Honestly I'm surprised we even got that, and I think we might not have except that other researchers independently figured out synthetic captions around the same time.
currentscurrents
joined 1 year ago
I'm interested to see how model-based RL could work for reasoning.
Instead of training a model to predict data and then fine-tuning it with RL to be a chatbot, you use RL as the primary training objective and train the data model as a side effect. This lets your pretraining objective be the actual objective you care about, so your reward function could punish issues like hallucination or prompt injection.
I haven't seen any papers using model-based RL for language modeling yet, but it's starting to work well in more traditional RL domains like game-playing. (dreamerv3, TD-MPC2)
[D] Skill Creep in ML/DL Roles - is the field getting not just more competitive, but more difficult?
I do know what BERT is and how RNNs differ from transformers. What buzzwords should I be putting on my resume to get these interviews?
This seems pretty sketchy. Lots of angry words, but few details.
Most of this has nothing to do with sexual abuse, but is rather family drama over their dad's will. She says that Sam and his lawyer were able to delay or withhold money she was supposed to inherit, but doesn't really provide details. There's not enough information here to judge the accuracy of her claims.
The sexual abuse allegedly happened when she was 4 and he was 13, but she didn't remember it until some kind of flashback in 2020.
Technological abuse - {I experienced} Shadowbanning across all platforms except onlyfans and pornhub."
Sam is certainly well-connected within the tech industry, but I'm doubtful that he could get that many platforms to ban her. Also, her posts seem to be up and visible right now.
One key difference is that they are not trained with end-to-end optimization but rather a hand crafted learning rule. This rule has strong inductive biases that work well for small datasets with pre-extracted features, like tabular data.
Their big disadvantage (and this applies to logical/symbolic approaches in general) is that they don't work well with raw data. Even easy datasets like CIFAR10. The world is too messy for perfect logical rules; neural networks are able to capture this complexity, but simpler models struggle to.
statistical
Note that learning is a fundamentally statistical process, so Tsetlin Machines are also statistics based.
They definitely can go deeper - with skip connections and normalization you can propagate gradients through any depth of architecture.
Adding more layers isn't free though, it requires more parameters and thus more compute. There's an optimal depth-to-width ratio for a given parameter count.
All the real datasets we care about are "special" in that they are the output of complex systems. We don't actually want to model the data; we want to model the underlying system.
Many of these systems are as computationally as complex as programs, and so can only be perfectly modeled by another program. This means that modeling can be viewed as the process of analyzing the output of a program to create another program that emulates it.
Given infinite compute, I would brute force search the space of all programs, and find the shortest one that matches the original system for all inputs and outputs. Lacking infinite compute, I would use an optimization algorithm like gradient descent to find an approximate solution.
You can see the link to Kolmogorov Complexity here, and why modeling is said to be equivalent to compression.
Jascha Sohl-Dickstein invented diffusion models, he's a pretty big name in the field.
anyone who doesn’t have copious amounts of programming experience should not be involved in academia related to machine learning
ML research is very heavy on math and statistics. In general, the skills necessary for ML are not very similar to the skills necessary for programming.
Learned optimizers look promising - training a neural network to train neural networks.
Unfortunately they're hard to train and nobody has gotten them to really work yet. The two main approaches are meta-training or reinforcement learning, but meta-training is very expensive and RL has all the usual pitfalls of RL.
Could this be why Google search (which has used a Bert-based retrieval system for the last couple years) ranks LLM-generated SEO fluff so highly?
As much as I agree misinformation is a problem, automatic misinformation detection systems seem only one software update away from automatic censorship systems.
Whoever built the system can decide what truth is, and what ideas are allowed to be discussed. And hey, maybe you agree with them. But what if somebody like Elon Musk or Peter Thiel buys out your social media company? I certainly don't agree with their ideas of reality.
I don't think there can be a systemic solution to misinformation that doesn't compromise free speech. It's the frying pan or the fire.
TL;DR they finetuned LLaMA on a bunch of Chinese scientific papers. As a result, it's pretty good at answering questions about science. Especially in Chinese.