fcatalan

I haven't had an useful meeting in years. All the important collaboration and decision making has happened organically in text chat, which is great because it's all searchable and dated, and I do refer to that a lot. In fact they recently moved my main collaborator from another building into the next desk and we agreed to keep the work stuff in chat as much as possible so it isn't lost. So we chitchat about our kids but still type out our debate about the best version launch date.

Every meeting in person or via Zoom I have been in has been either an useless sales pitch, grandstanding by some manager, brown-nosing by some upstart or some other form of toxic socialization, scheming or conspiracy. I detest all those and avoid them, which is probably why I've become kind of an unpromotable pariah, which is ok, as a promotion would mean attending more of them.

I use it a lot for reducing friction. When I procrastinate about starting something I ask the AI to come up with a quick plan. Maybe I'll just follow the first step, but it gets me going.

Sometimes I´ll even go a bit crazy on this planning thing and do things a bit similar to what this guy shows: https://www.youtube.com/watch?v=XY4sFxLmMvw I tend to steer the process more myself, but typing whatever vague ideas are in my mind and ending up in minutes with a milestone and ticket list is very enabling, even if it isn´t perfect.

I also do more "drive by" small improvements:

- Annoying things that weren't important enough for a side quest writing a shell script, now have a shell script or an ansible playbook.

- That ugly CSS in an internal tool untouched for 5 years? fixed in 1 minute.

- The small prototype put into production with 0 documentation years ago? I ask an agentic tool to provide a basic readme and then edit it a bit so it doesn´t lie, well worth 15 minutes.

I also give it a first shot at finding the cause of bugs/problems. Most of the time it doesn't work, but in the last week it found right away the cause of some long standing subtle problems we had in a couple places.

I have also had sometimes luck providing it with single functions or modules that work but need some improvement (make this more DRY, improve error handling, log this or that...) Here I´m very conservative with the results because as you said it can be dangerous.

So am I more productive? I guess so, I don't think 4x or even 2x, I don't think projects are getting done much faster overall, but stuff that wouldn't have been done otherwise is being done.

What usually falls flat is trying to go on a more "vibe-coding" route. I have tried to come up with a couple small internal tools and things like that, and after promising starts, the agents just can't deal with the complexity without needing so much help that I'd just go faster by myself.

To be honest not visiting some websites is one of my main uses of Perplexity.

For example I like to watch F1 and I like to know the times for all sessions in my timezone during the weekend.

It's surprisingly hard to find this information, because the Google search is SEOed to hell and back by sites that hide the information behind endless articles full of irrelevant AI slop and 2 million intrusive ads, and that's if they have it right or at all.

Perplexity wades through all that shit, gives me a neatly formatted table and has never been wrong so far.

So I can see where the BBC is coming from but I also don't really want them to win.

I brought over the source of the Dear imgui library to a toy project and Cline/Gemini2.5 hallucinated the interface and when the compilation failed started editing the library to conform with it. I was all like: Nono no no no stop.

Yes there's a feedback page for each round explaining the pic and showing the location, also a breakdown page after every 5 round match.

Yes, score rankings are missing. There's a leaderboard for the daily, but only shows accounts you have friended.

For reference, I tend to hover around 44k for the daily. 48k is a lucky/easy day, 40k some disaster happened. Under 40 is very rare. I like to think that would be above average, but without rankings I don't know :)

I do the timeguessr daily every day almost as a ritual, it's fun, satisfying and educative.

It's also very interesting to watch people try after you've done it: how they catch on stuff you missed, their strategies to find and date things, or how they infuriatingly ignore whatever was the giveaway to you.

I have observed my 16 year old incorporate LLMs into her workflow and most of it has been an improvement.

For a start, let's be realistic, before LLMs, unless some subject specially caught her fancy, most of her output was trite rewordings of Wikipedia plus some extra bits from the first couple of sites that came up in Google. The myth of "your own personal essay" was always a myth even before the digital age. Things came from the paper encyclopedia, then Encarta then Wikipedia. At college level, that and the reading list from the course.

Nowadays she prompts and reprompts deep research mode, checks the sources to avoid hallucinations and be able to defend her work if challenged, prunes and reworks the outline to her liking, brings things to her level and manually removes LLM blandness...

All in all I think she gets more than before from the whole exercise and the output is much better.

She also does things like fixing bad study material she gets handed, asking for clarifications and alternative explanations for stuff she is initially confused by, generating practice quizzes...

By her account, her peers handing in straight cut&paste slop from ChatGPT are the ones that previously didn't hand in anything at all. I've also seen her occasionally do that as retaliation against bad faith or pure make-work assignments meant as collective punishment, which I find... fair?

I concur that SICP 4.4 is very approachable. I once took a class that had a simple Prolog assignment, I recall we were given some building plans and had to program a path solver through the building. I thought it was a bit too easy and I wanted to dig deeper, because just doing the task left you with a taste of "this is magic, just use these spells".

I looked at how to implement Prolog and was stumped until I found that SICP section.

So I ported it to JavaScript and gave it a Prolog-like syntax and made a web page where you could run the assignment but also exposed the inner workings, and it was one of the neatest things I've ever handed in as coursework.

I need to have a closer look at this. Mostly because I was surprised recently while experimenting with making a dieting advice agent. I built a prompt to guide the recommendations "only healthy foods, low purines, low inflammation blah blah" and then gave it simple tools to have a memory of previous meals, ingredient availability, grocery ticket input and so on.

The main interface was still chat.

The surprise was that when I tried to talk about anything else in that chat, the LLM (gemini2.5) flatly refused to engage, telling me something like "I will only assist with healthy meal recommendations". I was surprised because nothing in the prompt was so restrictive, in no way I had told it to do that, just gave it mainly positive rules in the form of "when this happens do that".

9/11 was a big turning point in my experience. American conservatives that I considered online friends were simple impossible to reason with within days and completely alien beings after a few weeks.

But Rogue One ends literally setting the opening chase at the beginning of A New Hope, so no space for a straight sequel there. Maybe some Kleya "John Wick in Space" sidequest?

Roman elites spoke a lot of Greek among themselves, a bit similar to how Tsarist nobility used French

The stove must be touched, there's no other way

Oh I didn't mind it, the response is in fact right: it's not very realistic to extrapolate early diet results, and people come up with all kinds of potentially harmful crazy diets, so better to add the warning. I just wanted to emphasise that I deliberately avoided to drop any early clues about the nature of the numbers as I just wanted the (very probably wrong) results without any further comments, and it was interesting (maybe not really surprising) that the LLM would still easily guess what they were about when prompted

Oh the usual "linear weight loss predictions might not hold", "if you are on a restrictive diet make sure you are supervised by a doctor" and so on.

I'm trying to lose some weight, and while bored I pasted a few data points into Gemini to do some dumb extrapolation, just a list of dates and weights. No field names, no units.

I specifically avoided mentioning anything that would trigger any tut-tutting about the whole thing being a dumb exercise. Just anonymous linear regression.

Then when I finished I asked it to guess what we were talking about. It nailed it: the reasoning output was spot on, considering every clue: The amounts, the precision, the rate of decrease, the dates I had been asking about and human psychology. It briefly considered the chance of tracking some resource to plan for replacement but essentially said "nah human cares more about looking good this summer".

Then it gave me all the caveats and reprimands...

You can also see that the hanging yellow part of the headscarf, he just winged it, effective as it might be.

I paint as a sort of weekly ritual, just 2 hours every Wednesday evening, and did an inept copy of this as my first serious try. Months of staring closely at every little detail of it leave you in a sort of communion with the work and the artist.

One thing you quickly learn is that the old masters were "impressionists" too. If you overwork stuff trying to perfect every shape with hundreds of precise brushstrokes, you end up with a naive, infantile looking painting that feels "unpainterly".

Trying and failing to mimic that single quick brushtroke that fools the eye leaves you in awe, fully appreciating the mastery.

I have a similar philosophy for the systems I manage. We have always been severely understaffed, so I treat any user support request that repeats twice as a bug.

If the decision to push a button is yours, I'll give you the button. If you need some data more than once, you get a button too. My ideal user never needs to know who manages the system or how to contact us.

This has even got me a "why do you guys have almost no tickets? You aren't doing anything!" talk a couple times. Music for my ears.

This reminds me of my first "paid" job. I was contracted with a friend to fix the decaying IT infrastructure of a mid-sized local business. We were 18, early 90s. The contract was through a friend of my friend and was a shady affair from the start: nothing in writing or signed.

We agreed upfront to get about 600$ each, which was woefully little for the ordeal we had to go through. The place was a mess of DOS PCs, AIX workstations and ancient early macs. The backups were a cardboard box of unlabeled 5 1/4 inch floppies. We even had to solder our own cables to get some ancient serial printers to work and get the new IP network talking to stuff that wasn't built for it at all.

The owner was paranoid about security so we passworded everything that could be and compiled a long list of passwords.

When everything was working we sat with him to get our money, and then he said he had understood that the deal was $600 shared between us. He didn't budge, so we got nothing and he didn't get the password list.

Yes I understand the encodings will be close and that helps, I guess that's why they produce good translations, but I'm intrigued by the LLM having so much control of the switching without even explicit prompting, just a one-shot example. I also guess I'm easily impressed.

Those texts aren't nonsense. The prompt has a meaning, the LLMs are able to understand it, and are able to reply with coherent and understandable responses crafted in the same way the prompt was written. For me it's a very clear example of something that is very far from any training data coming out of the models. Intelligent? No, but for me it points to the idea that "language is solved".

Another thing that LLMs are (surprisingly?) good at is randomly mixed language conversations, here's something I tried a few days ago:

Prompt: Kan du comprender questo text? Anata wa quelque inkling habe about lo que se parla dans la presente oraçao? Repondez med sam way of goboriu.

ChatGPT:

Ja, ich kann verstehen este mix de idiomas! Tu es usando une combinación de várias línguas pour expresar una pregunta. Je vais responder en the same modo:

Si tu veux kommunicieren en cette manière, c'est possible, aber il peut être ein bisschen confusing pour andere people! 你觉得这个方式有趣吗？ O préfères-tu un idioma más claro?

Deepseek:

Ja, jeg kan forstå en lille bit av ciò que tu está dizendo. Anata no message wa molto interessante, pero un poco confusing aussi. Si tu peux écrire in una lingua sola, sería plus facile para mi at comprender. Danke, merci, arigatou!

Claude:

Oui, je kan entender algo of this texte mezclado! C'est un interessante modo de communicare using différentes languages together. Je trouve ça quite divertido y puedo responder en la misma manera mixta. Ist das what you were looking for? Watashi wa happy to continue questa conversation in whatever linguistic style anata wa prefer!

Perplexity:

Ja, ich kann diesen Text verstehen! Vous avez créé una frase muy interesante con palabras from different languages. Let me rispondere using a similar mishmash of lingue:

Questo è un esempio fascinante di code-switching e language mixing. Watashi wa impressed by la creatività y el modo en que vous avez combiné various idiomas in one sentence. Es ist ein poco challenging zu decifrar, aber anche molto divertente!

Ich würde sagen que el mensaje principal è about compréhension linguistique et la capacité de entender verschiedene Sprachen when they're mixed together. C'est une manière intéressante de tester language skills et cognitive flexibility.

Grazie for questa stimolante question! J'espère que mi respuesta was equally entertaining und in lo mismo estilo multilingüe.

"after doing his time"

So then have BigBalls at least do his time before handing him the keys to the kingdom.

The friction was high enough for me to not do it: Coming up with a week-long list of stuff that checks all the boxes: healthy, balanced, non repetitive, feasible in my cooking time, ingredients available in my store and I'll actually eat it... it's not a trivial effort, then condensing it into a shopping list is additional clerical work. This was just refining the prompt a couple times to see if I could get all those aspects considered.

Another factor is that having the menus come from an outside source keeps them 'honest', I wouldn't 100% trust myself not to include some of the crap I was precisely trying to get rid of If I made them from scratch

The result was surprisingly good... maybe not 100% right, more like 90%, but good enough to overcome that friction.

I wanted to get my diet in order, so I asked Perplexity for a nutritionally balanced weekly menu and a shopping list compatible with it and the store I go to, and it did pretty well in both tasks. On top of eating healthier and more varied, after a few weeks I noticed my weekly food bill was consistently down by about 20%

You can just prompt for it:

"KI-Philologenstaunverunsicherung"

Breaking it down:

KI – (Künstliche Intelligenz, Artificial Intelligence) Philologen – (Philologists, scholars of language and linguistics) Staun – (from staunen, to marvel or be astonished) Verunsicherung – (a sense of uncertainty or unease)

https://chatgpt.com/share/67d80661-4bb4-8012-9328-77d56af52b...

I think some Game Theory 101 might be relevant here. Yes it's good to cooperate initially but after that, tit for tat is the thing to do:

https://en.wikipedia.org/wiki/Tit_for_tat

Hungary, Turkey, India, Venezuela come to mind. Poland, Brazil are also recent near misses. South Korea also had a recent oops. So a well trodden path.

Duolingo is a great kickstarter from zero knowledge, in my experience it also leads to very fast reading proficiency.

I'm convinced I can make sense of a newspaper within weeks and read novels within months in any language just using Duolingo (maybe longer if I also have to learn a different writing system). I did it with Danish years ago. I could read newspapers after a month and after four months I read a few novels at a slow but still enjoyable pace.

Or another example: my daughter was struggling with English at school and I bribed her to keep a 200-day Duolingo streak. That was enough to allow her to understand sources in English about her hobbies and interests, and now after a couple years her level is way above what is taught at school, she aces every test effortlessly.

Fluency... English is not my first language, but I can write, read and understand it pretty well. I even "think in English" at times, but I have never spent more than 4 consecutive days in an English-speaking country, and my fluency suffers a lot comparatively.

I can speak about anything, even give a talk or teach, but my accent is very thick, I'll stammer and stutter and stop here and there to find the right turn of phrase, or struggle to pronounce words that I have written a thousand times but never said aloud.

So in my opinion if you want to speak well, you have to speak a lot and I don´t think Duolingo is very good for that.