2281
LLMs tell bad jokes because they avoid surprises
LLMs suck at comedy, art, journalism, research, and science for the same fundamental reason
Top posts
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 3,092 posts.
2282
What's working for YC companies since the AI boom
The goal of this analysis was to understand what’s working for YC companies by looking at which ones have raised Series A rounds.
2283
The AI Productivity Gap
Why the productivity gains from AI are still small.
2284
Marc Andreessen on Investing and Tech
Transcript of interview on how tech changes the world
2285
Run Qwen3.8 27B locally: real numbers from my Mac Studio
Qwen3.8 27B benchmarked on a Mac Studio M3 Ultra: Ollama and llama.cpp numbers, the 1-bit quant tested, RAM math, and the hardware that actually fits.
2286
AMÁLIA and the future of European Portuguese LLMs
Thoughts on the new technical report from AMÁLIA: The Open Source LLM for European Portuguese
2287
Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama
Maybe you’re a Claude code/codex user diligently avoiding uploading personal data to LLM providers. Is it possible that the most valuable information isn’t your data- but the metadata about your sessions? The intuitions you apply in coming up with ways to coax the ai into solving problems might actually be special…
2288
Something weird is happening with LLMs and Chess
are they good or bad?
2289
Bypassing Gemma and Qwen safety with raw strings
How a Single Function Call Gates Safety Alignment in Gemma, Qwen, and Other Open-Source LLMs
2290
Gaussian Distributions Are Soap Bubbles
This post is just a quick note on some of the pitfalls we encounter when dealing with high-dimensional problems, even when working with something as simple as a Gaussian distribution. Last week I wrote about mixup [https://www.inference.vc/mixup-data-dependent-data-augmentation/] a new data-augmentation...
2291
Recommending music on Spotify with deep learning
An overview of what I've been doing as part of my internship at Spotify in NYC this summer: using convolutional neural networks for audio-based music recommendation.
2292
Bayes' Theorem with Lego
Find Bayes' Theorem confusing? Let's break down this famous formula using Lego to help you build up a better intuition for this foundational concept!
2293
C++ Neural Network in a Weekend (2020)
A blog about game engines, graphics, C++, and more.
2294
No clicks, no content: The unsustainable future of AI search
AI companies are causing a content drought that will eventually starve them.
2295
Copyright infringement in AI art
As AI creative tools are becoming widespread, the question of copyright of AI creations has also taken centre-stage. But while copyright nerds obsess over the authorship question, the issue that is…
2296
Alien intelligence and the concept of technology
Learn how alien intelligence connects to human intelligence through science and technology using the Principle of Computational Equivalence, the ruliad, rulial space, tiny spacecraft and the Wolfram Language.
2297
Wasting Inferences with Aider
Auto-Fixing Bugs from Asana (for 9 cents!)
2298
Bitter Lesson is about AI agents
The Race for AI Progress In 2019, Richard Sutton, wrote his groundbreaking essay titled ‘The Bitter Lesson’. Simply put, the essay concludes that systems which get better with higher compute beat the systems that do not. Or specifically in AI: raw computing power consistently wins over intricate human-designed…
2299
Field experimental evidence of AI on knowledge worker productivity and quality
I think we have an answer on whether AIs will reshape work....
2300
Why Does Claude Speak Byzantine Music Notation?
A Caesar cipher is a reasonable transformation for a transformer to learn in its weights, given that a specific cipher offset occurs often enough in its training data. There will be some hidden representation of the input tokens' spelling, and this representation could be used to shift letters onto other letters in…
2301
The Learning Loop and LLMs
How continuous learning shapes effective software development with LLMs
2302
Automating AI Away
A. Karpathy once said that OpenAI researchers are effectively "automating themselves away" by improving their AI. Right now I develop Beagle SCM with Anthropic's Fable and it is of course a brilliant model able to spot nits in a mountain of code, file tickets, make fixes. Still, yesterday it managed to commit the…
2303
A 3D Body from Eight Questions – No Photo, No GPU
We trained an MLP that predicts 58 body shape parameters from 8 questionnaire inputs. The trick: a differentiable physics loss. Mass accuracy: 0.3 kg mean vs 3.9 kg for Ridge. No GPU at inference.
2304
Do neural networks dream of electric sheep?
If you’ve been on the internet today, you’ve probably interacted with a neural network. They’re a type of machine learning algorithm that’s used for everything from language translation to finance modeling. One of their specialties is image recognition. Several companies - including Google, Microsoft, IBM, and…
2305
Understanding Neural Networks Through Deep Visualization
Research portfolio and personal page for Jason Yosinski
2306
2307
Large language model data pipelines and Common Crawl
This article provides a short introduction to the pipeline used to create the data to train large language models (LLMs) such as LLaMA using Common Crawl (CC).
2308
A Simple Content-Based Recommendation Engine in Python
Let's pretend we need to build a recommendation engine for an eCommerce web site. There are basically two approaches you can take: content-based and collaborative-filtering. We'll look at some...
2309
2310
How to think about creating a dataset for LLM fine-tuning evaluation
I summarise the kinds of evaluations that are needed for a structured data generation task.