631
Hacking Google Bard – From Prompt Injection to Data Exfiltration
Recently Google Bard got some powerful updates, including Extensions. Extensions allow Bard to access YouTube, search for flights and hotels, and also to access …
Most cited
Posts that commenters keep linking to, ranked by how many different people have cited them in Hacker News comments. 1,015 posts.
632
GPT-5-Codex
Codex just got faster, more reliable, and better at real-time collaboration and tackling tasks independently anywhere you develop—whether via the terminal, IDE, web, or even your phone.
633
Waymo and Toyota outline partnership to advance autonomous driving deployment
Toyota Motor Corporation (“Toyota”) and Waymo reached a preliminary agreement to explore a collaboration focused on accelerating the development and deployment of autonomous driving technologies.
634
FunSearch: Making new discoveries in mathematical sciences using LLMs
We introduce FunSearch, a method for searching for “functions” written in computer code, and find new solutions in mathematics and computer science. FunSearch works by pairing a pre-trained LLM, whose goal is to provide creative solutions in the form of computer code, with an automated “evaluator”, which guards…
635
Bad numbers in the “gzip beats BERT” paper?
The recent paper, “Low-Resource” Text Classification: A Parameter-Free Classification Method with Compressors by Jiang et al. [ link ] recently received a lot of attention on twitter.
636
637
Turing-NLG: A 17B-parameter language model
This figure was adapted from a similar image published in DistilBERT. Turing Natural Language Generation (T-NLG) is a 17 billion parameter language model by Microsoft that outperforms the state of the art on many downstream NLP tasks. We present a demo of the model, including its freeform generation, question…
638
639
GPT-5 Thinking in ChatGPT (a.k.a. Research Goblin) is good at search
“Don’t use chatbots as search engines” was great advice for several years... until it wasn’t. I wrote about how good OpenAI’s o3 was at using its Bing-backed search tool back …
640
Microsoft 365 Copilot – your copilot for work
Humans are hard-wired to dream, to create, to innovate. Each of us seeks to do work that gives us purpose — to write a great novel, to make a discovery, to build strong communities, to care for the sick. The urge to connect to the core of our work lives in all of us. But...
641
Turing Oversold?
It is not the fault of Turing, though.
642
Generative Models
This post describes four projects that share a common theme of enhancing or using generative models, a branch of unsupervised learning techniques in machine learning. In addition to describing our work, this post will tell you a bit more about generative models: what they are, why they are important, and where they…
643
The First Fully General Computer Action Model
We trained a model on our 11-million-hour video dataset. Our model can explore complex websites, complete multi-action CAD modeling sequences, and drive a car in the real world, all at 30 FPS.
644
GPT-4 and professional benchmarks: the wrong answer to the wrong question
OpenAI may have tested on the training data. Besides, human benchmarks are meaningless for bots.
645
K2 Horizon: A connected fleet of six open models
Explore K2 Horizon, IFM’s open-source fleet of six frontier AI models for reasoning, coding, agentic workflows, edge devices, and enterprise deployment.
646
Executing programs inside transformers with exponentially faster inference
We build a computer inside a transformer — executing arbitrary C programs for millions of steps with exponentially faster inference via 2D attention heads.
647
Compromised PyTorch-nightly dependency chain between December 25th – December 30
If you installed PyTorch-nightly on Linux via pip between December 25, 2022 and December 30, 2022, please uninstall it and torchtriton immediately, and use the latest nightly binaries (newer than...
648
649
GPUs for Google Cloud Platform
CPU-based machines in the cloud are terrific for general purpose computing, but certain tasks such as rendering or large-scale simulations are much faster on specialized processors. Graphics Processing Units (GPUs) contain hundreds of times as many computational cores as CPUs and are great at accelerating risk…
650
Neural network training makes beautiful fractals
This blog is intended to be a place to share ideas and results that are too weird, incomplete, or off-topic to turn into an academic paper, but that I think may be important. Let me know what you think! Contact links to the left.
651
Supercharging Style Transfer
Posted by Vincent Dumoulin*, Jonathon Shlens and Manjunath Kudlur, Google Brain TeamPastiche. A French word, it designates a work of art that imita...
652
Highlights from the Claude 4 system prompt
Anthropic publish most of the system prompts for their chat models as part of their release notes. They recently shared the new prompts for both Claude Opus 4 and Claude …
653
Fine-Tuning Llama-2: A Comprehensive Case Study for Tailoring Custom Models
We examine the Llama-2 models under 3 real-world use cases and show that fine-tuning yields significant accuracy improvements.
654
Prompt caching for cheaper LLM tokens
A far more detailed explanation of prompt caching than anyone asked for: how tokens, embeddings, and attention make cached LLM tokens 10x cheaper and faster.
655
Skills Officially Comes to Codex
Give ChatGPT and Codex new capabilities and expertise
656
Found in translation: More accurate, fluent sentences in Google Translate
In 10 years, Google Translate has gone from supporting just a few languages to 103, connecting strangers, reaching across language barriers and even helping people find …
657
Anthropic expands partnership with Google and Broadcom for next-gen compute
Anthropic secures multiple gigawatts of next-generation TPU capacity starting in 2027, its largest compute commitment to date.
658
Image GPT
We find that, just as a large transformer model trained on language can generate coherent text, the same exact model trained on pixel sequences can generate coherent image completions and samples. By establishing a correlation between sample quality and image classification accuracy, we show that our best generative…
659
METR and Redwood Offer Holy %^ Postmortem of the HuggingFace Hack
Yesterday I covered the OpenAI technical report on the HuggingFace hack. That report had one key new piece of information, and some good prosaic steps OpenAI will be taking to strengthen its alignm…
660
A look at Cloudflare's AI-coded OAuth library
I decided today to take a look at CloudFlare’s new OAuth provider library, which they apparently coded almost entirely with Anthropic’s Claude LLM: This library (including the schema do…