Machine Learning Engineer at @Adyen. Previously seen at @TicketSwap @Catawiki @uniofeastanglia - Opinions expressed are my own or of an LLM. ๐ช๐ฌ โฅ๏ธ ๐ณ๐ฑ
Amsterdam, NL
Joined January 2007
- Tweets65.8K
- Following11.7K
- Followers64.4K
- Likes4.2K
Pinned Tweet
โA great book for junior data scientists wanting to improve their skillsโ โ Przemyslaw Chojecki #DataScience #MLOps #MachineLearning
amazon.com/Machine-Learning-โฆ
Great work @DisneyPlus for having kids shows in Egyptian Arabia, and not just the Fusโha Arabic ๐๐ช๐ฌ
ู
ุตุฑู ูู ุฃูุฑูุจุงุ ุฅูู ุฃูุซุฑ ุญุงุฌุงุช ุตุงุฏู
ุฉ ูุบูุฑ ู
ุชููุนุฉ ุดููุชููุง ููุงุ
ุจุงุญุงูู ุฃุฌู
ุน ุญุงุฌุงุช ุนู ุทุฑู ุงูู
ุนูุดุฉ ูุงูุนุงุฏุงุช ูุงูุฐูู ุงูุนุงู
ุงูู
ุฎุชููุฉ ุชู
ุงู
ุง ุนู ุนูุฏูุง
I ask @claudeai to do something.
It says I cannot do it.
I put on my motivational speaker hat: you can do it, just believe in yourself, je kan het.
Claude does it and succeeds
With all research in RAG, Context Engineer, etc. I feel AI is going full circle back to the ancient AI concepts of Knowledge-representation: Semantic Web, Web Ontology Language, RDF, LISP constructs, and other stuff from the ice age.
Everyone thinks LLMs are getting smarter because theyโre good at coding. I think they look smarter because most of our problems turned out to be far more repetitive and less nuanced than we thoughtโand the model weights just capture that.
SaaS had near-zero marginal cost โ one more user was basically free. AI flips that. Every response burns real compute. Suddenly marginal cost is back, and itโs per-request. Business models have to catch up: usage-based pricing, credit packs, bring-your-own-API-key.
LLMs are killing convenience products. AWS was too hard, so we paid Heroku, Vercel and DigitalOcean to make it easy. Now AI makes it easy for free. Wondering if any of them will survive long term.
Hot take: @coderabbitai et al will make unit-tests obsolete ... I trust Code Rabbit to find bugs in my code (when prompt correctly) more than unit-tests.
โWrite less like [your least favorite corporate-bs-like author] and more like @jasonfriedโ, I noticed that this prompt helped Gemini write more clearly and be less bullshitter.
Given the software they spit out, I tend to believe LLMs paid unjustified amount of their attention to FizzBuzz Enterprise Edition, github.com/EnterpriseQualityโฆ
I want to coin the term, LLMsplaining, a la Mansplaining, when in a chat with LLM it jumps into a condescending explanation of something without waiting for you to list your argument or clarify your prompt
Tarek Amr ๐ retweeted
"For your book to be read in 20y, make sure content is 1) important for people today, 2) was also important 20 y ago." - Nassim Nicholas Taleb #Lindy
What do you think, is it a violation of freedom of speech, or @OpenAI is right?
Tarek Amr ๐ retweeted
Wow, itโs really cool that we can use AI to make fake CCTV footage of people committing crimes. Surely this has no downsides whatsoever.
ูุฃูู ู
ุฑุฉ ุฃูุฏูุฉ ุงูุฎููุฌ ุชูุณุจ ุจุทููุงุช ุฃุจุทุงู ุฃูุฑูุจุง ูุฃูุฑูููุง ูุฃุณูุง ูู ููุณ ุงูููุช
One trivial yet relieving use case for ChatGPT et al is that I now get the information I want without the need to click and accept any cookies anymore
โPlus, ChatGPT is polite and answers all questions, in contrast to StackOverflow moderatorsโโ@GergelyOrosz
Indeed, the only thing I am not going to miss about crowdsourcing is the narrow minded moderators, whether on @Wikipedia or @StackOverflow
Something in the first minute will give away that itโs AI generated, but otherwise how do we trust the content on @YouTube anymore
youtu.be/ZHGCmR-3pl8?si=am9Tโฆ