Andy Dupertuis

Why cloud providers can’t let your cost control be less expensive than long term commitments #finops #cloud linkedin.com/feed/update/ur…

@RealJimChanos No disagreement

Meanwhile, as everyone rushed to build new data centers, the old data centers continue to deteriorate. In the LTM $EQIX’s pre-tax ROIC was only 5.8%.

@QuinnyPig Not concrete. Cold shell cost is comparatively trivial

Huh. Amazon's free cash flow has collapsed by 95%, because basically every dollar in the door goes back out as concrete, GPUs, and substations. CapEX went up $59.3B. Net income was $30.3B. $16.8B of that is a pre-tax paper gain from marking up their Anthropic stake

@felixleezd *ahem youtu.be/EUXnJraKM3k?si…

If you open a Chinese app for the first time, you’ll probably think it’s badly designed. Too many icons and features. Everything crammed onto one screen. If you grew up on Western apps, your instinct is immediate: this is cluttered. But it works. In the U.S., we’ve been trained to associate good UX with minimalism. In China, density often signals value. Open WeChat or Alipay, and it feels overwhelming at first. Information-heavy, feature-packed. But to local users, that density means capability. It says: everything you might need is already here in front of you. If you enter a new market assuming your design taste equals good UX, you’ll misread the signal. Good design is contextual.

@skdh *ahem

The problem with time travel has nothing to do with Einstein's theories in particular, it's simply a consistency problem. One has the same issue in quantum field theory, which is why we need to introduce a time-order relation to make sense of it. Incidentally, if that wasn't so (that time travel has a conceptual problem rather than a problem with general relativity) one could argue that quantum gravity might make time travel possible. Time travel really only makes sense conceptually if you travel back in time into some sort of parallel universe and that isn't compatible with the current theories of physics. Warp drives btw are compatible with Einstein's theories, that is the entire point. The problem with warp drives is that to go faster than the speed of light you need some sort of negative energy which doesn't seem to exist. x.com/JonesDanny/sta…

@BLACKWELL154380 @JasonBotterill Actual answer- ai compute capacity is ephemeral. There is no concept of storing it for distribution later. Volatility has to be managed in realtime. Demand is spikey and controllable. Supply is steady and fixed. So your lever is demand mgt

@JasonBotterill Not being able to use /fast mode still feel crazy to me. Why do they care how I want to spend my usage

The ChatGPT Pro plan gives way more value than Claude Max now. GPT-5.4 thinking is high taste and unlimited, much more Codex limits, and you actually get /fast mode as part of the subscription. Anthropic doesn’t let you use /fast mode even on the $200 tier.

@akshen121 @TheRealAdamG @TheRealAdamG does the excel plugin benefit from the Pro subscription tier? Or are pro/plus the same? @OpenAI

Chatgpt Vs Claude in Excel TLDR: Gave ChatGPT and Claude the same credit risk modeling task in Excel. ChatGPT followed proper methodology, did variable clustering, train/test split, Excel native model, AUC 0.627 on test data. Claude skipped clustering, inflated IV with sparse categories, ran the model in JavaScript not Excel, reported 0.706 AUC on training data with no split. Chagpt 5.4 nailed a real world Banking and Finance modelling case study. I gave both ChatGPT and Claude the same banking credit risk dataset with the same prompt: bin the variables, calculate Information Value, do variable clustering, select features, build a model in Excel, generate ROC. Basically an end to end scorecard development workflow inside Excel. ChatGPT followed the brief properly. Quantile-based WoE binning, IV ranking across all 33 variables, correlation-based variable clustering at 0.75 threshold to remove redundant features, picked 6 representative variables, built a 2-variable Excel-native decision tree with proper train/test split via ID mod 10, scored leaf-level bad rates, ROC on held-out test data. AUC: 0.627. Claude looked more impressive on the surface. 28 sheets, individual binning for every variable, logistic regression with 27 features, gradient descent at 500 epochs, full coefficient table with importance bars, confusion matrix, precision/recall/F1. AUC: 0.7066. Sounds like it won right? So, The logistic regression was computed in JavaScript via execute_office_js and results were pasted as static values. That's not Excel-native, against the task. The 0.706 AUC? On the full training set, no train/test split. That number is meaningless. Variable clustering? Skipped entirely even though the prompt explicitly asked for it. It "selected" 31 out of 33 features which is barely feature selection. The entire feature ranking was built on a methodological error. I've done this exact dataset myself with deeper feature engineering and ensemble technique hit 0.70 AUC. The fact that ChatGPT got to 0.627 with less feature enigineering and naive modelling is genuinely impressive. Binning and IV based feature selection is one of the most important techniques in credit risk modeling and it nailed the workflow. One thing claude did well is the workbook is visually impressive. One area ChatGPT can improve, the visuals and formatting could be sharper, the workbook is functional but not polished.

@sama en.wikipedia.org/wiki/The_Last_…

"What is the hardest question I could ask you that you might get right?"

@emollick Difference: copilot has Byzantine licensing requirements. You can’t just try it

My Excel toolbar right now. They are all different from each other in ways that are only clear when you use them a lot, and which also differ from the results if you ask Claude or ChatGPT on their websites to create an Excel sheet, or if you use Cowork or Codex. Complicated!

@koltregaskes -> youtu.be/C0W4H6jMLKg?si…

AI productivity psychosis is becoming a real issue. We are running so many autonomous agents in parallel that the cognitive load of just monitoring them is breaking people. You can delegate all day, but the fatigue of constantly directing tasks and neglecting personal downtime is catching up with the workforce. I'm running 10+ sessions simultaneously right now and I'm getting completely confused between what one is doing compared to the other. The mental overhead of tracking which agent is working on which task, what state each one is in, what needs reviewing - it's exhausting. We've automated the work but created a new bottleneck: ourselves. We're the ones watching, redirecting, approving, context-switching between parallel streams. That's not productivity. That's just distributed cognitive load. The tools let us spin up endless agents. But nobody's solved the human side - how do we actually manage this without burning out? We need better orchestration layers, not just more delegation.

@mattturck Make the printer work right before an important meeting

The real AGI will be when laptops reliably connect to the big screen in the conference room without having to call IT every time

@Geiger_Capital @Citrini7 Wut? >”The inverse of the convex”

My favorite part of the @Citrini7 piece India is going to be absolutely decimated due to their entire economy being reliant on providing cheap white-collar workers to the West. Probably spot on, actually.

@buccocapital *yes *yes *sorta *deterministic

Do you realize just how much fucking money OpenAI and Anthropic have raised? Do you realize how much is at stake? They cannot become commoditized intelligence accessed via API They mathematically and strategically must attack the application layer

@NWischoff I gave a talk about this at a major academic institution years ago

Had to sit down for this one

@zachlloydtweets x.com/andydupertuis/…

Switched to Claude Code for a week. I get the appeal... but man, I couldn't stop missing a proper GUI. Simple things like moving your cursor by clicking, multi-cursor, dropdowns, and scroll views are all just nicer to use. We deserve hi-fi coding agents

@dittycheria Actual answer. ‘msft’ is an equity, but Mcrosoft is a loose amalgamation of large, mid, and small size companies. Each has its own culture and mechanism to make stuff, sell stuff, and count the money. Coordination between them occurs in a partner metaphor

I’m genuinely perplexed by MSFT. I needed Teams since some orgs use it instead of Zoom or Meet. Went to the Microsoft site, downloaded it, installed it, then got told Teams Classic is no longer supported. Why allow users to download deprecated software at all? How is this acceptable UX?

@simphiweyinkoc_ “I think we’re saying the same thing” or “I think we’re aligning”

How do you say “did I stutter?” in corporate language

@BradRTorgersen Put a capable, pressured captain on an isolated ship, give them a mission with political consequences, then force a decision where competence and ethics collide

Been meaning to finish what I started months ago: a complete re-watch of TOS from the inception. Given the present fracas over SFA, tonight seemed like the time. It's been 61 years since network execs had a first look at Roddenberry's so-called Wagon Train to the Stars. "The Cage" would say a lot about what was to come. For starters, there are no wink-and-nod cutesy references to (then) 1965 pop culture. I am sure it never even crossed Roddenberry's mind. This was going to be a serious show about serious ideas. And the ship is crewed by serious people. Pike and Co. are serious men and women on a mission. The best scene (for my money) is Pike's lamentation to the surgeon: "You bet I'm tired! I'm tired of deciding who lives . . . and who dies." It establishes the stakes. The Enterprise is not a safe space. Its personnel sacrifice their lives in the line of duty. Nothing adolescent about that. As foes go, the Talosians aren't the best baddies Trek's ever seen. But the predicament created is still an interesting one. What's better: to prefer hard reality? Or let oneself drift into pleasant illusion? The episode doesn't make a final determination. It basically says, "Depends." And leaves it at that. (A conundrum that would be addressed thirty five years later with the first Matrix movie.) Most important of all, "The Cage" makes it crystal clear this is a space navy. For all the nonsense later writers tried to insert about how Starfleet is not a military force, the original pilot features military rank and discipline aboard a military vessel armed with military weaponry. Including personal phasers, crew-served phaser cannon, etc. The space navy setting is *essential* to getting Star Trek right. Any episode attempting to ignore or strip this away, does a disservice to the franchise. And the franchise works best when its space navy aspect is front and center—as with later TOS episodes like "The Doomsday Machine" or movies like The Wrath of Khan.

@corbtt Hey neighbor

Claude Code just processed a return for the shoes I ordered on Amazon and ordered the next size up for me. Completely autonomously, from a 2-sentence task description. HAVE YOU UPDATED YOUR WORLDVIEW THAT BROWSER AGENTS ACTUALLY WORK YET?

@trengriffin (48.6178883, -123.1701583). Careful, shallow in places

I didn't see any of you walking along this beach in waders holding a rake and a flashlight catching Dungeness crab at low tide just now

@sethbannon If you bring back blue books and oral exams it reduces revenue and margin upside because total cost to serve goes up, margin goes down, and pipeline velocity slows. Probably not worth it unless you’re arguing that schools have a higher purpose?

This is brilliant. A professor noticed take home assignments coming back suspiciously good. Like McKinsey memos. So he started cold calling the students asking why they made certain choices in their submissions. They couldn't explain even basic choices! Clear copy/past from LLMs. So he fought AI with AI -- an oral final exam run by a voice agent and evaluated by a council of LLM graders. > 36 students examined in 9 days > ~25 min avg per exam > $15 total all‑in (≈ $0.42/student) > Full transcripts, audit trail, and super actionable feedback This works because you can paste into ChatGPT and copy the output, but you can’t fake coherent, real‑time reasoning about your project when someone keeps drilling. Interesting that the LLM grading committee actually converged after deliberation and exposed a teaching gap (A/B testing was the weak spot across the class). Students using AI killed take home exams. Very clever to fight fire with fire and use AI to bring back oral exams. Perhaps not surprising, only 13% of students preferred the AI oral format 😂 Oral exams used to be the gold standard in education but were replaced by more scalable written exams. With AI, oral exams are scalable again. Will be interesting to see how this changes education.