OpenAI Says Next Reasoning Model Coming in a "Couple of Weeks"

0:00Today on the AI Daily Brief, the divergence between opinion of the general public and AI experts when it comes to AI. Before that, in the headlines, OpenAI is pushing out some new models much faster than we thought. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:23Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes. Quick announcement before we dive in. For a while, some of you have been asking me for an ad-free version of the show. Obviously, there are a million platforms that allow you to do that now. And so I am happy to share that I will be now offering an ad-free version via Patreon. At the beginning, the Patreon will just be for the ad-free version, but over time, if people are interested, I'll potentially expand it to bonus content and interactive stuff and all the things that Patreons normally have.

0:51But for now, I'm just trying to match the set of you who have said things like, I would pay to not hear the ads, which I totally understand as a listener. You can find it at patreon.com slash aidailybrief. Now to our topics, though. As I was prepping this show, we got an announcement from Sam Altman that was pretty interesting, and I think supports the feeling that many of us are having that we are definitely in another moment of acceleration. Altman tweeted, Change of plans. We're going to release 03 and 04 mini after all, probably in a couple of weeks, and then do GPT-5 in a few months. There are a bunch of reasons for this, but the most exciting one is that we're going to be able to make GPT-5 much better than we originally thought.

1:27We also found it harder than we thought it was going to be to smoothly integrate everything, and we want to make sure we have enough capacity to support what we expect to be unprecedented demand. We were really able to improve on what we previewed for O3 in many ways. I think people will be happy. So then we are weeks away from updated OpenAI reasoning models, and it sounds like just months away from their first model that integrates reasoning and non-reasoning models altogether natively. They definitely seem confident in the quality. Not only are they moving up releases, but they're also explicitly saying that things are better than they thought.

1:59This is the opposite of trying to hedge expectations. This is going in the other direction. Sam is talking about unprecedented demand, which coming off of their insane announcement that in the last week alone, 150 million ChatGPT users have generated over 700 million images is a pretty reasonable bet, but still it's extremely notable. What's more, you have folks on the OpenAI team who are even giving away some little hints. Rohan Pandey writes, If you enjoyed reading DeepSeek Chains of Thought, I think you'll absolutely love watching O3 do its thing. Not only is this interesting in terms of giving us some insight around how O3 functions, but it's also an OpenAI staffer explicitly acknowledging that people really liked seeing DeepSeek thinking.

2:36When someone responded, still waiting for the open source release, which you'll remember OpenAI said that they were going to release their first open weights model in some time, Rohan said also on its way, which at the risk of overreading into a small choice of wording, suggests that none of these are that open source release. Some folks out there just started speculating around or sharing what they wanted out of these models. ARC Prize President Greg Kamrat writes, my wishlist, reasoning by token limits, not buckets, parameter for samples, pricing from day one, reasonable rate limits, webhook support.

3:05I'm okay giving up latency. I want horsepower. Maybe even a more common reaction was the assumption that this was about competitive pressure. Khalifa Manna writes, bunch of reasons equals Gemini 2.5 and Sonnet 3.7 beating them in just about everything. Entrepreneur Bindu Reddy writes, Google is forcing OpenAI to change plans. OpenAI will change plans and release 03 and 04 mini instead of shipping GPT-5, which was supposed to combine all their models. The big reason is Gemini 2.5 Pro, a thinking model that is very good at benchmark maxing, being number one on benchmarks. Unfortunately, we have found that all thinking models don't work very well on complex real-world coding tasks and don't do well at tool use.

3:41Plus, no one has PhD-level reasoning tasks that the benchmarks measure these models on. All this means that Sonnet is still number one IRL. Hopefully, Google and OpenAI will train better base models shortly. Now, one thing that's not clear, Bindu is assuming that the change of plans Altman is talking about is GPT-5 not combining the reasoning and non-reasoning models. But that's not super clear to me. Although it is one possible read, I'm sure we'll get clarification there. And in any case, more models are on their way. For my money, though, to the extent that OpenAI is feeling competitive pressure, it is not from their American peers.

4:14It's from DeepSeek and from China. Given that basically all of China is rallying around DeepSeq and integrating their approaches into other models as well, I understand why that might be making them feel a little pressure. Whatever the reason for the change of plans, I am here for it. Can't wait to try these new models. Bring on the acceleration, baby. Next up, what was supposed to be our title story in the headlines. There are more rumblings that Microsoft is pulling back from data center investment. Bloomberg is reporting that Microsoft is getting cold feet about data center construction with delays or cancellations affecting six sites worldwide.

4:45Sources say that sites in Indonesia, the UK, Australia, Illinois, North Dakota, and Wisconsin have been impacted. Earlier this year, market analysts reported that Microsoft had canceled options on data center leases, representing a couple of hundred megawatts of capacity. The number later ballooned to two gigawatts as TD Cowan rumored that Microsoft had passed on additional options. The research firm said that these deals likely represented, quote, data center oversupply relative to its current demand forecast. Just prior to the CoreWeave IPO, we learned that Microsoft had failed to pick up a$12 billion option with the company.

5:14This being a huge problem for CoreWeave as Microsoft represented over 60 % of their revenue. OpenAI stepped in to fill the void, inking a deal of the same size a week before the company went public. Aside from any potential drama there, Microsoft seems to be sending a pretty clear signal that they don't want to be expanding their data center footprint in the current environment. A spokesperson for the company failed to explain the moves, stating, we plan our data center capacity needs years in advance to ensure we have sufficient infrastructure in the right places. As AI demand continues to grow and our data center presence continues to expand, the changes we have made demonstrate the flexibility of our strategy.

5:46Now, you might remember back in February, during his interview on the Dwarkesh podcast, Microsoft CEO Satya Nadella guaranteed that data centers will be overbuilt. He compared the infrastructure to internet cable and even railways, noting that overbuilding is a constant feature of frontier tech. He said that this is why, quote, I'm so excited to be a leaser. Now, it's understandable if Microsoft doesn't feel the need to be left holding the bag with excess infrastructure, but some of these deals represent serious sunk costs. The company has already spent$262 million to expand their site in Wisconsin, with$40 million going to the concrete pour alone.

6:17The total cost of the project is$3.3 billion, but a large portion of that will be the purchase of GPUs to fill the racks. The company claimed that the construction is merely delayed, not cancelled, with the data center to come online next year. Microsoft said that the$1.7 billion Indonesian site is also on pause, but will be completed in the second quarter of 2025. The company is still committed to their $80 billion of spend on AI infrastructure during the fiscal year, which winds up in June, but have flagged that the following year should feature a slowdown in CapEx as the focus shifts from new construction to fitting out existing facilities.

6:48CoreWeave CEO Michael Entrader said that the slowdown is more specific to Microsoft than the broader industry, commenting, it's pretty localized and their relationship with OpenAI has changed. So it stands to reason that there would be some noise. Ed Socia, a director at intelligence firm Data Center Hawk, said he's observing a shift in priorities across the industry. He claimed that AI companies are tweaking their data center plans to cut costs and prioritize projects that can come online sooner, adding, you may have initially thought one data center project would be the fastest speed to market, but then you realized that the labor supply chain and power delivery wasn't as quick as you thought, then you would have to shift in the short term to focus on other markets.

7:21An analyst going by TechFundOnX claimed that this is precisely the issue Microsoft has run into, posting,

7:40Candy Factory responded,

7:48So what are the possibilities here? One is that Microsoft thinks that the data center buildout is overblown and that we're in a bubble. That's certainly what a lot of people are trying to make of this. It looks to me more like, though, that Microsoft, likely led by Amy Hood, yes, just doesn't want to be the weakest firm out there if demand starts collapsing. I think it's far less likely that we're seeing some native slowdown in demand, especially given how absolutely hungry for power companies like OpenAI seem to be. But I think, frankly, every company is thinking about the potential for tariff-induced recession.

8:17I think big companies are starting to think about cost-cutting because of that, and I think Microsoft seems more worried than most about the risk of demand falling short. Anyway, worth continuing to watch, but that's my read for now. Finally today, TSMC has reportedly formed a tentative partnership with Intel to revitalize the struggling U.S. chipmaker's manufacturing wing. The information reports that the two companies will form a joint venture to operate Intel's chip foundries. Other unnamed U.S. semiconductor firms will also take part in the new partnership. Talks are in the earliest stages and the exact structure of the venture hasn't been settled, but this still is very good news for Intel.

8:49Reports state that TSMC could take a 20 % stake in the new organization. they would provide Intel with the information to build TSMC's chips on their equipment, along with training assistance for Intel personnel. Partnering with TSMC was the widely reported preference of the Trump administration earlier this year to avoid losing the only U.S. owned chip producer. Keeping TSMC as a minority stakeholder would also ensure that a foreign company doesn't control the one-time U.S. industry champion. Since then, the firm has appointed industry veteran Lip Bhutan as CEO and given him a mandate to turn the company around.

9:19Earlier this week at the Intel Vision conference, Tan said that he was planning to spin off assets that weren't core to the mission of the company. During his keynote speech, he said, As we strengthen our Intel products, we are equally committed to building a great foundry. Global demand for chip production is growing, and you need supply chains that are flexible, resilient, and secure. Intel Foundry plays a crucial role. We will continue to advance our foundry strategy to meet our needs. Now, Intel chip foundries have been bleeding cash for years. The division posted an operating loss of$13.4 billion last year, and the company says they don't expect to turn a profit until 2027.

9:49Amid all this news, Intel was a singular spot of green in a sea of absolute red yesterday as tech stocks sold off. The stock was up 6.6 % on the news, with investors likely thinking that any plan is a good plan as Intel struggles to survive. For now, that is going to do it for today's AI Dealer Brief Headlines edition. Next up, the main episode. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded. Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex.

10:25That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC 2 and ISO 27001. Centralized security workflows complete questionnaires up to 5x faster and proactively manage vendor risk. Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company. Join over 9 ,000 global companies like Atlassian, Quora, and Factory who use Vanta to manage risk, improve security in real time.

11:02For a limited time, this audience gets$1 ,000 off Vanta at vanta.com slash nlw. That's v-a-n-t-a dot com slash nlw for$1 ,000 off. Today's episode is brought to you by Superintelligent and our friends at Lindy. You've heard me talk about Lindy a couple times over the past few weeks. They are an agent builder platform that can help you build agents to automate a huge variety of functions, workflows, services, from data entry to lead generation to customer service. They're basically a platform for building all of the agents that you might want in the future. We had previously had an offer where for AI Daily Brief listeners, if you email us at agent at bsuper.ai with the word Lindy in the title, we'll connect you to the Lindy team who can help you build a specific custom agent in a matter of days or weeks for less than$20 ,000 a year.

11:52It's a great way to dive in and just get your feet wet with agents. Today, however, Lindy announced something very cool, specifically agent swarms. If you've heard my Dr. Strange theory of AI agent work, this is basically that brought to life. Instead of an agent doing research on sales leads one at a time, you could spin up a swarm of 200 Lindy agents, all doing different research and writing custom emails so that in a matter of seconds, you could be sitting on 200 custom emails ready to go for each of those different leads. We've been playing around with swarms for research, for content production, for sales, and it is very much a glimpse of the future.

12:26So again, if you are interested in Lindy, send us a note at agent at bsuper.ai, put Lindy in the title, and we will get right back to you. Hey, listeners. Want to supercharge your business with AI? In our fast-paced world, having a solid AI plan can make all the difference. Enabling organizations to create new value, grow, and stay ahead of the competition is what it's all about. KPMG is here to help you create an AI strategy that really works. Don't wait. Now's the time to get ahead. Check out real stories from KPMG of how AI is driving success with its clients at kpmg.us.ai. Again, that's www.kpmg.us slash AI.

13:05Now, back to the show. Welcome back to the AI Daily Brief. Artificial intelligence is one of those areas where the general public has had to become aware of and dig into a new technology way faster than they normally have in the past. Now, obviously, this is expressed in terms of the number of different people who are users of these tools. Chat GPT is already at 500 million weekly users, for example, which is just an absolutely insane growth curve, unlike anything we've ever seen. But there are still a huge amount of questions around AI, even among people who are users of the technology. It's evolving at such a fast pace that it's quite natural, I think, for people not to be sure about how it's going to play out.

13:43On top of that, you have people screaming about IP theft over here or robots taking over over there. And so trying to get a handle on how people actually think and feel about this space can be pretty tricky, which is why I think that this new research from Pew is very valuable. Pew Research has just dropped a study they call How the U.S. Public and AI Experts View Artificial Intelligence. Now, Pew has been tracking AI adoption actually since before ChatGPT, and so is in a better position than most to actually have a sort of longitudinal view of things. This study was a deep dive on how the U.S.

14:14public views AI and how that differs from the opinions of experts in the field. Now, on the public side, Pew surveyed a sample of 5 ,400 people, evenly distributed between genders, and with a good spread of age and ethnicity. The way they defined AI experts was actually kind of interesting. They write, We defined AI experts as individuals who demonstrate expertise via their work or research in artificial intelligence or related fields. We focused on only those who live in the United States. To identify these individuals, we created a list of authors and presenters at 21 AI-focused conferences from 2023 to 2024.

14:45This list was developed in consultation with project advisors. Now, obviously, there are always reasons to quibble with any statistical methodology, but I think that this is a fairly decent approach. I think expert is an inherently subjective term, and you could do worse than deciding that expert means people who other people thought were worth inviting to speak at a conference. Now, once again, unfortunately, the biggest gripe I have with this, and my goodness, I wish that surveyors would increase their speed to reports. I don't know, maybe use some AI or something, because this data was all collected last fall.

15:15In other words, we had barely touched agents. OpenAI had just released, and not even widely, their first ever reasoning model. No one had ever heard of DeepSeek. It was just, in short, a different world. And so I think that is a big grain of salt you have to take all of this with. Still, to the extent that we view this as an interesting case study in the gap between experts and the general public, I think it's pretty interesting. Pew started with the biggest topic. Basically, is AI going to be a net negative or a net positive thing? AI experts were wildly more optimistic than the general public.

15:4556 % of experts said that AI was going to be positive versus just 17 % of the normal public, who I will occasionally call normies, which is absolutely a term of endearment. Now, in those normie sections, 35 % said that they thought that AI would be more negative, as compared to 15 % of AI experts who said that it would be more negative than positive. About a third of US adults said that it would be equally negative and positive, versus 23 % of AI experts. So really stark difference here. The reasoning might be in the eye of the beholder? Is it that the more you know about AI, the more optimistic you are?

16:17Or is this just people being excited about their bags? That's a bit of a glass-half-full, glass-half-empty debate, but it is notable that about 76 % of AI experts think that AI will benefit them personally, as opposed to just 24 % of US adults who think that AI will benefit them personally. Now, asking somewhat more granular questions, when asked whether AI would make humans more productive? A resounding majority of AI experts, 74%, said yes. That's compared to just 17 % of general US adults. To me, this is one of the questions that most dramatically shows the experience gap between these two groups.

16:50Hold aside any expertise. Now, one area where no group was particularly optimistic was when answering the questions, will AI make humans happier? Only 6 % of non-expert US adults said that it was extremely or very likely, but only 22 % of experts said that it was very or extremely likely, with 35 % of experts saying that it was not to or not at all likely that AI would make people happier. Now, happier is one of the more subjective terms, so I don't know how much you can read into this, but it certainly shows that some types of optimism are at least a little bit limited. In general, experts think AI is going to be a lot more impactful than normies do.

17:25For example, just 23 % of US adults said that AI would be somewhat or very positive in terms of how people do their jobs, as opposed to 73 % of experts. Only 21 % of U.S. adults thought that AI's impact on the economy would be somewhat or very positive, as opposed to 69 % of AI experts. The one area where there was the most optimism was around medical care. 84 % of AI experts said that AI would be very or somewhat positive, and a full 44 % of general U.S. adults also said the same. The area where there was the most alignment of AI's negative impact was around news, with just 10 % of general U.S.

17:59adults and 18 % of AI experts saying that AI would have a positive impact on the news. and the lowest surveyed area was around elections, with only 9 % of general U.S. adults and only 11 % of AI experts saying that AI would have a positive effect. Now, as you might imagine, one of the big questions was around jobs. Experts have a fairly mixed view on how AI is going to impact jobs. When asked what they thought the impact of AI on jobs would be, 39 % of experts said that there would be fewer jobs versus 33 % who said that there would be not much difference versus 19 % who said that there would be more.

18:30A full 64 % of non-expert U.S. adults thought that AI would mean fewer jobs. Now, I think that this intuitively reflects people's sense that companies are always looking for ways to cut costs. And if AI really does get as good as it seemed like it's going to, of course, there's going to be fewer jobs. What that perspective doesn't take into account, and what it's likely that many of the AI experts are taking into account, is the countervailing forces of new jobs and new industries that will be created by the technology. One area of commonality was inaccurate information, misinformation, impersonation, data misuse.

19:0066 % of adults overall and 70 % of experts are highly concerned about people getting inaccurate information from AI. Interestingly, the general public is way more worried than AI experts are about the loss of human connection. 57 % of the public is highly concerned about that outcome of AI, as opposed to just 37 % among experts. 55 % of both groups say that they're highly concerned with bias in decisions made by AI. And again, the one area of biggest divergence between these two groups is around job loss. Just 25 % of AI experts were extremely or very concerned about AI-related job loss, as opposed to 56 % of general U.S.

19:34adults. Overall, even though this is from October, I think it's an extremely useful survey. Now, if I had to guess, what do I think would have changed in that period of time? My guess is that overall, we would see a slight convergence of opinions, simply on the basis of the fact that more people over the last six months have been exposed. ChatGPT has basically doubled its usage in that time. and so I would expect to see some changes in opinion from the U.S. public. Unfortunately, based on the way that people still do surveys, we'll probably have to wait another year to find out. Anyways, friends, that is going to do it for today's AI Daily Brief.

20:06Appreciate you listening as always, and until next time, peace.