0:10
AI art as curation
When I make an AI image or song, I don’t feel the kind of satisfaction I do when I draw, or play something on the guitar. Angry memes about how “no one making AI art is an artist” just bounce off me, I never felt that way in the first place.
But AI images and music have turned out to be pretty emotionally resonant with me at a level I didn’t expect, for another reason. They let me curate very specific vibes that matter to me and would be hard to assemble anywhere else. Curation isn't creation, but it's a separate and (I think) equally valuable outlet.
0:49
StockQuakes - AI Stock Analyzer - 5000 Stocks Daily
*PAID = Stock Symbol Viewable with Paid Subscription
Get 7 day free trial
Subscribe now
StockQuakes Top 10 Average Index Charts
StockQuakes identify noteworthy changes in the market. StockQuakes sizes can be small, medium, large, and huge. The Daily Top 10 UP Average Index is an aggregate value of noteworthy changes in stocks that ended the day with stock prices higher than they opened per our AI algorithms. The Daily Top 10 UP Average Index is LARGE today with a StockQuakes level of 6.51. Here is a chart of the Daily Top 10 UP Average Index values for the last 10 trading days.
Similarly, the Daily Top 10 DOWN Average Index is an aggregate value of noteworthy changes in stocks that ended the day with stock prices lower than they opened per our AI algorithms. The Daily Top 10 DOWN Average Index is MEDIUM today with a StockQuakes level of -5.21. Here is a chart of the Daily Top 10 DOWN Average Index values for the last 10 trading days.
StockQuakes Daily Detail Charts and Reports
Please subscribe to see the many stocks contributing to the averages shown above including the following:
StockQuakes Daily Top 10 UP Stocks Chart, showing stock symbols and values (example).
StockQuakes Daily Top 10 DOWN Stocks Chart, showing stock symbols and values (example).
StockQuakes Daily Top 100 UP Stocks PDF Table Report, showing stock symbols, names, and values (example).
StockQuakes Daily Top 100 DOWN Stocks PDF Table Report, showing stock symbols, names, and values (example).
Get 7 day free trial
Read more
2:33
THE DEEPER DIVE: Plunging Deeper into Plunging Oil Prices
I want to start off by talking once again about an economist who hallucinates more than AI. Her logic board seems to have fried because every point she makes counters her own argument. I came across an article on Oilprice.com that claimed the oil supply crunch from the Iran war could take crude prices down to $40/bbl! It’s been two decades since we saw prices that low, and we have certainly never seen such low prices due to an oil shortage. We have, however, seen prices that low cause oil producers to reduce supply because it wasn’t worth producing anymore.
You would think, if you were going to make a claim that is peculiar at best, you’d be armed with some careful facts and logic. I read the article to find out how on earth the writer came to the conclusion that a supply crunch would crush prices, and to such a low level. I wondered what on earth I was missing in my own predictions that run the opposite direction. I found out what I was missing was the drugs she’s apparently taking. She starts off reasonably enough:
There have recently been many warnings about near-term oil shortages stemming from the conflict in Iran. Most analysts assume that shortages mean higher prices.
Except that she starts off with that in order to say that this one reasonable statement is unreasonable. Most of us understand this dynamic intuitively. We call it “supply and demand.” Run low on things, people will bid up the price to get the scarce item. Yet, somehow the oil addled brain that wrote that piece thinks a shortage of the most critical commodity for the global economy to function is going to result in prices plunging. Well, in the new world of “alternative facts,” who knows?
Big claims require big proofs, though, so I read on:
As I will explain, the dynamics of a self-organizing economy suggest the opposite outcome — lower prices, deepening recession, and shortages of goods and services that have little to do with price.
There is a formula buried within that statement, which I’ve said we may eventually see, but clearly shortages of goods and services do not lead to an outcome of lower prices UNLESS AND UNTIL they lead to an outcome of recession. Recessions do almost always lower some prices. So, stay in recession long enough and go deep enough, and prices could fall, but that is way down the road.
In forecasting stagflation—the combination of a stagnant or even receding economy with rising prices—I’ve said the present unfolding economic collapse may take us into a depression that ultimately could lower prices by destroying the economy so badly no one can afford to buy anything; but that all starts with inflation getting so bad it crushes the life out of the economy.
Now, the point here isn’t to take down what the writer on Oilprice.com said, but to use it to show in yet another way why things are not going to go that way and how they will go. With that in mind, we get to her breakdown of the dynamics, which disproves itself as it goes:
Rather than high prices, my major concerns are recession and the disappearing availability of goods and services that we rely on. This might be similar to the empty shelves that many stores experienced in 2020 and 2021. There may also be new government restrictions, intended to work around the reduced oil supply in a way that will allow essential services to continue to operate normally. Oil prices are likely to fall below $40 per barrel, as they did in 2020 with Covid restrictions.
The recession that is coming will be partly due to tariffs, which raise prices. That cuts back commerce and can cause a recession; but that recession doesn’t remove the tariffs that get built into prices; so, you likely remain stuck with those. Profit margins have already been cut to extremes to avoid passing on the tariffs as much as possible, so they won’t get cut back more. People may stop buying, then the lack of demand can reduce prices, but that happens after the economic wreckage, and there are other dynamics here that make that unlikely for awhile, too.
The major factor causing a deep recession/depression is going to be the energy crisis. I can’t think of any situation where a shortage of fuel has caused fuel prices to fall. Prices plunged very briefly during Covid, but not due to shortages. It was because no one was driving for a short time. Eventually, high prices during fuel shortage may cause fuel demand to fall; and that can cause prices to stop rising further or maybe to fall eventually if demand drops off enough; but that still mandates that, first of all, you have to endure the shortages that drive up prices enough to change purchasing behavior.
As prices go higher and higher, people start economizing; so, yes, that kind of recession may eventually have a self-correcting mechanism through pricing; but that is after you are deep enough into the pain to force people to change their consumption.
7:52
Intel’s CPO Packaging Roadmap: From Glass Couplers to Pluggable Optical Connectors
Over five days, DAC provides over 300 technical presentations and sessions that are selected by a committee of electronic design and university research experts offering information on recent developments and trends, management practices and new products, methodologies, and technologies in the electronics industry.
The conference is sponsored by the Association for Computing Machinery (ACM) and the Institute of Electrical and Electronics Engineers (IEEE), and is supported by ACM’s Special Interest Group on Design Automation (SIGDA) and IEEE’s Council on Electronic Design Automation (CEDA).
Co-packaged optics, or CPO, is becoming an important technology direction for AI data centers and high-performance computing systems. As the scale-up network bandwidth of GPU clusters and AI accelerators continues to increase, traditional copper interconnects are gradually reaching limits in reach, power consumption, and density. As a result, the industry is pushing optical I/O closer to the chip package.
This article reviews two Intel-related ECTC 2026 papers: “V-groove Based Edge Coupling Enabled by Optical Glass Coupler Attach for Co-packaged Optics” and “Multi-channel and multi-scale Optical Performance for a Detachable Edge-Coupling Connector with a Glass Coupler and Expanded Beam in CPO.”
The first paper focuses on the passive alignment and attachment process between a glass coupler and the PIC V-groove, including material selection, dimensional metrology, and reliability. The second paper further integrates the glass coupler with an expanded-beam detachable connector, establishing multi-scale optical simulation, multi-channel Monte Carlo analysis, and measured insertion-loss testing.
Taken together, the two papers reveal the core challenge of CPO optical coupling packaging: it is not simply about connecting fiber to a silicon photonics chip. The real task is to build an optical-electrical packaging architecture that is manufacturable, testable, repairable, low-loss, and reliable.
Network connectivity in AI data centers can be roughly divided into scale-out and scale-up. Scale-out usually connects different servers or racks, where optical communication is already widely used. Scale-up, by contrast, provides extremely high-bandwidth and low-latency interconnects within a single rack or between adjacent accelerators. Historically, this domain has relied heavily on high-speed SerDes and copper interconnects.
As 200 Gbps SerDes over copper backplane begins to be deployed, electrical interconnects may still extend for some time. However, they will eventually face limitations in reach, power, and density.
The basic idea of CPO is to move the electrical-to-optical, or E/O, conversion module closer to the switch ASIC, GPU, XPU, or accelerator package, and in some cases even integrate it deeper inside the package. This shortens the high-power electrical channel, reduces I/O PHY complexity and power consumption, and increases bandwidth density around the package edge or chip perimeter.
However, CPO also introduces a major packaging challenge: how does light enter and exit the package?
The waveguide mode size on a silicon photonics PIC is very small, while the mode field diameter of an optical fiber is relatively large. If a traditional fiber array unit, or FAU, is directly attached to the PIC V-groove, it may provide low optical loss, but it also sacrifices repairability, assembly flexibility, and yield management.
If CPO is to become a high-volume manufacturing technology, it needs an optical interface that can deliver low-loss coupling while also supporting pluggability, testing, and replacement.
These two papers are essentially answering that question. Paper 19 first establishes the fundamental process for attaching a glass coupler to a PIC. Paper 12 then connects this glass coupler to an expanded-beam connector, forming a more complete detachable edge-coupling CPO architecture.
Review of Paper 19: V-groove and Glass Coupler Attachment Process
Materials and Epoxy Design
Passive Alignment Attachment Process and Dimensional Metrology
Process Optimization: Bonding Force and Epoxy Shrinkage
Review of Paper 12: Detachable Expanded-Beam Connector and Multi-scale Analysis
Why Expanded Beam Matters
Integrated Interpretation: From Process Unit to System Architecture
12:37
Choppy AI Trade, Fed Outlook Drive Asia Markets | Insight with Haslinda Amin 07/03/2026
Asian equities drifted lower on Tuesday, the market’s mood set by two headlines: a wobble in AI‑related trade flows and fresh Fed commentary that kept rate expectations in flux.
In Japan, the Nikkei slipped modestly as tech firms reported mixed earnings, while Chinese manufacturers saw inventory builds that nudged the Shanghai Composite down a touch.
South‑east Asian currencies, led by the baht and ringgit, edged weaker against the dollar, reflecting investors’ cautious stance ahead of the Fed’s next policy meeting.
Overall, the session was a reminder that even without dramatic moves, sentiment can shift quickly when AI hype meets monetary‑policy uncertainty.
13:29
🎥 June Subscriber Q&A is live
G’day Folks,
Thanks to our Premium subscribers for another fantastic batch of questions this month!
In this Q&A session, James and Alec cover everything from bull market targets and cycle theory to Strategy, BIP-110, and where we think this market is headed, including:
Next bull market upside from a $50–60k bottom
Is the four-year cycle still intact?
Does the data support a $40k BTC bottom?
Is Strategy’s capitulation phase almost over?
James’ view on BIP-110
What would make us turn bullish?
Where do we expect Bitcoin to bottom?
Are long-term holders still selling?
Are diminishing returns the new reality?
James’ favourite research sources
We hope you enjoy the session,
James & Alec
14:19
You’re probably ignoring the most important number in your company
Public companies in America are required by law to publish one number: the ratio between what it pays its CEO and what it pays its median worker. The rule has been in force since 2017. The number sits in every proxy statement on the exchange.
Most leaders cannot tell you their own.
When you do look, the numbers are not subtle. The average across the S&P 500 was 285 to 1 in 2024. At the hundred lowest-paying large firms it averaged 632 to 1. At Starbucks it once reached 6,666 to 1. In 1965 the same ratio was about 21 to 1.
The easy way to read these stats is moral. A company that pays its CEO 285 times its lowest worker has decided some people are worth 285 times other people. That frame feels true, and it ends the conversation. It puts horns on one side and a halo on the other, and nobody on the inside recognizes themselves in it—because who wants to identify with horns?—so nothing changes.
Here is the more useful read. The number is the sum of compensation decisions made one at a time by people who were mostly not thinking about the ratio at all. Yet, even if the ratio is the result of several decisions compounding rather than a deliberate choice, it is a powerful indicator of a company’s culture.
So why don’t we change it?
The easy answer is that the people setting the ratio benefit from it. For boards, that is mostly not true. Directors are not protecting their own pay when they approve a CEO package. They are trying to land the person they believe will make the company worth more. And they set that number the way almost every board does. They pull comparison data. They look at what peer companies pay their CEOs. They land near the middle, because nobody wants to explain why they paid below market for the person who runs everything.
The flaw is built in because the benchmark only ever compares CEO pay to CEO pay. No one in that room compares the ratio to other ratios.
It is a habit that shows up everywhere people set pay. We benchmark the top against other tops and we never benchmark the relationship between the top and the bottom. The ratio is the one number nobody’s job is to look at.
But it doesn’t have to be like this.
Mondragon, the cooperative federation in the Basque region of Spain, caps the ratio between its highest and lowest paid to six. It employs tens of thousands of people. It has run for roughly seventy years. It is profitable. It has outlasted recessions that took down conventional firms. The cap is not charity. It is a design choice that someone decided to benchmark and defend.
You can make the same choice with one question.
What ratio between your highest-paid and lowest-paid employee would you defend in writing, with your name on it, to your entire workforce?
If you do not know your current ratio, that is information. If you know it and would not defend it openly, that is more important information.
A ratio you will not say out loud implies you are hiding something.
People are not asking to earn what you earn. They never were. Decades of research on workplace fairness land in the same place. People will live with a gap, even a wide one, if they believe the process behind it was honest and someone can explain it without flinching. What they will not forgive is a number no one will account for. When the reasoning is missing, they do not assume the best. They fill in the blank, and they fill it in against you with distrust, fear, and anxiety.
A ratio you can defend out loud is, by definition, a ratio your people can live with. Those are the same number. Fairness was never sameness. Fairness is a gap somebody is willing to put their name on.
The architecture is the problem. That is hard to accept, because architecture feels permanent and blaming individuals feels more satisfying. But a system that was designed can be redesigned.
18:30
Why saying yes early in your career pays off later
There’s a version of career advice that often gets handed down: Find your passion early, specialize fast, build your personal brand. It’s tidy, and looks good on paper, but it bears almost no resemblance to how most successful careers actually unfold. Mine certainly didn’t go that way. What I’ve come to believe, after 25 years of leading technology companies through growth and transformation, is that the thing most likely to determine your trajectory isn’t your credentials or your clarity of vision. It’s your willingness to say yes, even before you feel ready.
I left school at 16 without a grand plan. I wasn’t particularly academic and, like most 16-year-olds, I didn’t have a carefully mapped career path. So, saying yes to jobs that presented themselves was my first priority, no matter what they entailed. My first job was on Romford Market [in Havering, a London borough], selling bananas from a small barrow next to a fruit and vegetable stall. The days were early, the weather was cold, and the work was repetitive; suffice it to say, I wanted to move on from there as quickly as possible.
Not long after, I said yes to a more exciting opportunity: taxi driver. I passed my driver’s test in the morning and was driving a taxi by the afternoon. In that role, I met every kind of person—happy people, angry people, distracted people, you name it. It was challenging work, but it taught me a skill that’s guided me throughout my life: how to build trust with someone in a very short amount of time.
At 21, I was driving taxis, had a young child, and wasn’t looking for a way out. Then one day, as I was driving a Ford executive to the airport, he asked me what I planned to do with my life. I told him I was planning on driving taxis. He asked whether I’d ever considered a career in computers. I didn’t have the background, the credentials, or the confidence to imagine myself in that world, but he gave me his card and told me to come see him. Eventually, I said yes.
That yes changed everything. It led to work with Ford’s communication network across Europe, starting with auditing screens and expanding into back-end systems, networking, and how information moved across teams and countries. None of it was part of a plan, but each opportunity gave me more context, and over time those experiences built something harder to manufacture than expertise: range.
Range is especially important in a work culture that often pushes people to define their lane too quickly. In my view, the early stages of a career should be less about fine-tuning a narrow focus and more about understanding how businesses really work.
Some of the most valuable opportunities won’t look impressive at first. In fact, they may look inconvenient, messy, or beneath your job description. But I’ve found that the work nobody wants to do is often where the biggest learning lives. And it’s usually where problems are visible, urgency is high, and people notice who steps up.
Case in point: Earlier on in my career, newspapers needed to get from Essex to Paris [more than 300 miles by road], and, one day, the truck meant to deliver them broke down. This easily could have been treated as someone else’s problem. Instead, I offered to get a van and drive them there myself. Was that the most strategic-looking career move? Probably not. But people remembered it because it showed ownership, that when something mattered, I was willing to help solve it.
That kind of moment builds credibility. Not through heroics, but in demonstrating how you operate when the outcome matters more than optics.
There’s another reason to say yes early: You can’t challenge a system that you don’t understand. I’ve seen plenty of smart, ambitious people form strong opinions before they have enough context. They want to disrupt the process, question the decision, or point out what leadership is getting wrong. Sometimes they’re right. But often, they’re missing the history, tradeoffs, constraints, and relationships that explain why things work the way they do.
Perspective comes from pattern recognition, and pattern recognition comes from repetition. The more you expose yourself to different parts of a business, the better you become at seeing what is truly happening. You start to understand the task and, more importantly, the entire system around it.
Of course, there is a limit. Saying yes to everything is not a strategy, and can be a fast path to burnout. Early-career professionals shouldn’t confuse growth with accepting every unreasonable request or tolerating poor treatment.
23:28
Freemium: GPT-5.6 & Sonnet 5: The agentic race heats up
The artificial intelligence landscape in mid-2026 is undergoing a structural transition as development pipelines shift from generic, cloud-hosted endpoints to specialized, multi-tiered infrastructure architectures. This week’s primary advancements illustrate a deep convergence between highly optimized local hardware, custom-designed ASICs, and natively adaptive model reasoning. Developers, founders, and enterprise technical leaders are navigating a rapidly evolving space defined by shifts in API token economics, the emergence of encoder-free on-device multimodal execution, and stricter regulatory constraints worldwide.
In today’s edition, the analysis details the engineering and economic implications of OpenAI’s GPT-5.6 series, Anthropic’s adaptive Claude Sonnet 5, and the Broadcom-partnered Jalapeño custom inference platform. Additionally, this issue evaluates open-source tools and frameworks designed to provide persistent memory, local security scanning, and production-grade observability for autonomous agent workflows.
What Happened: OpenAI has launched a limited preview of its next-generation GPT-5.6 model family, introducing the flagship model Sol, the balanced everyday model Terra, and the fast, cost-effective model Luna. Designed to advance the frontier in software engineering, quantitative scientific analysis, and cybersecurity, GPT-5.6 Sol incorporates deep test-time compute scaling through customizable reasoning effort levels. On Terminal-Bench 2.1, a benchmark assessing command-line planning and tool orchestration, GPT-5.6 Sol Ultra achieved a record 91.9% score, outperforming GPT-5.5 (88.0%) and Claude Mythos 5 (84.3%). To manage security risks associated with biological and cyber capabilities, OpenAI is implementing strict pre-release evaluations and a staged rollout in coordination with the United States government. The preview is currently restricted to select organizational partners using the OpenAI API and Codex, with general availability scheduled in the coming weeks.
The Technical Primer: Conventional large language models rely on next-token prediction, which functions similarly to immediate, intuitive human speech. GPT-5.6’s advanced reasoning represents a deliberate internal monologue, allowing the model to construct, evaluate, and traverse a tree of possible analytical steps in latent space before committing to a final output sequence.
Why It Matters:
Generational Shift in Systems Orchestration: The implementation of “Ultra Mode” enables Sol to dynamically allocate and coordinate subagents, transforming the AI from a single instruction-following model into a self-correcting multi-agent compiler for complex tasks.
Rigorous Evaluation of High-Order Scientific Reasoning: The introduction of the GeneBench-Pro benchmark, consisting of 129 complex computational biology problems, establishes a new evaluation standard that tests an agent’s “research taste” and decision-making under clinical and genomic uncertainty, rather than simple rote recall.
National Security Alignment and Coordinated Deployment: The phased preview model highlights a growing trend of top-tier AI labs coordinating directly with federal bodies to validate cyber-safety boundaries prior to wide public release, signaling a future where frontier models are treated as strategic infrastructure.
Who Should Care:
Developers: Must review prompt caching configurations; GPT-5.6 cache writes are billed at 1.25x the uncached rate, while cache reads continue to receive a 90% discount.
Founders: Can design multi-agent workflows using Terra to achieve competitive GPT-5.5 capabilities while cutting token costs by 50%.
Enterprises: Need to coordinate with account representatives to establish security permissions and evaluate how real-time misuse classifiers impact dual-use defensive operations.
Researchers: Should analyze the design of synthetic, causal benchmarks like GeneBench-Pro to evaluate long-horizon reasoning without data contamination.
Creators: Can prepare for advanced, autonomous content pipelines by exploring the capabilities of the cost-efficient Luna model.
Key Takeaway: GPT-5.6 Sol represents a major leap in long-horizon agentic capability, proving that the future of frontier AI lies in structured, test-time compute scaling combined with rigorous, layered safety guardrails.
What Happened: Anthropic has launched Claude Sonnet 5, introducing a highly agentic model that approaches the capabilities of its flagship Opus 4.8 at Sonnet’s speed and pricing. The model features adaptive thinking natively enabled by default, an expanded 1M token context window, and a 128k maximum output token capability. To optimize performance, Sonnet 5 utilizes an updated tokenizer that processes text differently, resulting in approximately 30% more tokens generated for equivalent text relative to its predecessor, Claude Sonnet 4.6.
28:47
LVMH Wins Court Case Over Logo Used by China Tea Chain
$1.5 billion—that’s the penalty the Chinese milk‑tea chain just got slapped with for using Louis Vuitton’s logo.
A Shanghai court found the brand’s cups and signage infringed the luxury house’s trademark, ordering the company to pay the French group.
The ruling sparked a flood of comments on Weibo, with many users debating whether Chinese firms should be stricter about IP and how the decision might affect other local brands.
For now the tea chain faces a massive bill and a likely rebrand, while LVMH celebrates a win for its global brand protection.
