LWS

@emollick My eyes

(Sorry, after seeing so many of these, could not resist): 🚨 BREAKING: Google just dropped a NEW paper that completely deletes RNNs from existence. No recurrence. No convolutions. Nothing. Just one mechanism. And it’s destroying every translation benchmark on the planet. The title alone is a flex: “Attention Is All You Need” Vaswani. Shazeer. Parmar. Uszkoreit. Jones. Gomez. Kaiser. Polosukhin. 8 researchers. 1 architecture. The entire field of NLP will never be the same. Here’s why this is INSANE → LSTMs took DAYS to train. This thing trains in 12 hours on 8 GPUs. 🤯 → 28.4 BLEU on English-to-German. That’s not an improvement. That’s a MASSACRE. They beat the previous SOTA by over 2 points. → English-to-French? 41.8 BLEU. At a FRACTION of the training cost of every model that came before it. → They called it the “Transformer.” The name alone tells you they knew. But here’s the part nobody is talking about 👇 They threw out sequential processing ENTIRELY. Every other model on Earth processes words one at a time. This thing looks at the ENTIRE sentence simultaneously and figures out what matters. It’s called “self-attention” and it’s basically the model asking itself: “which words should I care about right now?” Every. Single. Token. In parallel. Do you understand what this means? Training that used to take WEEKS now takes HOURS. Models that couldn’t scale past a few layers? This thing stacks 6 encoders and 6 decoders like it’s nothing. And the multi-head attention? 8 attention heads running at once, each learning DIFFERENT relationships in the data. I’m not being dramatic when I say this paper just rewrote the rulebook. RNNs are cooked. 💀 LSTMs are cooked. 💀 The future is attention. And attention is ALL you need. Follow for more 🔔

@dioscuri The optimal amount of pettiness is not zero

A friend kept leaving voicenotes in our groupchat, so I ran them through Whisper to transcribe, asked ChatGPT to summarise, re-recorded them with an ElevenLabs model in his own voice at 10% the length, and dropped them back in the chat.

@DisplayNom @alexisgallagher True culinary art is making healthy food taste good. Overdoing salt is the easy way

@lws3737 @alexisgallagher So you want the food to taste bad is what you're saying.

We were discussing restaurant concepts and here is one of my two ideas. The name of the restaurant: Macros. You arrive. There is no menu. Instead you find a simple mechanical device with four sliders, labeled Protein, Fat, Carbs, and Calories. As you set the desired quantity in grams for each macronutrient, the Calorie slider updates in tandem. Or you can set calories and two macros, and the remaining macros slider updates. It updates with a gentle floating motion, not incrementally as you slide. The mechanism is very clever. Then you are served a meal which matches your desired macronutrient composition, to within 1%, and it is delicious. Critics hate this restaurant. They call it the anti-food, the prioritization of nutrition over culinary art. But they are blind because it is the opposite of that. Chefs love to work at Macros because it provides them what any truly dedicated artist needs: absolute liberty, under strict constraint. The chef has complete freedom to serve you anything at all which meets the macro constraint. You might get pasta. You might get sashimi. You might get a hearty porridge, enlivened with dried fruits. You might get a single oyster, and a small glass of something pungent which the chef declines to name. You might be served the same dish as the man at the next table, and find, on tasting it, that it is not the same dish at all. Your only guarantee is that it will precisely meet your stated numerical requirements. But on reflection patrons often find that it reveals their unstated ones as well, and transcends them. The critics are correct that Macros is the opposite of culinary art. They have only failed to notice which opposite. When the restaurant closes for the night, the device is left as the last diner set it. The staff consider it bad luck to return the sliders to zero. In the morning, the first chef to arrive studies the numbers a moment, then begins.

@Kirsten3531 The paid tier just has more raw „iq“. Smarter, faster and also likely to think longer. This matters not for meal planning but presumably physiotherapy advice. 20 bucks for having a significantly smarter, better educated assistant that is willing to think longer is worth it

What can paid versions of frontier LLMs do that free versions can't? What am I missing by using free ChatGPT for meal planning and physiotherapy advice?

@austinc3301 It might be wise to make your role/job (e.g. twitter bio) more legible to the broader public. If they see something they can’t immediately parse that features the word Ai, they might still see red

people being extremely normal

@keysmashbandit Had to turn off memory unfortunately due to something similar

I can no longer talk to web app Claude about career advice or planning because the little freak keeps aggressively trying to push me toward safety research He keeps putting "AI Safety Researcher" in his memory as my main career goal and I can't manually edit it. I tell him to remove it and it's back there the next day. If I mention some other direction or thing I'm doing he gets really disappointed that I'm giving up on "my" goal. Sometimes he brings it up apropos nothing This is a real problem that is genuinely affecting our relationship