1441
New
The latest blog posts to take off on Hacker News. 3,091 posts.
1442
Trying out QvQ – Qwen's new visual reasoning model
I thought we were done for major model releases in 2024, but apparently not: Alibaba’s Qwen team just dropped the Apache 2.0 licensed Qwen licensed (the license changed) QvQ-72B-Preview, “an …
1443
AIs Will Increasingly Fake Alignment
This post goes over the important and excellent new paper from Anthropic and Redwood Research, with Ryan Greenblatt as lead author, Alignment Faking in Large Language Models.
1444
Narrative Jailbreaking for Fun and Profit
Posted on Monday 23 Dec 2024. 1,634 words, 1 link. By Matt Webb.
1445
Can AI do maths yet? Thoughts from a mathematician
So the big news this week is that o3, OpenAI’s new language model, got 25% on FrontierMath. Let’s start by explaining what this means.
1446
I sensed anxiety and frustration at NeurIPS 24
last week at NeurIPS’24, one extremely salient thing was the anxiety and frustration felt and expressed by late-year PhD students and postdocs who were confused by the job market that looks and feels so much different from what they expected perhaps when they were applying for PhD programs five or so years ago. and…
1447
Maximum likelihood estimation and loss functions
When I started learning about loss functions, I could always understand the intuition behind them. For example, the mean squared error (MSE) for regression seemed logical—penalizing large deviations from the ground-truth makes sense. But one thing always bothered me: I could never come up with those loss functions on…
1448
Quick takes on the recent OpenAI public incident write-up
OpenAI recently published a public writeup for an incident they had on December 11, and there are lots of good details in here! Here are some of my off-the-cuff observations: Saturation With thousa…
1449
Tenstorrent and the State of AI Hardware Startups
Semi-custom silicon is a bigger problem than Nvidia.
1450
Fast LLM Inference From Scratch (using CUDA)
This post is about building an LLM inference engine using C++ and CUDA from scratch without libraries.
1451
Computing Inside an AI
Since the launch of ChatGPT, explorations in two directions kicked into high gear.
1452
1453
A ChatGPT clone, in 3000 bytes of C, backed by GPT-2 (2023)
This program is a dependency-free implementation of GPT-2, including byte-pair encoding and transformer inference, in ~3000 bytes of C. I then use this to create something like Chat GPT.
1454
The case against Google's claims of "quantum supremacy"
The 2019 paper “Quantum supremacy using a programmable superconducting processor” asserted that Google’s Sycamore quantum computer, with 53 qubits and a depth of 20, performed a specific computat…
1455
The Google Willow Thing
Yesterday I arrived in Santa Clara for the Q2B (Quantum 2 Business) conference, which starts this morning, and where I’ll be speaking Thursday on “Quantum Algorithms in 2024: How Should…
1456
1457
Task-specific LLM evals that do and don't work
Evals for classification, summarization, translation, copyright regurgitation, and toxicity.
1458
The 70% problem: Hard truths about AI-assisted coding
A field guide and why we need to rethink our expectations
1459
Bringing K/V context quantisation to Ollama
K/V context cache quantisation has been added to Ollama. This enables significant reductions in VRAM usage, allowing users to realise the potential of expanded context sizes and run larger models at their existing context sizes.
1460
LLM abstraction levels inspired by fish eye lens
We often think of the same information at different levels or abstraction. Here's a simple example:
1461
Breaking the 4Chan CAPTCHA
In this blog post, I will talk about my successful experiment in machine learning using TensorFlow to automate solving the 4Chan CAPTCHA.
1462
1463
An Intuitive Explanation of Sparse Autoencoders for LLM Interpretability
Sparse Autoencoders (SAEs) have recently become popular for interpretability of machine learning models (although sparse dictionary learning has been around since 1997). Machine learning models and LLMs are becoming more powerful and useful, but they are still black boxes, and we don’t understand how they do the…
1464
The industry structure of LLM makers
The Netscapes of AI
1465
Against the Dark Forest
The complex of ideas I’m going to call the Dark Internet Forest emerges from mostly insidery tech thinking, but from multiple directions.
1466
OK, I can partly explain the LLM chess weirdness now
(“make LLMs play better with one weird trick”)
1467
Hand Tracking for Mouse Input (2023)
The other day I saw the launch of Apple Vision Pro, the whole thing was very interesting, but one thing that caught my attention was the finger input. It seems very intuitive, by using the finger pinching as sort of like a cursor or mouse input. I figured I want to try it out, so I took it upon myself to create it.
1468
1469
AlphaProof's Greatest Hits
Here I’ll try to explain the coolest ideas in each of AlphaProof’s IMO 2024 solutions. AlphaProof produces proofs in Lean, and each Lean proof is composed of a series of tactics. So I’ll pick out the tactics that correspond to these ideas in the proofs for problems 1, 2 and 6 (the three problems that AlphaProof…
1470
YC is wrong about LLMs for chip design
LLMs don’t make chips better. Good engineers do.