Karsten Konrad 🇪🇺

What an odd thing to say for a computational linguist.

Skull issue.

Jetzt habe ich Schwaben gelesen und fand das sehr beunruhigend.

Surprisingly, that’s still orders of magnitude better than having no tests.

Algebraic creatures Made with #Python #NumPy #Matplotlib and
@marimo.io

Tom Cruise doing another movie?

You can not possibly fully appreciate Claude Code unless you talk to it in its native German.

* Actually, my Apple ][+ had only 32k of RAM when I got it for Christmas. I had to buy a special card from this random company called Microsoft that I'd never heard of that would expand the memory to 64k. (The 80 column and lower case card would come later.)

For some reason, nobody married data mining tools and optimization into the same tool before, I believe. It’s pretty obvious, isn’t it, that tools for preparing data and optimization fit well together?

We get as input sales data from customers, and design rule-based systems for pricing for that. Our task is to make sure that the new rules do not deviate too much from the original sales data (optimization) and are consistent - adding many rules may make them unsatisfiable. Z3 can do both.

I have integrated Z3 in my data mining app. I use my own specification language, which an LLM translates from natural language. I should ave looked into a more native format ;)

Anyone here knows a good book on constraint programming, SAT/SMT solving or MILP optimization? I want to brush up the basics, there must be better books now than what we had in the early 2000s.

Damn.

Me: "May I have two scoops of cookie dough please?"
Waiter: "Yes, you wanna spoon?"
Me: "Sure, what time do you get off?"

You reposted me with a very clear intend, to make fun of an argument without actually engaging. I consider that a breach of trust. Next time someone starts a conversation about a topic with you, maybe give it more time bf. telling all the internet how ridiculous you find it.

If you are wondering whether the AI discourse on Bsky is done in good faith, just cite one counterexample to some universal statement („LLMs are useless…“, „No LLM will ever…“, „All productivity gains are a lie or caused by inexperience“) and watch the moving of the goalpost. Every single time.

But I told you I was building apps for decades, and tbh, Claude can do a very good job in that part of the coding I am experienced it, at much higher Speed. I am not convinced your argument here carries much water. Very much „no calculator will be as fast as me and my slide rule“ vibes here.

Your bar of being experienced seems to be „full stack developer“. You are kind of an ass here.

In my app, yes. Machine learning, optimization. I know the algorithms and APIs for the core engine very well, and used these myself before in scripts. What I needed the AI for was gluing together, writing tests, all of UI and building an installer. And pandas to polars translation. I use TDD.

Sorry, I was unclear: I am a professional consultant now, but am a computer scientist.

Junior developers. I have been developing sw for 20+ years, mostly core ML algorithms. But stopped 15 years ago, and never did frontend development bc. I worked in teams with specific skillsets. I do not have a team available rn.

I am a business consultant, not a developer. I spend about 300h on this, a lot in my spare time. Barely managed to find that time. 3000h would be me investing 3 years - the opportunity costs become to high to justify the whole project.

10x pretty much covers it, and I am not inexperienced. I am a bit rusty, but it seems I have a good intuition of how tasks can be modularized and explained to an LLM. My main project is a 50k LOC data mining app I’d never been able to complete myself.

I don’t have a tattoo, but this qualifies as a candidate.

Use the blue to red coloring, not red to green. Colorblindness is something to consider when you only use colors,

As has everyone else... thank you for your thoughts. I am still not 100% sure that the article is genuinen, that is not pulling our collective legs. I truly hope not.

Does anyone use balanced replication weights & multiple matrix design in surveys in real life, outside of PISA / TIMSS?
We are having to learn this stuff on my Masters course: am wondering how useful it is. Thanks.

An eigen-starfish!
Made with #python #numpy #matplotlib and @marimo.io

Btw., I am not a native speaker, but wouldn’t „You can find my PhD on natural language semantics in the library of congress“ fit my linguistics degree into less than 300 characters?

You were arguing that LLMs only do syntax. I gave you a simple method to show LLMs can do semantic reasoning. Here, you move the goalpost to pragmatics. Want to bet LLMs can reason about that too? They can’t count words very well, I give you that.