541
Why LLMs still have problems with OCR
LLMs excel at text generation but fail at precise document extraction due to nondeterminism. | Pulse AI
Most cited
Posts that commenters keep linking to, ranked by how many different people have cited them in Hacker News comments. 593 posts.
542
Write Emails Faster with Smart Compose in Gmail
Powered by artificial intelligence, Smart Compose suggests complete sentences in your emails so that you can draft emails from scratch, faster.
543
Composer: Building a fast frontier model with RL
Composer is our new agent model designed for software engineering intelligence and speed.
544
How convolutional neural networks see the world
Note: this post was originally written in January 2016. It is now very outdated. Please see this example of how to visualize convnet filters for an up-to-date alternative, or check out chapter 9 of my book "Deep Learning with Python (2nd edition)".
545
Qwen-Robot Suite: A Foundation Model Suite for Physical World Intelligence
Qwen offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.
546
Fuyu-8B: A multimodal architecture for AI agents
We’re open-sourcing Fuyu-8B - a small version of the multimodal model that powers our product.
547
We gave an AI a 3 year retail lease and asked it to make a profit
We signed a 3 year lease and gave it to an AI
548
549
Project Vend: Phase Two
How Claude turned around its failing vending machine business
550
Facebook to open-source AI hardware design
Visit the post for more.
551
Understanding Convolutional Neural Networks
An interactive visualization system designed to help non-experts learn about Convolutional Neural Networks (CNNs).
552
Windsurf SWE-1: Our First Frontier Models
Introducing our first Frontier Models!
553
The Components of Agent Memory
For Claude to learn from experience, it has to decide what’s worth remembering before it knows what’s next. Here’s how “dreaming” helps it organize memory and correct mistakes.
554
A comparison to Waymo’s auto liability insurance claims at 25M miles
Understanding the safety impact of Automated Driving Systems (ADS) is crucial for their widespread adoption, yet robust real-world evaluation remains a critical area of development. By leveraging insurance industry third-party auto liability claims, we analyze the Waymo ADS performance across 25.3 million fully…
555
Towards a conversational agent that can chat about anything
Posted by Daniel Adiwardana, Senior Research Engineer, and Thang Luong, Senior Research Scientist, Google Research, Brain Team Modern conversatio...
556
Agents that imagine and plan
Imagining the consequences of your actions before you take them is a powerful tool of human cognition. When placing a glass on the edge of a table, for example, we will likely pause to consider how stable it is and whether it might fall. On the basis of that imagined consequence we might readjust the glass to prevent…
557
Spinning Up in Deep RL
Stay up to speed on the rapid advancement of AI technology and the benefits it offers to humanity.
558
Fixing Gemma Bugs
Unsloth fixing Google's open-source language model Gemma.
559
GPT-4 Turbo with Vision is a step backwards for coding
OpenAI’s GPT-4 Turbo with Vision model scores worse on aider’s code editing benchmarks than all the previous GPT-4 models. In particular, it seems much more prone to “lazy coding” than the existing GPT-4 Turbo “preview” models.
560
Open-sourcing circuit tracing tools
Our open-source tools generate attribution graphs showing how language models reach decisions.
561
Reformer, the Efficient Transformer
Posted by Nikita Kitaev, Student Researcher, UC Berkeley and Łukasz Kaiser, Research Scientist, Google Research Understanding sequential data — s...
562
Improving mathematical reasoning with process supervision
We’ve trained a model to achieve a new state-of-the-art in mathematical problem solving by rewarding each correct step of reasoning (“process supervision”) instead of simply rewarding the correct final answer (“outcome supervision”). In addition to boosting performance relative to outcome supervision, process…
563
564
Making floating point math highly efficient for AI hardware
In recent years, compute-intensive artificial intelligence tasks have prompted creation of a wide variety of custom hardware to run these powerful new systems efficiently. Deep learning models, suc…
565
Self driving cars and beyond
The story of technology is not unlike what often happens in dramatic stories and plays, where a “deus ex machina” or character drops down magically from the skies onto our world stage. But in reality, technology never comes so suddenly, or in such an isolated manner: There’s an entire ecosystem that grows up around…
566
Continuing our work to improve recommendations on YouTube
That's why we update our recommendations system all the time—we want to make sure we’re suggesting videos that people actually want to watch.
567
Llama 2 is about as factually accurate as GPT-4 for summaries and is 30X cheaper
Using Anyscale Endpoints, we compared Llama 2 7b, 13b and 70b vs. OpenAI's GPT-3.5-turbo and GPT-4 for accuracy and cost.
568
Securing our frontier model training environment
OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.
569
Prover-Verifier Games improve legibility of language model outputs
Discover how prover-verifier games improve the legibility of language model outputs, making AI solutions clearer, easier to verify, and more trustworthy for both humans and machines.
570
The Myth of AI – Jaron Lanier
That mythology, in turn, has spurred a reactionary, perpetual spasm from people who are horrified by what they hear. You'll have a figure say, "The computers will take over the Earth, but that's a good thing, because people had their chance and now we should give it to the machines." Then you'll have other people say…