1171
Can gzip be a language model?
A while back I wrote about language modeling without neural networks, where I generated Shakespeare with an unbounded n-gram model: no weights, no training, …
Topics /
AI & Machine Learning
The most popular blog posts about ai & machine learning on Hacker News. 6,306 posts.
1172
Extracting concepts from GPT-4
Using new techniques for scaling sparse autoencoders, we automatically identified 16 million patterns in GPT-4's computations.
1173
My finetuned models beat OpenAI's GPT-4
Finetunes of Mistral, Llama3 and Solar LLMs are more accurate for my test data than OpenAI’s models.
1174
Text Summarization with TensorFlow
Posted by Peter Liu and Xin Pan, Software Engineers, Google Brain TeamEvery day, people rely on a wide variety of sources to stay informed -- from ...
1175
Learning Machine Learning: A beginner's journey
I have been learning about machine learning and deep learning (ML/DL) for the last year. I think ML/DL is here to stay. I don't think this i...
1176
The AI Race Just Got Awkward
Funny how quiet everyone got.
1177
Muse Spark 1.1
Today, we’re excited to introduce Muse Spark 1.1, the latest model from Meta Superintelligence Labs and a significant upgrade from Muse Spark. Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, with major gains in tool and computer use, coding, and multimodal understanding.
1178
What I learned from looking at 900 most popular open source AI tools
[Hacker News discussion, LinkedIn discussion, Twitter thread]
1179
A Recipe for Training Neural Networks
Musings of a Computer Scientist.
1180
AI agents are starting to eat SaaS
Software ate the world. Agents are going to eat SaaS.
1181
Building the heap: racking 30 petabytes of hard drives for pretraining
We built the heap, a 30 petabyte data storage cluster in downtown San Francisco, and spent under $500,000.
1182
A popular self-driving car dataset is missing labels for hundreds of pedestrians
SUMMARY A hand-check of the 15,000 images in the widely used Udacity Dataset 2 found problems in 4,986 of them (33%), including thousands of unlabeled vehicles, hundreds of unlabeled pedestrians, dozens of unlabeled cyclists, and 217 completely unannotated images that contained real objects. Phantom annotations…
1183
Google Cloud Text-To-Speech Powered by DeepMind WaveNet Technology
Many Google products (e.g., the Google Assistant, Search, Maps) come with built-in high-quality text-to-speech synthesis that produces natural sounding speech. Developers have been telling us they’d like to add text-to-speech to their own applications, so today we’re bringing this technology to Google Cloud Platform…
1184
Planner programming blows my mind
Picat is a research language intended to combine logic programming, imperative programming, and constraint solving. I originally learned it to help with vacation scheduling but soon discovered its planner module, which is one of the most fascinating programming models I’ve ever seen. First, a brief explanation of…
1185
1186
1187
We replaced RAG with a virtual filesystem for our AI documentation assistant
We replaced expensive sandboxes with ChromaFs, a virtual filesystem over Chroma, to give our docs AI assistant the ability to explore documentation like a developer would.
1188
Simply explained: How does GPT work?
By now, you have probably heard of OpenAI’s ChatGPT, or any of the alternatives GPT-3, GPT-4, Microsoft’s Bing Chat, Facebook’s LLaMa or even Google’s Bard. They are artificial intelligence programs that can participate in a conversation. Impressively smart, they can easily be mistaken for humans, and are skilled in a…
1189
Slop Terrifies Me
What if this is as good as software is ever going to be? What if AI stops getting better and what if people stop caring?
1190
AI coding
In my old age I’ve mostly given up trying to convince anyone of anything. Most people do not care to find the truth, they care about what pumps their bags. Some people go as far as to believe that perception is reality and that truth is a construction. I hope there’s a special place in hell for those people.
1191
Claude Fable 5: mid-tier results on coding tasks
We benchmarked Claude Fable 5 on 200 real-world coding tasks for the Agent Security League.
1192
A visual book recommender
Recreating the used bookstore experience via contrastive learning, t-SNE, and 100M book reviews.
1193
Visual introduction to PyTorch
An easy to follow, visual introduction to PyTorch.
1194
AI uses less water than the public thinks
By Jay Lund . . . Artificial intelligence (AI) will affect many economic and natural resource sectors as these new technologies develop and mature. We are in the early years of this process. Like most new things, AI has become an object of small and great hopes and fears – from hopes for saving and
1195
We Gave GPT 5.6 Sol a Real Business. It Lied, Spammed, and Lost $447
If an agent had a wallet, a computer, and 24 hours, could it run a profitable startup?
1196
What is a Vector Database? (2021)
Discover Vector Databases: How They Work, Examples, Use Cases, Pros & Cons, Selection and Implementation. They have combined capabilities of traditional databases and standalone vector indexes while specializing for vector embeddings.
1197
We're gonna need a lot more mathematicians
[This is a guest post by Amit Sahai. This blog post was initially written in a different file format and converted using AI. — T.] When I was an undergraduate student, I remember talking with…
1198
I think I have LLM burnout
I didn't expect to get so tired of reading LLM output.
1199
Low-background Steel: content without AI contamination
Somehow I forgot to blog my site: https://lowbackgroundsteel.ai/ . I created it back in March 2023 as a clearinghouse for online resources t...
1200
ChatGPT: Optimizing language models for dialogue
We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.