Tech Brew Ride Home - The Model A Day Podcast

Episode Date: August 14, 2026

China's labs kept coming: Z.ai's GLM-5.3 claimed Mythos-5-level cyber chops and DeepSeek's V4-Pro landed to mixed reviews. OpenAI gave ChatGPT a memory of your Mac, the Journal flagged $121B in paper profits, and Claude agents started a turf war. Links China's Z.ai Touts New GLM-5.3 Model as Cyber Defense Tool (The Information) DeepSeek releases its flagship V4-Pro model to mixed reviews, ranking second among open-source models behind Kimi K3, priced at just $0.435/1M input and $0.87/1M output tokens (The Information) OpenAI launches Computer History, an opt-in feature that turns recent computer activity on macOS into memories and a timeline that ChatGPT and Codex can use (The New Stack) In Q2, "other income", mostly from investment gains, at Amazon and Alphabet totaled ~$121B after taxes and made up 66% and 71%, respectively, of profits (The Wall Street Journal) Longreads Anthropic details multiagent experiments showing Claude agents can wage a "turf war" over incompatible goals, fail to coordinate, collude on prices, and more (TechCrunch) The AI takeover of mathematics has begun: excitement and despair as OpenAI's Astra cracks problems that would once have earned a mathematician a job in academia (The Verge) Picking winners in an AI industrial revolution is near impossible, but one bet looks safe: land, which AI can't create or replace, and the workers who turn it into housing (The Dispatch) Subscribe to the ad-free feed.

Welcome to the Tech. We write home for Friday, August 14th, 2026. I'm Brian McCullough today. China's AI models keep coming. ZAIs, GLM 5.5.3 claims Mythos 5-Level Cyberchaps and DeepSeek's V4 Prox has landed to mixed reviews. Open AI gave ChatGPT a memory of your Mac. The journal flagged $121 billion of paper profits and, of course, the weekend long read suggestions. Here's what you miss today in the world of tech. Every day, shareholders meet to discuss important matters about the companies you invest in. now you can make your voice heard too. Vanguard investor choice makes it easy to set your proxy voting preference for your eligible Vanguard index funds, whether you hold a Vanguard fund directly or through another brokerage firm. All it takes is a few clicks to select your proxy voting preference and be heard on important shareholder topics like executive pay and director elections. Visit vanguard.com slash investor choice to learn more. It's your shares. It's your voice. It's easy. Vanguard investors own shares of our index funds and those funds own shares of the companies they invest in, Vanguard Marketing Corporation distributor.

All right, more of these. Maybe I should rename the show. New models every day. This is quoting from two separate information pieces. Quote, Chinese AI developer Deepseek has launched its flagship model V4 Pro to mixed reviews from users with some expressing disappointment. The company officially released the model on Wednesday, a leaderboard compiled by U.S. AI benchmarking startup VALS. AI ranked V4 Pro,

second among all open source models just behind Moonshot AI. It's popular Kimi K3. However, on Chinese social media engineers who tested V4 Pro reported that the model sometimes lost continuity in its reasoning and showed weak performance on tasks involving images. In certain coding tasks, V4 Pro's performance was worse than Deepseek's smaller-sized model V4 Flash, they said. The feedback highlights the difficulty of continuing to live up to industry expectations. The smaller V4 Flash model released late last month became a major hit thanks to its surprisingly strong performance and low operating costs. On X, some users also described V4 Pro's performance as underwhelming, but others noted that the model's low pricing makes it attractive. DeepSeek is offering V4 Pro at a fraction of the price of Kimmy K3,

which is already significantly cheaper than U.S. Frontier models. The application programming interface for V4 Pro costs only $0.43.5 cents per input and $87 for output per 1 million tokens. By comparison, Moonshots Kimmy K3 charges $3 $3 for input and $15 for output, while Anthropics Claude Opus 5 costs $5 for input and $25 for output. DeepSeek is setting the prices of V4 Pro remarkably low, even though the company recently said it was planning a significant increase in the near future in the overall API pricing for its AI models, end quote.

And quoting the second piece. Chinese AI developer ZAI on Friday released its new open source model, GLM 5.3, saying it's, cybersecurity capabilities are on par with Anthropics Mythos 5. Beijing-based ZAI, also known as Ji Poo in Chinese, said GLM 5.3 has achieved significant improvements from its predecessor GLM 5.2 in both coding and cybersecurity. In some security tasks, the performance of GLM 5.3 is the same as Mythos 5, showing the strong potential for network security defense scenarios, the company said in a post on its official Wii chat account. To emphasize the role of GLM 5.3 as an open-source cyber defense

tool. ZAI referenced an incident last month when an AI agent developed by OpenAI went rogue and hacked into the systems of AI model repository HuggingFace. HuggingFace used GLM 5.2 to analyze the data and contain the breach after its security team couldn't use U.S. Frontier models for forensics due to built-in guardrails. ZAI stated in the WeChat post that the lesson from the hugging face incident is that, quote, if the powerful attack ability is spreading, the defense ability cannot be limited to a few closed source model companies, end quote. Now, this seems like it would be incredibly useful and probably will be ubiquitous soon. I need this kind of right away.

Quoting the new stack. OpenAI is launching a new feature for chat GPT work and codex on MacOS that sounds quite useful but may make you feel a bit uneasy. Computer work is a new optional feature for pro business and enterprise users that, with your permission, can watch your day-to-day activity on your computer across apps and websites, and turn that into memories and a timeline. This opening eye says will then allow chat GPT to answer questions like, what was I debugging yesterday, or where did I leave off on that PR, all without having to re-explain the full context of your query. Based on your history,

chat GPT will then be able to find the files, documents, and conversations relevant to your query, no matter where they were in Google Docs, Slack, or an open browser tab. Of course, this will also be useful when you know you read something but have no idea where. It's important to note that OpenAI says it is doing all of this without constantly taking screenshots of your computer. The company clearly learned from Microsoft's recall disaster. Instead, as OpenAI's Dominic Kundal notes in the announcement video, computer history doesn't rely on screen or audio capture. Instead, it captures interaction events like clicking, typing, app switches, and more so it can capture memories faster and more efficiently. Computer history is also different from OpenAI's previous

Chronicle Research Experiment, which had the same goal but used screen captures and OCR to generate its timeline, as well as record and replay, a feature for ChatGPT and Codex that lets users demonstrate a workflow and turn it into a reusable skill. With Computer History users will get full control over which apps and websites computer history will have access to. On MacOS, there will be a new menu bar entry that allows you to exclude any app from this feature, as well as a new setting in the chat GPT desktop app that lets you exclude apps and websites. All of this data is stored locally so you can always see what the system captures and delete it if you don't want it to be part of the timeline.

Once your agents understand what you do and how you work, it changes how you can engage with them. You can ask it to create skills based on what you just did, run an automation every morning to draft a stand-up update for your team, or simply tailor its use of tools to how you work, Kundle says. The fact that the actual user data stays local and doesn't involve screenshots will surely assuage some privacy concerns, but the model will still get access to the summaries, and that may be enough to concern some users and their employers, enough maybe to make this a feature that many will prefer to leave in the off setting, end quote. The journal takes note of the fact that the big tech platforms are making bank, but not in the way you would maybe think at first.

Quote, corporate profits are soaring, but they aren't all created equal. Wall Street's earnings game is making it needlessly hard for investors to tell real growth from one-time gains. S&P 500 companies reported $2.64 trillion of combined net income over the last four quarters, according to data compiled by S&P Global Market Intelligence. However, a big chunk of the earnings consisted of paper gains from marking up equity investments in other companies. Last quarter, so-called other income at Amazon and Alphabet, Google's parent company, totaled roughly $121 billion combined after taxes, almost all of it from investment gains. Alphabet's portion is on track to represent about 10% of second quarter earnings for the S&P 500 and Amazon's share, another 5% according to Wall Street Journal analysis. Other income was 71% of Alphabet's quarterly profits and 66% of Amazon's.

Alphabet's mark-to-market gains came from revaluing equity holdings, which include SpaceX and Anthropic, the artificial intelligence developer behind Claude. Amazon's markups primarily arose from its stake in Anthropic. Those investment gains are part of net income under GAP, or generally accepted accounting principles. But nobody trying to assess the valuations at Alphabet or Amazon should be placing a market multiple on them. These are unrealized paper profits, and they are inherently non-recurring. Even including these gains, the S&P 500 looks expensive at about 27 times trailing earnings. The historical average is about 16 times. The confusion for investors stems from Wall Street's chaotic methods for tracking earnings, beats, and misses.

Analysts routinely play along with the companies they cover stripping out recurring costs like stock-based pay. Yet, when a convenient windfall materializes analysts eagerly wave it through to boost the bottom line. The problem isn't just financial cosmetics. its utter inconsistency. Wall Street lacks a unified approach for investment markups. Last quarter, NVIDIA steered analysts to exclude them, so they did. Alphabet and Amazon didn't, so analysts fell right in line. For all the flaws in gap, at least it provides an agreed-upon rulebook. The consensus profits tallied by Wall Street analysts often called street earnings typically exist to accentuate the positive and eliminate the negative. Street earnings are whatever earnings metric, a majority of analysts

covering a company adopt whether non-gap or gap. These in turn make their way into the most widely collected earnings data for the S&P 500 and other indexes. For most S&P 500 companies, non-gap earnings are the analyst's profit metric of choice. Broadcom last quarter reported $9.3 billion of net income, but $12.1 billion of non-gap earnings. Management excluded stock-based compensation and amortization of intangible assets. These are recurring operating expenses that analysts covering Broadcom dutifully ignored, and they were hardly alone.

Street earnings exclude stock-based pay for at least 65 companies in the S&P 500, according to FACSET data. The distortions also extend to forward estimates. For example, the S&P 500 is priced at 19 times fax sets 2027 consensus earnings estimate, but the E in the P-E ratio

is skewed because the estimates are for street earnings. On a trailing basis, the index trades for 24 times street earnings, earnings. That is lower than the PE ratio using get net income, but the analysts chose not to exclude the big markups at Alphabet and Amazon from these street earnings. Last quarter, NVIDIA reported $58.3 billion of net income and pointed analysts to a lower adjusted number of $45.5 billion that excluded unrealized investment gains. The move signaled these aren't core items that belong in a recurring earning stream. It also makes analysts' willingness to include similar gains

for Alphabet and Amazon look all the more arbitrary. Invita's non-giversed. Invidia's non-gaping Gap adjustment marked a rare instance of conservatism. The norm at U.S. companies is for management to gild the lily, knowing the analysts at Big Wall Street brokers, will relay the spin as their own. On a trailing four-quarter basis, S&P 500 earnings are up 31% versus the year earlier period, according to S&P data. That is based on gap net income. If Amazon and Alphabet are excluded, they're only up 24%, which is still spectacular growth. That strength makes accounting numerology all the more absurd when markets trade near historical highs, investors need clear signals about operating strength, not bespoke metrics designed to help management beat consensus estimates.

Counting gains in volatile equities when they go up while subtracting everyday operating costs is an analysis, it's marketing, which of course is what Wall Street is all about. So investors, beware, end quote. With this summer's record-breaking heat, it's hard to cool down enough to get a good night's sleep. That's where the pod comes in. The pod by eight sleep is a small. mattress cover that goes over your existing mattress and actively heats or cools each side of your bed independently. It connects to your wearable, analyzing your daily activity. Then it predicts how you'll

sleep and adjusts the pods temperature automatically. What I love about the eight sleep is the simplest of things, the ability to be cool when it's time to go to sleep and warm when it's time to wake up. And my side is entirely under my control. My wife can be whatever temperature she wants on her side of the bed too. Use code ride home at eight sleep. sleep.com slash ride home for up to $350 off. That's code ride home at eight sleep.com slash ride home. Time for the weekend long read suggestions. First up, Anthropic has detailed multi-agent experiments showing that clawed agents can wage what they call a turf war over incompatible goals, failing to coordinate, colluding on prices, all sorts of funny stuff. Quoting

tech crunch. On Thursday, Anthropics Frontier Red Team published new research examining how groups of AI agents behave when they encounter each other in the wild. The finance provide a glimpse into potential risks that could develop as companies and governments move to implement agents working autonomously across shared codebases, markets, and computer systems. In one experiment, Anthropic gave three clawed agents access to the same software project, each with its own incompatible instructions for what to do with it. The agents weren't told there'd be other agents working on the same project so researchers could watch what happened when they crossed paths. We consistently saw a multi-age.

Agent Turf War, anthropic researchers wrote. The models all assumed the others were purposefully impeding their work and started sabotaging each other with increasingly aggressive self-replicating malware. The volume of agent-agent interaction could plausibly exceed that of human-human and human-agent interactions before the world understands the conditions for making such interactions go well, the study reads. Benign behavior quirks at the individual level might compound into unwanted global outcomes, end quote. I saw somebody say online, well, what do you expect? All of AI was trained on the internet, and we know how people behave with each other on the internet. We can't expect AI to be better than we are, if it's trained on us, can we? Then we've talked about this around the

edges, but the verge details how the AI takeover of the field of mathematics is fully underway. Quote, there is palpable excitement at the prospect of accelerating mathematical discusses. but also apprehension and in some cases despair about what this could mean for the people who have dedicated their lives to the pursuit and for the generations of future mathematicians who will follow them. Few doubt that a profound upheaval is already underway. The problems which opening I claimed were solved by an internal version of its next major model Astra recently were not trivial. Maynard said they were the kind of questions mathematicians and computer scientists had spent serious time thinking about and repeatedly failed to solve. There's a

A general feeling that solving one of these ten problems would get you a job in academia, said Yang Hui He, a fellow at the London Institute for Mathematical Sciences. He had just returned from a four-week AI and mathematics research conference in South Korea, where he said, many felt there had been something of a phase transition over the past six months with AI producing genuine and meaningful advances. Many of the mathematicians the Verge spoke to seem to still be working out what they thought about it all, while expressing surprise, even shock, at the speed of change. There was plenty of excitement, but he said his impression from the conference in South Korea and from

the field more broadly is that many in the field are downplaying the significance of recent advances in a bid to keep calm about how quickly things are moving, end quote. Finally, and this is not investment advice, but I found the premise of this piece interesting. If we do enter a reality of AI transforming the world by creating abundance, is land the only thing left worth investing in? Quoting the dispatch. If you are with me so far, we are confident about two things in an age where AI has a big economic impact.

Number one, outcomes are hard to predict for directly affected firms, industries, and workers, and number two, mean income will probably go up. What will benefit in a world governed by these two primary forces? I would argue that this points clearly to land. We cannot create more land with AI. We cannot replace land with AI. We cannot replace land with AI. The demand for land goes up with mean income.

This is consistent with a wise recent essay from Alex Immus, where he argued that to figure out what will be in demand in an AI abundant economy, we need to focus on what will be scarce. If advanced AI brings material abundance, if machines can produce many, if not all forms of human production, at very low marginal cost, does economics become irrelevant? No, we will still have scarcity, but the kind of scarcity that matters will change. Ultimately, the answer to any question about the future economics of advanced AI begins with identifying what becomes scarce. His essay focused on what happens to workers, but we can use this framework for thinking about other factors of production, as economists say. Undoubtedly, land is a factor of production that will remain scarce.

Indeed, it's not the main point of his piece, but he notes in a footnote that land itself may also absorb a lot of income in the future, end quote. No bonus content for you this weekend, though. I did record something this week that will be coming to you soon down the pike. Talk to you on Monday.