Everyone's Using AI But No One's Quite Sure What to Think About It

0:00Today on the AI Daily Brief, everyone's using AI, but no one quite knows what to think about it. Before that end of headlines, Claude gets a new million token context window and many, many more updates. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

0:20Hey hello friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Blitzy and Superintelligent. To get an ad-free version of the show, go to patreon.com slash ai daily brief. And speaking of KPMG, a quick announcement today. KPMG has just launched their You Can With AI podcast. It's a seven-part podcast hosted by me. Basically, the idea of this thing was to give people inside enterprises or organizations a kind of primer across a whole slew of issues that relate to AI in the actual enterprise. So there's an episode on strategy, an episode on data readiness, an episode on agents in practice, trust and governance, and a bunch more.

1:02And while I think it'll be really useful for folks like you who are regular listeners of the AI Daily Brief, where I think it might be even more valuable is for your peers and colleagues that you are trying to get up to speed fast. That's kind of how we design this thing. Anyways, you'll be hearing about this more in the coming weeks. I believe on Saturday in this slot that is normally empty, because you know we only do six shows a week, I'll be dropping at least part of an episode so you can get a feel and a flavor for You Can with AI. But definitely check it out. you can get it at kpmg.us slash AI podcasts.

1:34So that's www.kpmg.us slash AI podcasts. Now, finally, a note about today's episode. Yesterday, of course, it was all just GPT-5 prompting tips. Because of that, there were a lot of headline stories to cover. So today, we're kind of reversing the flow. The headlines are actually longer than the main episode, but there is a headlines and a main episode. So without any further ado, let's get into all the news from Anthropic, OpenAI, Gemini, XAI, and more. It has been OpenAI, OpenAI, OpenAI for the last week, and Anthropic is saying, actually, if you really want the juice, come and look at what we got for you over here.

2:09In the next escalation of the AI coding wars, Anthropic has now unleashed a million token context window. The company announced earlier this week that its Claude Sonnet 4 model, which is the preferred model for many software engineers, can now process up to a million tokens of context in a single request. That is a 5x increase that will allow people to look at and ingest entire codebases, media documents, etc. without having to break it into smaller chunks. In fact, this is equivalent to a codebase of something like 75 ,000 lines of code. Now, both OpenAI and Google offer million-token context windows already, however, Anthropic claims their model outperforms.

2:48They said that they saw 100 % performance on internal needle in a haystack evaluations, demonstrating Claude's ability to accurately search the entire context window. Claude's product lead, Brad Abrams, said, This is really cool because it's one of the big barriers I've seen with customers. They have to break up their problems into these small chunks with our existing context window, and with a million tokens, the model can handle the entire scope of the context. In other words, handle problems at their full scale. In addition to working in large codebases, long context is also useful for more complex agentic tasks.

3:19This could, for example, make Claude uniquely useful for lengthy tasks that run autonomously in the background, which is quickly becoming a more common part of AI coding agents. Still, overall, the upgrade highlights how cutthroat the competition for the AI coding market is. Remember, in advance of GPT-5, Anthropic also released Opus 4.1. Abrams told TechCrunch that he expects the AI coding platforms will get a lot of benefit from the upgrade. And when he was asked if GPT-5 has eaten into Claude's API usage, he downplayed the concern, commenting that he's quote, really happy with the API business and the way it's been growing.

3:52When asked whether the launch of GPT-5 prompted this change, Abrams responded, look, we're moving at a fast clip here and just listening to customer feedback. Just two and a half months ago, we launched Opus 4 and Sonnet 4, and one week ago, we launched Opus 4.1, and now we're launching this 1 million context. I think it's just showing how our enterprise customers are really eager to get these improvements and we're doing the best we can to get them out. Now, currently, the feature is only available to certain customers through the API, namely high-paying customers in Tier 4 and with custom rate limits, but Anthropic promised a broader rollout in the coming weeks.

4:23Now, as we've been discussing, in addition to just performance, the other dimension of the AI coding wars is around cost, or at least it might be at some point. Right now, it hasn't hurt Anthropic's popularity that they're comparatively more expensive, but some pointed out that for context over 200 ,000 input tokens, they actually doubled the price, leading some to wonder if they can sustain these comparatively high prices relative to cheaper competitors. A couple other Anthropic stories. First, the company has price-matched OpenAI and will be offering Claude to the government for just a dollar.

4:54Just about a week after OpenAI offered to provide their models to the government for a nominal fee, Anthropic has matched. As a kicker, while OpenAI only made their offer to federal agencies, Anthropic is extending it to all three branches of government, including the Judiciary and Congress. Writes Anthropic, As AI adoption leads to transformation across industries, we want to ensure that federal workers can fully harness these capabilities to better serve the American people. By removing cost barriers, we're enabling the government to access the same advanced AI that's already proving its value in the private sector.

5:24Like OpenAI, Anthropic was added to the General Services Administration schedule, making them eligible for streamlined procurement. The Verge writes, As the government struggles over how and whether to regulate AI, there could also be a soft power benefit to getting its workers familiar with and reliant on these services, and perhaps, by extension, more reluctant to kneecap them. Now, the one other Anthropic story that I think is probably more interesting than many others will is that the company has acquihired the team behind Humanloop. Humanloop is a five-year-old startup working on prompt management, evals, and observability for enterprise LLM deployments.

5:58An Anthropic spokesperson said that they hadn't acquired any of the startup's assets or IP. However, the founders and staff will bring their expertise in building tools that help enterprises run reliable and safe AI systems at scale. Now, what's interesting about this is that I think that it highlights that there are multiple things going on inside these foundation model companies all at the same time. On the one hand, they are looking for continued model advances to ensure that Claude 5 is better than GBT 5. But at the same time, they're competing in a traditional business environment as well, where they are competing to win customers.

6:30Anthropic currently has an interesting wedge because of people's preference for its coding tools that's giving them really strong access to the enterprise. Humanloop to me represents the idea that these companies are going to build not only really high-performance models, but also the tooling around them that allows those models to integrate and actually serve enterprise customers. Evaluation tools are right now one of the biggest gaps when it comes to the enterprise ecosystem. I think it makes sense that Anthropic would try to have more of that capability natively, and I wouldn't be surprised if this is the start of a broader full-stack approach to being able to speak to the LLM and agentic infrastructure needs of those enterprise customers as well.

7:06And now that we've got our Anthropic stories out of the way, let's move over to OpenAI and ChatGPT, where we continue to see updates following the GPT-5 rollout. It feels like kind of one by one, OpenAI is unwinding the big product decisions from the GPT-5 launch. The company, as we saw, reinstated GPT-4-0 after an intense wave of online criticism, and now they're returning control back to the users, giving back a version of the model selector. On Tuesday, Sam Altman wrote, Updates to ChatGPT, you can now choose between Auto, Fast, and Thinking for GPT-5. Most people will want Auto, but the additional control will be useful for some people.

7:40Now just to give you a sense of this, the model picker is now, if anything, more extensive than it used to be. Basically, you have two separate model selectors, one for GPT-5 and one for legacy models. Under GPT-5, you now have Auto, Fast, Thinking Mini, Thinking, and Pro. OpenAI does give you a little guidance around what each of those means, with, for example, Auto deciding how long to think, and the others all thinking progressively longer as you go from Fast to Pro. Under Legacy models, you still have access to 4.0, 4.1, 4.5, 0.3, and 0.4 Mini. And it kind of brings up the question for me of whether, in hindsight, instead of getting rid of the model selector, the right approach might have been to just make the model selector UX better.

8:20This is a really tricky product question. The Steve Jobs school of things would be to basically do what OpenAI did, get rid of complexity and who cares if the pro users complain. But LLMs just might not work that way. It may be that because different use cases require different models and different approaches, it would be a better approach to just help people learn which model is going to work for them but leave them with some amount of control. I'm not sure and ultimately it's really easy to armchair product management, but the point is for now, the model selector is back and bigger than ever.

8:51The company did also acknowledge that in the future, before deprecating models, they would give more notice. OpenAI's head of ChatGPT, Nick Turley, said, in retrospect, not continuing to offer 4.0, at least in the interim, was a miss. He added that they were surprised at the level of attachment people had to the model, saying, it's not just that change is difficult for folks, it's about the fact that people can have such a strong feeling about the personality of a model. Now, one of the conversations you've seen a lot is that maybe GPT-5 was all about cost-cutting. That basically they were forced to do this and try to present it like a brand new model, but really it was all about just getting the cost of compute down.

9:26Now, as I tweeted earlier, I think that cost-cutting is a pejorative way of saying what you could also describe as prioritizing efficiency as AI workloads become ubiquitous. What I mean by that is that the more the time goes on, the more things people are using AI for. AI use, in other words, is compounding. You use a little of it and you want more of it, and then you want a lot more of it. And at some point, patterns are going to shift and people are going to have to choose to prioritize efficiency and cost over just the state of the art, even if they're not totally there yet. Regardless, the team at OpenAI has been very clear that this was not a cost issue.

9:58Turley said in that same Burge interview, It definitely wasn't a cost thing. In fact, the main thing we were striving for and we've been striving for for a long time is simplicity. He went on to reiterate that normal people, i.e. the people who weren't perpetually on Twitter or Reddit, had given them a lot of feedback that it was overwhelming to have to figure out what model to use. One takeaway from the whole thing is that it's probably not going to work to totally deprioritize power users for the sake of empowering regular users. The Wall Street Journal presents a set of anecdotes on how GPT-5 was received in the business community.

10:29Juliet Haas, an account strategy coordinator at a communications and crisis management agency, discussed revisiting a business development prompt, writes the WSJ, with GPT-4, the response suggested that she build strong industry connections and emphasize the importance of relationship building, while GPT-5 delivered a checklist. Haas said, the AI treated finding distressed companies more like a data science problem rather than understanding the fundamental considerations of relationships and timing. Yet more evidence that the model issue is not just a divide between business and non-business uses.

10:59Now, in the wake of GPT-5, one of the big conversations is whether existing architectures can actually get us to AGI. That's prompted a much bigger conversation around things like memory, which is something we're going to be getting into in the next couple of days, but on the memory front, Google has finally rolled out an automatic memory feature for Gemini. With the feature turned on, Gemini will now automatically remember user preferences and recall previous conversations. Until now, Gemini users had to specifically prompt the chatbot to put something in memory. The same UX change was made by OpenAI in April of this year, with Anthropic following suit last week as well.

11:33Michael Solisky, the Senior Director of Product Management for the Gemini app, said that the change was part of plans to make it more personalized. In the announcement blog post, he wrote, At IO, we introduced our vision for the Gemini app, to create an AI assistant that learns and truly understands you, not one that just responds to your prompt in the same way it would anyone else's prompt. Now, I will say that this is a feature that is not only essential, but also creates significant moat. I've been talking to O3 about a particular strategic consideration, one that I cannot talk about fully here yet, and because it has had that persistent memory, I can jump into a new thread at any point, and it basically has all of the previous context.

12:10That means I don't have to reintroduce it to the context over and over again every time, which is incredibly, incredibly valuable and time-saving. In fact, as I've been trying out alternatives like Grok4, it's made it hard to make a real comparison because I simply don't want to take the time to give Grok4 all of the different context. It's entirely possible that if Grok4 had all that context, it would be as good or better, but frankly, O3 has created a little moat for itself just by having that background. In other words, it's good to see this becoming just total table stakes for these models.

12:41One more today in this extended headlines. XAI's co-founder has left to start a venture firm. On Wednesday, Igor Babushkin wrote, Today was my last day at XAI, the company I helped start with Elon Musk in 2023. I still remember the day I first met Elon. We talked for hours about AI and what the future might hold. We both felt that a new AI company with a different kind of mission was needed. Now, Babushkin had been a leading researcher at Google DeepMind in the early days, and also worked for OpenAI in the lead-up to the release of ChatGPT. The post on X is very long, but in explaining his future plans, he wrote, As future models become more agentic over longer horizons and a wider range of tasks, they will take on more and more powerful capabilities, which will make it critical to study and advance AI safety.

13:24I want to continue on my mission to bring about AI that's safe and beneficial to humanity. I'm announcing the launch of Babushkin Ventures, which supports AI safety research and backs startups in AI and agentic systems that advance humanity and unlock the mysteries of our universe. Now, there's an interesting dimension of this, which we're not going to go too deep into here because there's limited information available, but it sort of seems like Igor maybe got a little bit AI safety-pilled. He said, As I'm heading towards my next chapter, I'm inspired by how my parents immigrated to seek a better world for their children.

13:51Recently, I had dinner with Max Tegmark, founder of the Future of Life Institute. He showed me a photo of his young sons and asked me how we can build AI safely to ensure that our children can flourish. I was deeply moved by his question. Now for XAI, this is their second major departure in a little over a week. Last Tuesday, Robert Keel stepped down as chief legal officer. He posted at the time,

14:20Now of course, two departures in a very short period of time has led to rampant speculation around whether there's more to this story than two people just making personal decisions for themselves. It's totally possible. But at the same time, Rob's explanation was pretty simple. He said, I love my two toddlers and I don't get to see them enough. For anyone who has kids that age and will be in the position to make that sort of decision, I'm sure they can relate. Still, obviously, we will keep an eye on the comings and goings of XAI. For now, they continue to put out top quality models that lots and lots of people are coming to use.

14:52That's however, we're going to do it for today's AI Daily Brief Extended Headlines edition. Next up, a more limited main episode. What if AI wasn't just a buzzword, but a business imperative? On You Can With AI, we take you inside the boardrooms and strategy sessions of the world's most forward-thinking enterprises. Hosted by me, Nathaniel Whittemore, and powered by KPMG, this seven-part series delivers real-world insights from leaders who are scaling AI with purpose, from aligning culture and leadership to building trust, data readiness, and deploying AI agents. Whether you're a C-suite executive, strategist, or innovator, This podcast is your front row seat to the future of enterprise AI.

15:31So tune in at www.kpmg.us slash AI podcasts to start transforming possibility into performance. You can with AI, you can with KPMG. Again, that's www.kpmg.us slash AI podcasts. This episode is brought to you by Blitzy, the enterprise autonomous software development platform with infinite code context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale code bases with millions of lines of code. Enterprise engineering leaders start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan, then generates and precompiles code for each task.

16:14Blitzy delivers 80 % plus of the development work autonomously while providing a guide for the final 20 % of human development work required to complete the sprint. Public companies are achieving a 5x engineering velocity increase when incorporating Blitzy as their free IDE development tool, pairing it with their coding co-pilot of choice to bring an AI-native SDLC into their org. Blitzy is providing a limited-time, 30-day free proof-of-concept for qualifying enterprises. The team will provide a 5x velocity increase on a real development project in your org. Visit Blitzy.com and press Book Demo to learn how Blitzy transforms your SDLC from AI-assisted to AI-native.

16:49That's B-L-I-T-Z-Y dot com.

17:19important use cases with what we call our use case planning reports. These reports are going to help you understand what sort of technical preparation you need to do to be ready for a use case, what challenges you might face in implementation, and whether you should be thinking about building, buying, partnering, or some combination. After that, you can even get a spec document in what we call our technical blueprint that gives either your developers or the developers of the partner you work with what they need to build exactly the agent that you're looking for. If you want to learn more about Superintelligence Agent Planning Suite, we built a custom GPT to answer your questions.

17:50Just go to bit.ly slash super super agent. That's bit.ly slash super super agent, all one word. And if you have any questions, the agent can even help you book an appointment with our team. Welcome back to the AI Daily Brief. Today, we're going kind of deep cut. I just recently found this survey, mostly at a Northeastern university, around AI attitudes across America. And what I think is really valuable about this is that a lot of the surveys or research that we talk about on this show comes from a highly enfranchised context, right? It's all about people who are using AI a lot or who are in positions to use AI a lot versus an incredibly broad cross-cutting survey.

18:34And that's really more what this represents. And I think there's a couple interesting TLDRs on this. I titled the show, Everyone is Using AI and No One Knows What to Think of It. And as you'll see, what that means is that there is really widespread adoption of this technology. There is a very strong sense, broadly speaking, that it is going to be hugely significant, but exactly how and in what ways and what we should do about it, there's way less clarity. It's not particularly political so far, or at least not partisan, but it definitely seems to people to be something that's going to be hugely significant.

19:08So where does this study come from? This came out of the Civic Health and Institutions Project, with researchers from Northeastern, Mass General, Rutgers, Harvard, and the University of Rochester, and is called AI Across America, Attitudes on AI Usage, Job Impact, and Federal Regulation. The survey was conducted between April 10th and June 5th of this year, so not brand new but also not that old, and about 21 ,000 people across all 50 states were asked about AI. So let's talk first about the big takeaways. I think very clearly, Ubiquity is the main one. The researchers write, Artificial intelligence has reached a tipping point in American society.

19:48Half of US adults report using at least one major AI tool. They also note that adoption is widespread. Every state reports usage levels of at least 40%, except West Virginia, where still a third of adults had used AI. I feel like sometimes these numbers seem low because it's hard to imagine that not everyone is using these things, but for a technology to get to this sort of penetration in two and a half years since it came online is just completely without precedent. What's more, it's not just that people are using it, they're convinced of its significance. Again, the researchers write, substantial majorities across all 50 states anticipate AI will impact their job within five years.

20:28There are also geographic patterns in this. When it comes to who anticipates the greatest AI impact in the workplace. It's a combination of knowledge economy hubs, California, New York, Massachusetts, as well as Sunbelt states like Texas, Georgia, and Florida. Companies whose economies are more agricultural and traditional industry have lower expectations of disruption. And when it comes to regulation, on the one hand, stats we've seen previously suggesting that the United States is less optimistic than, and for example some of its Asian or Middle Eastern peers when it comes to AI, are definitely validated.

21:03In every single state, the percentage of people who are concerned about too little regulation outweighs those who are worried about too much regulation. That said, it is also very clear that views on appropriate AI governance are still being formed. More than one-third of people across the whole are uncertain about appropriate regulatory approaches. And while those attitudes may vary geographically along some of those lines that we just heard, they are not strictly partisan based on other political issues. So let's look at a few other highlights before we talk about some of the big takeaways. When it comes to awareness, no big surprise, ChatGPT is the runaway winner, with almost two-thirds of people having at least heard of ChatGPT.

21:44Now that said, Google should be pretty stoked on this because Gemini is all the way up now at 50%. That is a real uptick given that for most of the last couple of years, ChatGPT has been pretty synonymous with AI. Still, one of the most interesting things about this awareness chart is that above Grok, Claude, Perplexity, Midjourney, etc. is DeepSeek. 17 % of respondents had heard of DeepSeek, which is kind of validating of how much people freaked out about it in December and January when it rocketed into the markets and to the very top of the app charts. Usage patterns follow the same distribution, although here at least Grok and Claude are a little bit closer to DeepSeek, suggesting that all the buzz around DeepSeq because of the whole China element outweighed even the usage of it, although of course the fact that it was the first free reasoning model that many people had used, did clearly give it some differentiation.

22:34Let's talk a little bit deeper about these attitudes towards AI regulation, because in many ways it feels like this is the major thrust of this survey. In fact, I could be wrong, and apologies to the research authors if this wasn't their intention, but it almost felt like to me that they wanted people to be more upset about AI regulation than they actually were. Still, overall, across all demographic groups, when asked the question, thinking about the use of AI in the United States, are you more concerned that the government will go too far regulating its use, not go far enough regulating its use, or not sure?