Ben

@david_erald @hagaetc @ejames_c mildly surprised no one caught it earlier. wondering if the numbers are inaccurate but close enough?

@hagaetc @ejames_c @benclmn -- Another episode of "Cedric was right"

Software can be used and tested. AI can write some code, you ship it and get feedback from users. You can quite quickly and straight forwardly figure if it was "correct" or not. For data on the other hand there is no similar feedback mechanism. How can a management team know that "Some data output produced" is correct? ... they can't! Unless someone has done the work to craft and understand everything underneath. Sure, anyone can now write some SQL with AI, but you need to know the context to know if you actually answered the question at hand with all the nuances and context that entails. h/t @bennstancil for this profound insight

@ejames_c 🤔

You know what’s the wildest idea from Darwin’s People? It’s this: You don’t have to talk about sample sizes or statistical power if you’re trying to prove causation. You only have to talk about those things when you’re trying to prove correlation.

Is it true that grace and high expectations are mutually exclusive? Not really. Still, that is what some people claim. Let me illustrate what I think their thinking is wrong. 1. Grace is about how you respond to mistakes, while High expectations are about what you believe a student can achieve. You can respond to a student’s struggle with patience, empathy, and support while still expecting them to meet academic and behavioral standards. 2. Grace removes barriers; expectations set direction. --Grace says, “I see why this was hard. Let’s adjust, support, or reteach.” --High expectations say: “And I still believe you can get there.” 3. Grace strengthens the relationship needed for high expectations to work. Kids don’t rise to expectations for adults who dismiss, shame, or punish them into compliance. But kids rise for people who show: “I’m on your side, and I won’t lower the bar for you, but I also won’t let you fall alone.” 4. Grace is not lowering standards. Grace is lowering barriers, fear, and shame. 5. Real-world learning uses both. Every adult who becomes skilled at anything has benefited from second chances, patient feedback, space to practice, and someone who believed they could succeed, so do students. So as a teacher, you can and should be able to have high expectations while knowing how to extend grace. To thrive in schools, kids need both.

My personal answer is "experiential conviction." When it comes to know-how that requires people to do things they're uncomfortable with or go against the status quo, the bottleneck is not anymore *knowing* what would work, but *believing* it is worth doing, all things considered. A practical example: both I and an AI may know that you should probably do X. But only I can look into your eyes, show you my scars, tell you how I was like you, thinking like you, fearing like you, believing like you that while X made sense it wasn't worth doing, and then I did it, and now I'm fully convinced it's worth it, and you can look into my eyes and see I truly believe it. The person who has tried something in the past and has seen it working, and has taught it to others and has seen it working for them, too, has a thinking, communication, and execution hedge compared to the one who suggests doing the same thing but hasn't experienced it. That's why, for instance, a good managerial consultant can basically make their know-how public and not fear much competition for the in-person part of their job, because the experiential conviction is the hedge. Others who merely read the know-how won't have the experience and conviction to repeat it to others in a way that gets them to use it with the necessary conviction to use it well.

I’m increasingly of the conviction that the most underrated attributes of any knowledge professional are: energy, networking, and learning how to learn.

An essential belief of mine is that one should treat pleasure as something of a skill — it can be learned, cultivated, appreciated. We can study our senses and more for the sake of pleasure. A few notes towards that end.

@arvidkahl vector + axiom

Say I have a single (rotating) logfile on a VPS that I want to archive and make searchable. What’s the cheapest way to do this without having to deploy a complicated stack on another VPS? 🙃

So about all this: fly.io/blog/kurt-got-…

@CJ0pp3l 👋

You never know where you end and your environment begins. You have to eat. You are inseparable from it. If your skin didn't hold it back, if you suffered the slightest pin prick, your body practically wants to merge with the rest of the world.

@david_erald @TheOneandOmsy @ejames_c calibration case method works

@TheOneandOmsy @ejames_c @benclmn You folks probably like this.

The Masa Son / Milner / Facebook / ByteDance lore: TLDR: - In '09, Zuck raising a round and Yuri Milner tells him to name a price - Zuck says $10B. Milner puts in $200M - Then buys another $100M off employees at $6.5b - Masa Son passes, thinks it's too expensive - Fast forward 3 years later - Facebook IPOs at $104b - Fast forward today - Facebook worth $1.7T (170x that $10b round) - Masa still views it as one his biggest misses - When Milner was asked how he got there, said he'd operated a social network in Russia and already studied adoption + monetization, giving him a pricing advantage over Masa and all the other investors - And for Masa, after missing Facebook, when he saw ByteDance, he didn't flinch. Recognized history from Facebook, and wrote as much as he could at $75B. And today, it's one of his better investments in size My takeaway - sometimes you just need to know what greatness should look like: - Milner saw it in Russia, so Facebook was obvious - Masa saw it in Facebook, so ByteDance was obvious - But if you've never seen it, you'll never know what's obvious H/t - Alok Sama and "The Money Trap" - good read on Masa and SoftBank

This is the type of person you want to fill your punchcard with (and the type of person you want to try to be like):

@crystalwidjaja which app is this

im a sucker for data presentations that alter held beliefs （ ・∀・） flights during gojek years were more than i thought

I wasn't joking on the Standup podcast when I said this. I have never bothered to look up what <3 was supposed to be, so I always read it as "testicle party" because that's what it looks like: a pair of testicles with a party hat on.

"It’s the integrity of their frame that I care about: someone who lives by their beliefs often emanates a peace and self assuredness that makes leaning in feel intuitively safe." open.substack.com/pub/skincontac…

@david_erald anthropic.com/news/tool-use-…

@benclmn Ohh wait, tell me more mas? By "custom function call" it means in the Claude/ChatGPT itself, right?

Can't believe I'm saying this but LLM combined with XmR chart might be a great way to introduce SPC to our company. It's have a weird space where it's still mysterious enough to be follow by the team blindly ("Our AI Lord say there's non-routine/anomaly variation in this period. Go look for it!") but if needed you can understand the underlying method ("Nah, it's legit non-routine one. I've actually check the data") So bullish on this.

@david_erald sounds like providing a custom function call is what u want mas

Write this one publicly so I can update it when the experiment is done > [More rule to handle] -- Need to know if, instead of simple "If the number cross UCL or LCL, then scream 'Non-Routine! Go check', it can handle Rule 2 and Rule 3 like what Cedric and Sam mentioned here: #rule-2-when-three-out-of-four-consecutive-points-are-nearer-to-a-limit-than-the-centre-line" target="_blank" rel="nofollow noopener">commoncog.com/becoming-data-… > [Can process raw data itself] -- Need to know if, LLM like Claude or ChatGPT can processed the raw data (.csv or .xlsx file) itself and don't need for me too screenshot the number everytime. Try to hit the context window limit. > [Know why start there?] -- Need to have strong grasp to understand why we start measuring in certain period ("Why you start giving data from March, not since January?"). This seems critical but I don't know to solve it. Probably give as much as data that you can? > [Collect the actionable lever] -- Need to have good mechanism to collect signal and the root cause ("We saw that this error code, 57, is spiking. After digging, it turns out because of X and Y") and categorized the root cause to "This is internal factor - we can solve" and "Hmm, outside of our control".

@Malcolm_Ocean i always recommend my friends who are trying to "develop taste" play a rapid game of <this is good / this is bad> basically, you just scroll through a diverse set of pictures, like the ones on midjourney dot com and say "good - bad - good - good - bad" instinctually.

@simonsarris which camera are you on?

These are 43mm. The field of view feels more natural. I'm exchanging vistas for detail and (ever more) constraint. I don't actually have a lot of shots with it yet, but I quite like it so far.

(going from a camera with only a 28mm lens to 43mm) I think there's a closeness that I'm looking for. With the 28 I found myself almost always cropping. It's certainly more versatile, but I want the feeling of closeness born out of restraint.

All the VC/founders musing about doing rollups and injecting AI into inefficient service jobs are going to learn this the hard way. I hope they enjoy spending 5 years learning about mortgage servicing compliance regulation or whatever!