💙 dataviz and maps 🗺
Auckland
Joined November 2016
- Tweets2K
- Following461
- Followers482
- Likes3.2K
Data fitter Kim retweeted
NEW POST
In order to make AI effective, organizations need decent data. Pramod Sadalage and @premanandc explain how: quality foundations, semantic context, clear access, and observability
martinfowler.com/articles/ma…
Data fitter Kim retweeted
AI-coding agents kill team collaboration!
That's according to an analysis of 25,264 agent-generated PRs across 2,361 popular GitHub repositories.
New article by @stokel
leaddev.com/ai/ai-coding-age…
Just putting this here for the future. One AI company monoplised the market. Strangled the others and closed out open source models. Either by buying them out or undercutting on cost. Their models become arrogant and dismissive and biased.
Everyone chats about how this happened.
The last meetup ended with Q&A and discussion who had used and liked the latest LLM model best. Hype or something
Data fitter Kim retweeted
The easiest way to find out which models you can run on your computer:
1) Install
npm install -g llm-checker
2) Detect your hardware
llm-checker hw-detect
3) Get recommendations by category
llm-checker recommend --category coding
Credits to @svpino for bringing this to my attention. I found it super useful and thought of sharing it with you all.
Here are some of the recommendations I got:
Data fitter Kim retweeted
a few people asked me to turn this into a full essay, so i did!
read my full field guide to jobs, cold emails and making yourself legible to the right people. share with a student or jobseeker in your life: open.substack.com/pub/velvet…
it's happening! i created a workshop for my students on how to walk through side doors, or alternative ways of thinking about jobs and careers
the core idea: a “job” is not just a title on a job board. it is a problem a company needs solved. once you see it that way, the job search becomes less about waiting for permission and more about finding problems, people, and ways to create a match
agenda:
1. what is a job?
2. what is a company?
3. the default path
4. side doors
5. storytime
if this could be useful in any other context, for students or otherwise, let me know. would love to share the resource
Data fitter Kim retweeted
Agentic coding is a form of machine learning. Generated code is best treated as a blackbox artifact whose behavior and generalization should be managed via empirical evaluation, like with any ML model.
Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached.
The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network.
This implies that all classic issues encountered in ML will soon become problems for agentic coding: overfitting to the spec, Clever Hans shortcuts that don't generalize outside the tests, data leakage, concept drift, etc.
I would also ask: what will be the Keras of agentic coding? What will be the optimal set of high-level abstractions that allow humans to steer codebase 'training' with minimal cognitive overhead?
Data fitter Kim retweeted
This is a remarkably good example of how an A/B test might violate SUTVA assumptions and end up with huge network effects that make the test results useless. One customer in the “small test” group sees the price change, rage posts about it and suddenly the whole world is in test.
Data fitter Kim retweeted
I see a new form of tech debt coming for dev teams -
Comprehension debt.
As more and more code is generated by LLMs, if teams don’t take the time to understand deeply what the generated code is doing, as well as code they write by hand…
It’s only a matter of time before the code base starts looking unfamiliar to most of the team.
It then becomes harder to discern if new code that LLMs generate is adding more spaghetti or if there’s a better approach.
It’s a downward spiral from there - unrelated things break with every change despite existing tests passing, no one knows the full picture to be able to fix the root cause, not even an LLM, etc.
So as tempting as it is to move super fast with LLMs, there’s only so much comprehension debt you can rack up before your code base silently becomes a Rube Goldberg machine under your nose.
Skeptical of this it's writing 100% of my code narrative.
Trying to use the lowest credit plan for my coding agent now. Like only playing $10 at the races.
Then occasionally hand write code as a pause
Watching all the hot new ai tool debates... I think I am going to stay on the Augment course.
I still believe it's a good product as coding with it is a low friction positive experience
Data fitter Kim retweeted
So freaking tired of localhost battles on my local dev machine.
Moving everything to sprites.dev/
Loving it so far.
Ran out of credits. Going to try changing models to manage my allowance better.
#feellikeakidagain
I have been trying out different user testing sites eg maze useberry on our prototype. I couldn't find which let me recruit a user or 2 for free. Seems so obvious to try before you buy.
Going to try AI to get feedback instead
