Ron Sperber

@KirkTexas @ThrillaRilla369 Haha all good until you got banned by Metallica

Anyone who surfed the early web between 1995-2009, what’s the one website you still think about?

@SGTWipper1Each @ThrillaRilla369 This is what I was going to respond with. I can still hear them singing in my head. 😅

@ThrillaRilla369 Who remembers this gem?

@KristenKylecast @ThrillaRilla369 Badger badger badger badger badger badger.... Mushroom

@ThrillaRilla369 Ebaum’s World 🤣🤣🤣

@AlfinCodes COBOL

Guess the programming language

@MichalYouDoing True, but what that's missing that the LLM would have some chance to address is asking questions. If you read a text and you don't understand something or need clarification, you are out of luck. You could ask an LLM questions.

Hot take: Before LLMs, you could have also read all the books relevant to a university course for a tiny fraction of the price. And your education would have been less full of hallucinations and rubbish.

@mischavdburg While this is true, the reason the models do well with Python is because there was a LOT of data with python code that they were trained on. If you try to go to a language (or other representation) with significantly less training data, it's not as likely to do well.

Python and Go were designed for us. Readable syntax, verbose on purpose, so a human could follow along. AI doesn't need any of that. I keep wondering if there's a point where the models settle on some representation that's just faster for them, and quietly stops being readable to us. Not obfuscated. Just not built for human eyes anymore. I don't know what that does to how we review code, or trust it, or teach it. Still sitting with that one.

@JerusalemDemsas @TheArgumentMag It's not that hard to understand. One word: "tokenization"

It's really weird that AI can both solve incredibly complex math problems and can't reliably tell you how many days of the week contain the letter "n". I asked @TheArgumentMag writer Kobe Yank-Jacobs about this and his answer helped me understand the form and function of human and AI memory. theargumentmag.com/p/why-ai-can-s…

@Pragati_200 Yeah definitely not. I had an example today where I asked Claude code to go through my codebase. It found something it said just needed a docstring. I looked at what was going on and saw it was a deeper issue. Was fixed, but if I didn't understand the codebase, I'd be SOL

Unpopular opinion: AI makes everyone a developer the same way calculators made everyone a mathematician.

@yadavji_codes Some version of a checksum

Ever wondered You entered a 16-digit card number, and the form instantly says "invalid card number ! There are billions of possible numbers How is this check done so quickly?

@najeebsulaiman_ "Go ahead and run your convolutional + LSTM network on video data on your CPU and tell me how that goes for you"

𝐖𝐡𝐲 𝐢𝐬 𝐚 𝐆𝐏𝐔 𝐛𝐞𝐭𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝐚 𝐂𝐏𝐔 𝐟𝐨𝐫 𝐀𝐈? A few months ago, I was explaining AI infrastructure to someone, and they asked me a simple question: "If my laptop already has a CPU, why do AI companies spend billions of dollars buying GPUs?" 🧵

@najeebsulaiman_ GPUs were designed for graphics, which uses matrix computations that are the basis of any type of neural network

@shub0414 @ConsciousRide LLMs can't just "pick a random number". That's not how they work

@ConsciousRide but their is the word random, not just pick a number between 1 to 100, i think it should consider that

AI Engineers why does chatgpt always pick 73?

@shub0414 Llms are predicting next tokens based on training data. They aren't actually generating random numbers

@mandylu What do you mean? Transformers, diffusion models, etc are well understood

we still have no satisfying theory for why AI works

@yourclouddude It will be 0. a .get() on a dict attempts to get the value associated with the key. If the key doesn't exist it returns the value afterwards (or None if no value is given)

What is the output of this? Reply with your answer + why 👇

@adityaseeks I absolutely loathe leetcode/hackerrank etc style tests. I've taken some where they forbid you to search for anything. It's just not realistic. I took one where I was expected to use a Random Forest. That meant if I forgot that it was in sklearn.ensemble, I'd be SOL. That's dumb.

I don’t think in this era of AI it is still relevant to ask questions from leetcode. The more focus should be on the understanding of architectures, algorithms and how does the thing works, and yeah knowing the concepts of coding languages.

Name a movie you've seen more than 10 times with just a GIF

@vivoplt for what I use AI for mostly? Qwen

No GPT. No Claude. No Gemini. What’s your go to AI model now?

@valigo Uber burning through it's whole 2026 budget in Q1 is...wild. And if they really had a leaderboard for their engineers for token usage intended as a positive thing, that was insane.

It's absolutely beyond comprehension that all big tech companies burned through billions of dollars in tokens in 2026 already. And their products haven't improved at all. They only gotten worse. Looks like the only ones actually benefitting from AI are indies and small teams.

@JamesSurowiecki I get most of your complaints, but it's certainly true that some restaurants have started adding "service fees" or "service charges" that aren't tips, and that is pretty obnoxious.

Ragebait tweet written in May 2026 claiming that a receipt from November 2025 happened "tonight," and then insisting that restaurant customers are supposed to pay 20% for a "service charge" and another 20% for a tip. This site is just riddled with crap.