Sinan Taifour

I tried to elaborate that they're not purely random (and we can probably predict a patient's visit time based on history and description of the case), and we can already do some optimizations based on average visit times. Unfortunately his opinion didn't change.

His response was "it can't be done", his argument being that patient visit times are random.

I always spend at least an hour waiting for my appointment at the doctor in Jordan. Today after a 2-hour wait, I offered the doctor to optimize waiting times without any loss of business, for free.

I think with time it's less meaningful to look for Arabic LLMs; SOTA LLMs are now doing great in Arabic. Maybe the only reasons would be if you want to deploy something isolated (e.g. on premise for a bank, or you want to control some answers). Our evaluation at @MaqsamHQ is that Open Source LLMs are far behind SOTA closed source models in Arabic. Since there aren't standard evaluation datasets or techniques, LLM developers will pick a dataset and metric (e.g. We use perplexity as a metric, based on a proprietary dataset). Some might decide to manipulate their pick to look better; you can always lie with accurate statistics 😀

I'm still on the hunt for the best Arabic LLM. Everyone claims they are the best on their website, even share stats where it shows they beat all other models. But if everyone is "scientifically" better than the other, something is wrong with that science.

I ended up trying @Shortwave, I was very excited, but it sucked too. Every time I ask anything it only checks a max of 100 email threads before it answers. Needless to say "organize my inbox" (a prompt they promote) is useless if I have ~1500 unread email and they read 100 only.

Come on, #AI integrated into #Gmail should be able to do better! Reminds me of when Google Home understood "Turn off the lights in the kitchen" but not "Turn off the lights in the kitchen and bedroom". When will assistants actually act as assitants?

What does this #Ruby code print (12, or 124), and how do you explain the behavior of break in #Ruby?

@3blue1brown explains what makes people engage with math in this great #TED talk: youtube.com/watch?v=s_L-fp…

How excited I was about #BTC at $7, it's hard to believe today.

@ahmad_alhour "Numerical Methods for Scientists and Engineers". It's written in the 60s, but Hamming already was discussing AI. Check out the last chapter numbered "N+1", literally.

@sinantaifour What's the name of the book, btw?

A couple of weeks ago, I wanted to pick up a new side project and decided to work on a distributed calculation of #Pi. (1/6)

Introducing complexity should be strongly justified. "I might need this extra abstraction in the future" is not strong enough. If you really believe you will need it, write code that can easily be refactored, but don't introduce the abstraction until it is necessary.

@fmardini Apart from helping me get unstuck, his book had a motto very relevant to what I'm doing: "The purpose of scientific computing is insights, not numbers". (6/6)

I got stuck in tracking my roundoff errors. I chatted with @fmardini about it. He pointed me to a book by Richard Hamming (whom I admire). (5/6)

@ayman60 Maybe 200-300 sqm. Do you have a lead to cut my search short? 😁

@sinantaifour How many square meters ?

I've been looking for an apartment with a view. My geeky solution: pull 30m resolution altitude data of the city, calculate the gradient at every point, overlay a map, and find the most likely places to have nice views.

@tambi_jalouqa @Aqabawe I've found a couple of unexpected places that are on the "wrong side" of a neighborhood border; amazing gradient and hopefully much cheaper. I wish we had public selling prices like in the US!

@Aqabawe @sinantaifour Look for cases on the edge where the view is zone A. But the building is zone B, thus cheaper.