify, this could make a small experiment easier to justify. You might try an assistant that sorts customer questions, or a tool that turns meeting notes into a draft project brief. Check the allowance your account actually offers, then give the experiment a clear goal. Now to the money behind the models. Alphabet and Microsoft both reported results yesterday for the quarter ending in September. Both showed strong cloud growth, alongside very large spending on infrastructure. Google Cloud revenue rose one hundred and three percent from a year earlier, reaching about thirty point seven billion dollars. Alphabet says sales of its own T P U systems contributed meaningfully to that growth. At Microsoft, Azure revenue grew forty-eight percent to about thirty-three point two billion dollars. Those are different reporting categories, so the comparison needs care. Neither figure tells us the revenue from AI alone. The cash flow numbers add another dimension. Across Alphabet, cash spent on property and equipment exceeded cash generated by operations. That left quarterly free cash flow at roughly negative nine point one billion dollars. Alphabet remained highly profitable. Profit and free cash flow measure different things, especially when a company is buying expensive infrastructure that it expects to use for years. Using operating cash flow minus cash purchases of property and equipment, Microsoft generated roughly eighteen point six billion dollars in free cash flow. These are company-wide figures. They cannot tell us the standalone profitability of either company's AI business. What they do show is why demand and investment have to be considered together. Fast revenue growth is encouraging. The next question is how much lasting value all that new capacity produces. Next, Nvidia and Hugging Face. Reuters reports, citing two people familiar with the matter, that the United States Justice Department is asking how the proposed acquisition would affect competition. Questions alone establish neither wrongdoing nor a decision to block the deal. Nvidia announced the roughly thirteen-billion-dollar agreement in September. Its filing says completion is expected in the first half of next year, subject to conditions including regulatory approvals. Hugging Face is a place where developers find, share and use models, datasets and applications. Its role makes this relevant well beyond the people buying Nvidia chips. Nvidia has promised that the platform will continue supporting different hardware, clouds and model providers, and that using Nvidia computing will not be required. For builders, the practical issue is whether those choices stay useful and easy to exercise. A model can have openly available weights, but the places where people discover and deploy it still matter. And finally, let's head to Amsterdam. The Rijksmuseum and Prosus announced a three-year AI partnership earlier this week. Specialists from the Prosus AI Talent Lab will work full time on museum projects, exploring ways to help people discover and experience the collection. The work starts by examining the museum's goals, available data and existing workflows. Specific applications follow from that. There is already a useful example of the problem they are trying to address. The museum added visual search to its online collection in December last year, helping people find works with similar shapes, colours, styles or kinds of objects. That feature predates this partnership. No new visitor product or public release date has been announced. But the idea is easy to appreciate. You might recognize a detail in a painting without knowing the artist, the period or the right art-history term. A useful tool could turn that curiosity into a starting point. After all those model launches and billion-dollar figures, that's a good question to keep asking: what becomes easier for the person using the technology? That's it for today. Thanks for listening. If you enjoyed the episode, follow AI World Today: Daily News to keep up with what's happening in AI. You'll find links to the stories and sources in the description, and more news and analysis at AI World Today dot com. This episode is narrated by an AI voice. See you tomorrow.