iiinova — where your digital self keeps evolving

Matan Hakim @matanhakim.bsky.social · 31/08/2026

Eric Nantz is in DS Lab?! Clear your desk and get ready for a good one!

130

Matan Hakim @matanhakim.bsky.social · 19/08/2026

And of course, I hope you'll feel better and you're family will have all the best going forward 🙏

000

Matan Hakim @matanhakim.bsky.social · 19/08/2026

Personally, since I do data, research and tech work in a non-tech environment (NGOs) this has been the best period ever. My outputs are through the roof, and my personal work time is more manageable. Stating this not to rub it in, rather to say that outside Big Corps there are other storylines.

100

Matan Hakim @matanhakim.bsky.social · 10/08/2026

🇮🇹 ➕ 🇮🇳 🟰 🇺🇸 ⁉️ ❌❗ 🇶🇦‼️

010

Matan Hakim @matanhakim.bsky.social · 08/08/2026

Thanks Eric and Mike @mike-thomas.bsky.social for your kind words, and for taking my thoughts as a place for discussion rather than as criticism🙏 I think the more we enter this era, the more we need to use meta-cognition like in my post, as this is our advantage over current LLMs.

021

Matan Hakim @matanhakim.bsky.social · 07/08/2026

I hope there's also something about camelCase and snake_case

101

Matan Hakim @matanhakim.bsky.social · 04/08/2026

I feel very much the same, only it really also includes the parts of: coordinating the escape room time for everyone, handling that one friend who is a bit awkward and still creating the situation for everyone to enjoy, and recognizing the effort of every friend afterwards.

110

Matan Hakim @matanhakim.bsky.social · 04/08/2026

LEGIT! My sister has a neighbour with a cat who likes to go rogue and visit other people's homes, like my sister's. When told about it, the neighbour was like "I'm sorry, but ain't nothing to do about it 🤷‍♀️"

000

Matan Hakim @matanhakim.bsky.social · 04/08/2026

So excited to be chosen to this week's #Rweekly Highlights! Shoutout to the podcast hosts @rpodcast.bsky.social and @mike-thomas.bsky.social and to this week's curator @batoolmm.bsky.social, and much appreciation for this recognition 🙏

A meme featuring a close-up of a white and tabby cat with wide eyes and a goofy smile. The top text reads: 'ME, SEEING MYSELF IN RWEEKLY HIGHLIGHTS:'

110

Matan Hakim @matanhakim.bsky.social · 02/08/2026

Obviously I hand the recording to Claude, which transcribes the recording with a local Whisper model, and then uses the transcript as a prompt.

010

Matan Hakim @matanhakim.bsky.social · 02/08/2026

I have moved my design and architecture phase to be vocal. I start a voice recording when I have free time and a clean mind - driving, walking etc. - and just spill everything to it. I try to spill as much thought and context as I can, rather than creating something structured. So far it's amazing.

120

Matan Hakim @matanhakim.bsky.social · 01/08/2026

All of which is a lovely excuse for etymology, since to encourage someone is literally to put courage into them. So Eric, Mike, consider yourselves en-couraged, and the longer version of the argument is here: www.matanhakim.com/posts/2026-0... #rstats

matanhakim.com

Cognitive Courage: The Real Bottleneck to Learning AI Tools

Learning a new AI workflow no longer means studying it first. The platform you work in is the platform you learn in, and the only thing between you and your first skill is the nerve to say ‘I don’t kn...

020

Matan Hakim @matanhakim.bsky.social · 01/08/2026

What's left is smaller and considerably harder, which is saying out loud, to a machine, that you don't know how to do this and would like to be shown. I've started calling that "cognitive courage", and I think it now quietly decides how fast any of us learn anything.

120

Matan Hakim @matanhakim.bsky.social · 01/08/2026

Which means the old barriers have quietly packed up and left. Not knowing how, not finding the information, and not managing to fit the technique into your own particular project are all things the agent now hands you on request, inside the file you already have open.

100

Matan Hakim @matanhakim.bsky.social · 01/08/2026

What happens now looks nothing like that. You're in the middle of something real, you notice you keep handing the agent the same instructions, you say "turn what we just did into a skill", and then you have one, having learned what a skill is by accidentally making one.

A two-part diagram. The top row, labelled THEN, learning was a separate project, shows four boxes in a straight line joined by arrows: Pick a skill, Study books and docs, Get stuck search ask, Finally apply, annotated weeks. The bottom row, labelled NOW, learning happens inside the work, shows a gold box labelled Cognitive courage feeding into three boxes, Try it in real work, Ask the agent, Learn, with a return arrow arcing over the top labelled repeat in minutes.

100

Matan Hakim @matanhakim.bsky.social · 01/08/2026

It caught me because that sentence belongs to the way learning used to work, where you picked a thing, read the book, built a toy project beside your real work, got stuck, searched, asked, and weeks later finally had it. Learning was its own project, on its own clock.

100

Matan Hakim @matanhakim.bsky.social · 01/08/2026

He mentioned that he's been very much a consumer of skills but hasn't authored one of his own yet, and that it's on his list for this year. Which is an entirely fair thing to say, and I suspect most of us have a version of it sitting on a list somewhere.

100

Matan Hakim @matanhakim.bsky.social · 01/08/2026

A few weeks ago @rpodcast.bsky.social was walking through @smachlis.bsky.social's AI coding tips on R Weekly Highlights, with @mike-thomas.bsky.social alongside him, and he said one small thing in passing that I have been turning over ever since. 🧵

121

Matan Hakim @matanhakim.bsky.social · 20/07/2026

There's a base and superstructure joke somewhere around here waiting to be called out loud

020

Matan Hakim @matanhakim.bsky.social · 27/06/2026

Current European civilization is probably the most advanced human society Earth has ever known, with the caveat of a weird ideological resistance to air conditioning

030

Matan Hakim @matanhakim.bsky.social · 26/06/2026

The floor rose. Technical toil is getting cheap; value moves up to judgment, context, and communication. We didn't have to retool for the AI era - the role was already shaped like it. Full post 👇 matanhakim.com/posts/2026-0...

matanhakim.com

Why Data Scientists Are Well Positioned for the AI Revolution

The three skills that have always defined the job – coding, statistics, and domain expertise – are exactly what it takes to direct AI tools like Claude Code, and to know when not to trust them.

020

Matan Hakim @matanhakim.bsky.social · 26/06/2026

Now zoom from *who* we are to *what we do all day*: the R4DS workflow - import, tidy, transform, visualize, model, communicate. AI takes over the toil (the boilerplate, that join you've written 100x). You bring the judgment: the edge cases, and communication - the part that was always the point.

A flow diagram. Six boxes run left to right: Import, Tidy, Transform, Visualize, Model, Communicate, joined by arrows. A frame labelled Understand encloses Transform, Visualize and Model, with a curved arrow labelled iterate looping from Model back to Transform. A larger frame labelled Program encloses the whole diagram.

110

Matan Hakim @matanhakim.bsky.social · 26/06/2026

Circle 3, stats. Under the hood an LLM IS a statistical model. A statistician never trusts a point estimate at face value - you ask how far to trust it, design a check, smell a bad quantitative claim. Call it calibrated distrust, the exact reflex for a tool that's confidently wrong sometimes.

110

Matan Hakim @matanhakim.bsky.social · 26/06/2026

Circle 2, domain expertise. This is what gives 'human in the loop' any teeth. Models hallucinate, confidently. Knowing your field is what lets you catch the plausible-but-wrong answer. It's the difference between running a model and being able to vouch for it.

100

Matan Hakim @matanhakim.bsky.social · 26/06/2026

Circle 1, coding. Data scientists aren't software engineers, but we've absorbed the principles: git, tests, reproducibility. So you don't just 'vibe' with an agent - you're in the driver's seat, reading the diff, asking for tests, catching the silent failure before it ships.

100

Matan Hakim @matanhakim.bsky.social · 26/06/2026

See the bottom overlap - coding + domain knowledge, minus stats? Conway called it the 'danger zone': you know enough to be dangerous, producing analysis that looks legit and is quietly wrong. LLMs hand coding skills to anyone who can type a sentence. The danger zone just got very crowded.

100

Matan Hakim @matanhakim.bsky.social · 26/06/2026

Two diagrams that predate LLMs by a decade describe the AI-era data scientist remarkably well. First, Drew Conway's 2010 Venn: data science sits where 3 skill sets overlap - coding, stats, and domain expertise. Holding all three is what lets you direct a model AND know when not to trust it.

A three-circle Venn diagram. The circles are labelled Math & Statistics, Hacking Skills, and Substantive Expertise. The pairwise overlaps are labelled Machine Learning, Traditional Research, and Danger Zone; the central overlap of all three is labelled Data Science.

130

Matan Hakim @matanhakim.bsky.social · 26/06/2026

Most of the AI conversation is about whose job is at risk. I want to make the opposite case: data scientists are about as well positioned as anyone to *benefit* from this wave of tools. And the reason is structural - the role was already shaped like the AI era's demands. 🧵👇 #Rstats #DataScience

140

Matan Hakim @matanhakim.bsky.social · 21/06/2026

In my latest experience it depends on the use case: 1. for interactive EDA best to use Posit(ron) Assistant; 2. and for package/app development or data analysis which is straightforward and with minimal gotchas best to use Claude Code via the terminal or VS Code extension

021

Matan Hakim @matanhakim.bsky.social · 21/06/2026

I am all for that, the problem starts when I am too loud cleaning the house after my wife goes to sleep so I get stopped 😭

100

Matan Hakim @matanhakim.bsky.social · 13/06/2026

Eyes rolling so hard they be scratching them glasses 🤓

100

Matan Hakim @matanhakim.bsky.social · 13/06/2026

A man can live in the US for decades and become practically American, but would always need his rational, intuitive and well-designed metric system for baking over ounces and pounds

030

Matan Hakim @matanhakim.bsky.social · 10/06/2026

You forgot to say that the fish is first brined and then fried🫣

010

Matan Hakim @matanhakim.bsky.social · 02/06/2026

Had to skip it due to reasons but when I saw the calendar notification right before I knew I was missing 🥺

110

Matan Hakim @matanhakim.bsky.social · 26/05/2026

Just letting y'all know that Data Science Lab by @posit.co is THE BOMB and if you haven't checked it out already you really should join sometime! Shout-out @ivelasq3.bsky.social @libbyheeren.bsky.social You're doing awesome work 💯🤟

2123