armcat

13 ms·

151.
by armcat 9mo ago
Super nice! I've been using Kokoro locally, which is 82M parameters and runs (and sounds) amazing! https://huggingface.co/hexgrad/Kokoro-82M
152.
by armcat 9mo ago
The new mental model actually is (1) skills based model, i.e. https://agentskills.io/home , and (2) where the LLM agents "see all problems as coding problems". Skills are a bunch of detailed Markdowns and correspon
153.
by armcat 9mo ago
I thought it was already well understood/researched that it's not the weights that matter, but effectively taking your sets to muscular failure. While one might think "I can do 50 reps with low weights" there is practica
154.
by armcat 10mo ago
The promise is similar to LLMs, if you pretrain on sufficiently large timeseries datasets with sufficiently large variance/characteristics, that you will be able to transfer the model to a completely different use case that exhibits so
155.
by armcat 10mo ago
I really like BAML but this post seems a little too much like a BAML funnel. Here are three methods that worked for me consistently since constrained sampling first came out: 1. Add a validation step (using a mini model) right at the beginn
156.
by armcat 10mo ago
Hi Simon! Love your work! Our of curiosity - how many pelican-cycling samples do you produce. Curious about the variance here. Thanks!
157.
by armcat 10mo ago
I agree with you (I have reviewed papers in the past), however, made-up citations are a "signal". Why would the authors do that? If they made it up, most likely they haven't really read that prior work. If they haven't,
158.
by armcat 10mo ago
These are fantastic insights! I work in legaltech space so something to keep in mind is that legal space is very sensitive to data storage and security (apart from this of course: https://alexschapiro.com/security/vulne
159.
by armcat 11mo ago
I think there is this troika of "Leadership", "Management" and "Followship". You don't have to be an engineering manager to be a leader, and just because you are a leader doesn't mean you have any &qu
160.
by armcat 11mo ago
One mistake in your README - groq throughput is actually 1000 tokens per "second" (not "minute"), for gpt-oss-20b.
161.
by armcat 1y ago
I couldn't immediately see in their graphs/tables any comparison against simple lexical/statistical based context compression, such as candidate selection of chunks using TF-IDF, word overlap etc. For most of us in the indust
162.
by armcat 1y ago
Can second Regression and Other Stories, it's freely available here: https://users.aalto.fi/~ave/ROS.pdf , and you can access additional information such as data and code (including Python and Julia ports) here: h
163.
by armcat 1y ago
One of the most interesting mathematical aspects to me are the fact that LLMs are logit emitters. And associated with this output is uncertainty. Lot of ppl talk about networks of agents. But what you are doing is accumulating uncertainty -
164.
by armcat 1y ago
When you use ChatGPT and it executes code, i.e. when you tell it to do something with a CSV file, it seems to run in a VM with certain tools and libraries available to it, and a sandboxed disk access; no internet access though. So it's
165.
by armcat 1y ago
Therac-25 was part of the mandatory "computer ethics" course at my uni, as part of the Computer Science programme, circa early 2000s.
166.
by armcat 1y ago
Amazing work on this, beautifully put together and very useful!
167.
by armcat 1y ago
Ah got it, it looks like it's a whole bunch of things so it can also interface with ollama, and other APIs.
168.
by armcat 1y ago
I've been asking bespoke questions and the timing is >2 seconds, and slower than what I get for the same questions to ChatGPT (using gpt-4.1-mini). I am looking at their call stack and what I see: "verifyOpenAIConnection()&quot
169.
by armcat 1y ago
I've been looking at the code on their chat playground, https://chat.inceptionlabs.ai/ , and they have a helper function `const convertOpenAIMessages = (convo) => { ... }`, which also contains `models: ['gpt-3.5
170.
by armcat 1y ago
There is a "small language model", and then there is a "small LARGE language model". In late 2018, BERT (110 million params) would've been considered a "large" language model. A "small" LM would
171.
by armcat 1y ago
This was previously reported 5 months ago: https://news.ycombinator.com/item?id=42415122 (84 comments). As an aside - I am a big fan of Luke Zettlemoyer and his team at the University of Washington. They've been doing
172.
by armcat 2y ago
Terminator 2 has aged so well. Even my son who is Gen Alpha is incredibly impressed with the movie. We are both very much looking to this game as well!
173.
by armcat 2y ago
I am not too sure about shortening the CoT tokens explicitly because different problems will require different length of proof - some require half a page, whilst others will require 10 pages worth of tokens. As the graphs in the paper indic
174.
by armcat 2y ago
It feels like lot of the reasoning tokens go to waste on pure brute force approach - plugging in numbers and evaluating and comparing against the answer. "Nope, that didn't work, let's try 4 instead of 6 this time", etc.
175.
by armcat 2y ago
Yes, the authors explicitly highlighted those two points in the abstract, in terms of them being the elicitation threshold for complex reasoning, namely, an extremely complete pre-trained foundation model, and a set of extremely high qualit
176.
by armcat 2y ago
Kevin's books tend to be more foundational based on battle-tested techniques (I love his probabilistic ML book series, https://probml.github.io/pml-book/ ). GRPO is a relatively new technique introduced by the Deep
177.
by armcat 2y ago
How I see LLMs (which have roots in early word embeddings like word2vec) is not as statistical machines, but geometric machines. When you train LLMs you are essentially moving concepts around in a very high dimensional space. If we take a c
178.
by armcat 2y ago
I asked it on my local Qwen 32B distilled version, and it duly obliged, very similar to the wikipedia entry.
179.
by armcat 2y ago
Maritime law is definitely not my forte, but I think the convention was related to the tolls imposed to pass through these waters (Sweden's Gota Kanal was built to bypass this). It's also related to "innocent passage", w
180.
by armcat 2y ago
Lot of shipping lanes in the baltic are international waters, but to get into baltic, you have to pass through Danish waters.

More ›