Full Width [alt+shift+f] Shortcuts [alt+shift+k]
Sign Up [alt+shift+s] Log In [alt+shift+l]
Articles [alt+h] Reading List [alt+r] [alt+s]

The Berkeley Artificial Intelligence Research Blog

Visit Blog

Sort By

Popular [alt+1] Recent [alt+2]

Change View

List View [alt+w] Expanded [alt+e]
The Berkeley...
Virtual Personas for Language Models via an Anthology of Backstories Anthology, a method for conditioning LLMs to representative, consistent, and diverse virtual...
3 months ago
52

Virtual Personas for Language Models via an Anthology of Backstories

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

3 months ago
Anthology, a method for conditioning LLMs to representative, consistent, and diverse virtual personas by generating and utilizing naturalistic backstories with rich details of individual values and experience. --> We introduce Anthology, a method for conditioning LLMs to...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Linguistic Bias in ChatGPT: Language Models Reinforce Dialect Discrimination Sample language model responses to different varieties of English and native speaker...
5 months ago
87

Linguistic Bias in ChatGPT: Language Models Reinforce Dialect Discrimination

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

5 months ago
Sample language model responses to different varieties of English and native speaker reactions. ChatGPT does amazingly well at communicating with people in English. But whose English? Only 15% of ChatGPT users are from the US, where Standard American English is the default. But...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
How to Evaluate Jailbreak Methods: A Case Study with the StrongREJECT Benchmark When we began studying jailbreak evaluations, we found a fascinating paper claiming that you could...
5 months ago
81

How to Evaluate Jailbreak Methods: A Case Study with the StrongREJECT Benchmark

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

5 months ago
When we began studying jailbreak evaluations, we found a fascinating paper claiming that you could jailbreak frontier LLMs simply by translating forbidden prompts into obscure languages. Excited by this result, we attempted to reproduce it and found something unexpected. The...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Are We Ready for Multi-Image Reasoning? Launching VHs: The Visual Haystacks Benchmark! Humans excel at processing vast arrays of visual information, a skill that is crucial for achieving...
7 months ago
93

Are We Ready for Multi-Image Reasoning? Launching VHs: The Visual Haystacks Benchmark!

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

7 months ago
Humans excel at processing vast arrays of visual information, a skill that is crucial for achieving artificial general intelligence (AGI). Over the decades, AI researchers have developed Visual Question Answering (VQA) systems to interpret scenes within single images and answer...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
TinyAgent: Function Calling at the Edge The ability of LLMs to execute commands through plain language (e.g. English) has enabled agentic...
8 months ago
94

TinyAgent: Function Calling at the Edge

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

8 months ago
The ability of LLMs to execute commands through plain language (e.g. English) has enabled agentic systems that can complete a user query by orchestrating the right set of tools (e.g. ToolFormer, Gorilla). This, along with the recent multi-modal efforts such as the GPT-4o or...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Modeling Extremely Large Images with $x$T As computer vision researchers, we believe that every pixel can tell a story. However, there seems...
11 months ago
120

Modeling Extremely Large Images with $x$T

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

11 months ago
As computer vision researchers, we believe that every pixel can tell a story. However, there seems to be a writer’s block settling into the field when it comes to dealing with large images. Large images are no longer rare—the cameras we carry in our pockets and those orbiting our...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
2024 BAIR Graduate Directory Every year, the Berkeley Artificial Intelligence Research (BAIR) Lab graduates some of the most...
11 months ago
93

2024 BAIR Graduate Directory

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

11 months ago
Every year, the Berkeley Artificial Intelligence Research (BAIR) Lab graduates some of the most talented and innovative minds in artificial intelligence and machine learning. Our Ph.D. graduates have each expanded the frontiers of AI research and are now ready to embark on new...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
The Shift from Models to Compound AI Systems AI caught everyone’s attention in 2023 with Large Language Models (LLMs) that can be instructed to...
a year ago
103

The Shift from Models to Compound AI Systems

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
AI caught everyone’s attention in 2023 with Large Language Models (LLMs) that can be instructed to perform general tasks, such as translation or coding, just by prompting. This naturally led to an intense focus on models as the primary ingredient in AI application development,...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Ghostbuster: Detecting Text Ghostwritten by Large Language Models The structure of Ghostbuster, our new state-of-the-art method for detecting AI-generated...
a year ago
94

Ghostbuster: Detecting Text Ghostwritten by Large Language Models

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
The structure of Ghostbuster, our new state-of-the-art method for detecting AI-generated text. Large language models like ChatGPT write impressively well—so well, in fact, that they’ve become a problem. Students have begun using these models to ghostwrite assignments, leading...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Asymmetric Certified Robustness via Feature-Convex Neural Networks Asymmetric Certified Robustness via Feature-Convex Neural Networks TLDR: We propose the asymmetric...
a year ago
75

Asymmetric Certified Robustness via Feature-Convex Neural Networks

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
Asymmetric Certified Robustness via Feature-Convex Neural Networks TLDR: We propose the asymmetric certified robustness problem, which requires certified robustness for only one class and reflects real-world adversarial scenarios. This focused setting allows us to introduce...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Goal Representations for Instruction Following Goal Representations for Instruction Following Figure title. Figure caption. This image is...
a year ago
101

Goal Representations for Instruction Following

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
Goal Representations for Instruction Following Figure title. Figure caption. This image is centered and set to 50% page width. --> A longstanding goal of the field of robot learning has been to create generalist agents that can perform tasks for humans. Natural language has...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Rethinking the Role of PPO in RLHF Rethinking the Role of PPO in RLHF TL;DR: In RLHF, there’s tension between the reward learning...
a year ago
133

Rethinking the Role of PPO in RLHF

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
Rethinking the Role of PPO in RLHF TL;DR: In RLHF, there’s tension between the reward learning phase, which uses human preference in the form of comparisons, and the RL fine-tuning phase, which optimizes a single, non-comparative reward. What if we performed RL in a comparative...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Training Diffusion Models with <br> Reinforcement Learning function reveal() { const replay = document.querySelector('.ddpo-replay'); ...
a year ago
70

Training Diffusion Models with <br> Reinforcement Learning

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
function reveal() { const replay = document.querySelector('.ddpo-replay'); replay.style.display = 'flex'; } window.onload = () => { const replay = document.querySelector('.ddpo-replay'); replay.addEventListener('click', () => { ...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
On the Stepwise Nature of <br> Self-Supervised Learning Figure 1: stepwise behavior in self-supervised learning. When training common SSL algorithms, we...
a year ago
85

On the Stepwise Nature of <br> Self-Supervised Learning

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
Figure 1: stepwise behavior in self-supervised learning. When training common SSL algorithms, we find that the loss descends in a stepwise fashion (top left) and the learned embeddings iteratively increase in dimensionality (bottom left). Direct visualization of embeddings...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Generating 3D Molecular Conformers via Equivariant Coarse-Graining and Aggregated Attention --> Figure 1: CoarsenConf architecture. (I) The encoder $q_\phi(z| X, \mathcal{R})$ takes the...
a year ago
71

Generating 3D Molecular Conformers via Equivariant Coarse-Graining and Aggregated Attention

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
--> Figure 1: CoarsenConf architecture. (I) The encoder $q_\phi(z| X, \mathcal{R})$ takes the fine-grained (FG) ground truth conformer $X$, RDKit approximate conformer $\mathcal{R}$ , and coarse-grained (CG) conformer $\mathcal{C}$ as inputs (derived from $X$ and a predefined...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
GPT-4 + Stable-Diffusion = ?: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with... TL;DR: Text Prompt -> LLM -> Intermediate Representation (such as an image layout) -> Stable...
a year ago
85

GPT-4 + Stable-Diffusion = ?: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
TL;DR: Text Prompt -> LLM -> Intermediate Representation (such as an image layout) -> Stable Diffusion -> Image. Recent advancements in text-to-image generation with diffusion models have yielded remarkable results synthesizing highly realistic and diverse images. However,...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Interactive Fleet Learning Figure 1: “Interactive Fleet Learning” (IFL) refers to robot fleets in industry and academia that...
a year ago
69

Interactive Fleet Learning

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
Figure 1: “Interactive Fleet Learning” (IFL) refers to robot fleets in industry and academia that fall back on human teleoperators when necessary and continually learn from them over time. In the last few years we have seen an exciting development in robotics and artificial...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Koala: A Dialogue Model for Academic Research In this post, we introduce Koala, a chatbot trained by fine-tuning Meta’s LLaMA on dialogue data...
a year ago
68

Koala: A Dialogue Model for Academic Research

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

a year ago
In this post, we introduce Koala, a chatbot trained by fine-tuning Meta’s LLaMA on dialogue data gathered from the web. We describe the dataset curation and training process of our model, and also present the results of a user study that compares our model to ChatGPT and...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Fully Autonomous Real-World Reinforcement Learning with Applications to Mobile Manipulation Reinforcement learning provides a conceptual framework for autonomous agents to learn from...
over a year ago
54

Fully Autonomous Real-World Reinforcement Learning with Applications to Mobile Manipulation

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
Reinforcement learning provides a conceptual framework for autonomous agents to learn from experience, analogously to how one might train a pet with treats. But practical applications of reinforcement learning are often far from natural: instead of using RL to learn through trial...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Keeping Learning-Based Control Safe by Regulating Distributional Shift To regulate the distribution shift experience by learning-based controllers, we seek a mechanism for...
over a year ago
59

Keeping Learning-Based Control Safe by Regulating Distributional Shift

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
To regulate the distribution shift experience by learning-based controllers, we seek a mechanism for constraining the agent to regions of high data density throughout its trajectory (left). Here, we present an approach which achieves this goal by combining features of density...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Reverse engineering the NTK: towards first-principles architecture design Deep neural networks have enabled technological wonders ranging from voice recognition to machine...
over a year ago
44

Reverse engineering the NTK: towards first-principles architecture design

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
Deep neural networks have enabled technological wonders ranging from voice recognition to machine transition to protein engineering, but their design and application is nonetheless notoriously unprincipled. The development of tools and methods to guide this process is one of the...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Why do Policy Gradient Methods work so well in Cooperative MARL? Evidence from Policy Representation In cooperative multi-agent reinforcement learning (MARL), due to its on-policy nature, policy...
over a year ago
49

Why do Policy Gradient Methods work so well in Cooperative MARL? Evidence from Policy Representation

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
In cooperative multi-agent reinforcement learning (MARL), due to its on-policy nature, policy gradient (PG) methods are typically believed to be less sample efficient than value decomposition (VD) methods, which are off-policy. However, some recent empirical studies demonstrate...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
FIGS: Attaining XGBoost-level performance with the interpretability and speed of CART FIGS (Fast Interpretable Greedy-tree Sums): A method for building interpretable models by...
over a year ago
47

FIGS: Attaining XGBoost-level performance with the interpretability and speed of CART

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
FIGS (Fast Interpretable Greedy-tree Sums): A method for building interpretable models by simultaneously growing an ensemble of decision trees in competition with one another. Recent machine-learning advances have led to increasingly complex predictive models, often at the cost...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
The Berkeley Crossword Solver We recently published the Berkeley Crossword Solver (BCS), the current state of the art for solving...
over a year ago
49

The Berkeley Crossword Solver

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
We recently published the Berkeley Crossword Solver (BCS), the current state of the art for solving American-style crossword puzzles. The BCS combines neural question answering and probabilistic inference to achieve near-perfect performance on most American-style crossword...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Rethinking Human-in-the-Loop for Artificial Augmented Intelligence How do we build and evaluate an AI system for real-world applications? In most AI research, the...
over a year ago
64

Rethinking Human-in-the-Loop for Artificial Augmented Intelligence

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
How do we build and evaluate an AI system for real-world applications? In most AI research, the evaluation of AI methods involves a training-validation-testing process. The experiments usually stop when the models have good testing performance on the reported datasets because...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]
The Berkeley...
Designing Societally Beneficial Reinforcement Learning Systems Deep reinforcement learning (DRL) is transitioning from a research field focused on game playing to...
over a year ago
67

Designing Societally Beneficial Reinforcement Learning Systems

from The Berkeley Artificial Intelligence Research Blog [alt+shift+b] in AI

over a year ago
Deep reinforcement learning (DRL) is transitioning from a research field focused on game playing to a technology with real-world applications. Notable examples include DeepMind’s work on controlling a nuclear reactor or on improving Youtube video compression, or Tesla attempting...
Direct Link [→] Remove from reading list Add to reading list [alt+a]
upvote [alt+ctrl+↑] downvote [alt+ctrl+↓]

The best way to read the internet.
Made for nerds, by a nerd.

The best way to read the internet

© 2023 Bored Reading. All rights reserved.

Terms Privacy Design by Kumail

bored reading

v. 229.328 compliant with
earth/mars-x8292 communication
protocol 29xjw899992

When Google killed Google Reader, the bloggosphere took a severe hit and the content quality went down because there weren't enough readers to justify the effort it takes to maintain a high quality blog.

What's left is a decaying wasteland of blogs, most abandoned, a lot are now SEO spam, and 99% are not worth reading. BoredReading is a way to read the remaining great blogs that survived the ice age.

To add your favorite blog, please suggest it here! I get emails every day which helps me maintain a high quality selection of blogs!

Avatar

Made by,
Kumail Hunaid

@kumailht on Twitter.
My website and email.

Close [esc]