5791
Which kinds of GPT startups will thrive?
How to separate the wheat from the chaff when GPT-powered startups are popping up everywhere
Top posts
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 6,335 posts.
5792
SynthID – A tool to watermark and identify content generated through AI
SynthID is a tool to watermark and identify AI-generated content, helping to foster transparency and trust in generative AI.
5793
Most users cannot identify AI bias, even in training data
When recognizing faces and emotions, artificial intelligence (AI) can be biased, like classifying white people as happier than people from other racial backgrounds. This happens because the data used to train the AI contained a disproportionate number of happy white faces, leading it to correlate race with emotional…
5794
5795
Hacking with Claude on a $27 smart watch
The Pine Time is a $27 smart watch that runs open source firmware. I’ve had one sitting in the back of a drawer for a...
5796
Leveraging AI for efficient incident response
We’re sharing how we streamline system reliability investigations using a new AI-assisted root cause analysis system. The system uses a combination of heuristic-based retrieval and large language m…
5797
How Underground Groups Use Stolen Identities and Deepfakes
The growing appearance of deepfake attacks is significantly reshaping the threat landscape. These fakes brings attacks such as business email compromise (BEC) and identity verification bypassing to new levels.
5798
We reproduced Anthropic's Mythos findings with public models
Anthropic framed Mythos and Project Glasswing as proof that frontier AI vulnerability research now needs gated access. We tested the public, patched cases with GPT-5.4 and Claude Opus 4.6 and found that the key building blocks are already accessible outside Glasswing, while reliable operationalization remains the real…
5799
SmolLM2
New from Loubna Ben Allal and her research team at Hugging Face: SmolLM2 is a family of compact language models available in three size: 135M, 360M, and 1.7B parameters. They …
5800
MTIA v1: Meta’s first-generation AI inference accelerator
In 2020, we initiated the Meta Training and Inference Accelerator (MTIA) family of chips to support our evolving AI workloads, starting with an inference accelerator ASIC for deep learning recommendation models (DLRMs).
5801
Asymmetric Quantization: Near-Lossless Retrieval with 97% Storage Reduction
Late interaction makes retrieval more precise but turns every document into many vectors. Asymmetric quantization keeps query vectors precise and stores…
5802
PaperBench
We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research.
5803
Design Patterns for Securing LLM Agents Against Prompt Injections
This new paper by 11 authors from organizations including IBM, Invariant Labs, ETH Zurich, Google and Microsoft is an excellent addition to the literature on prompt injection and LLM security. …
5804
5805
Quantifying the performance of the TPU, our first machine learning chip
We’re releasing a study that shares new details on our Tensor Processing Unit (TPU) custom chips which have been running machine learning applications in our data centers since 2015.
5806
Inversion: Fast, Reliable Structured LLMs
We create the world's fastest and most efficient AI.
5807
Real-time Drone object tracking using Python and OpenCV
After flying this past weekend (together with Gabriel and Leandro) with Gabriel's drone (which is an handmade APM 2.6 based quadcopter) in our town (Porto Alegre, Brasil), I decided to implement a tracking for objects using OpenCV and Python and check how the results would be using simple and fast methods like…
5808
Abusing vector search for texts, maps, and chess
Vector search isn't just for AI embeddings. Use HNSW for geo-spatial queries, stock covariance, chess positions, text tokens, and multi-modal recommendations with custom metrics.
5809
5810
Segment Anything Model and Friends
The Segment Anything Model (SAM) and its successors made a significant leap forward in computer vision, particularly in image and video segmentation. Along with SAM’s innovative approach to promptable segmentation, the literature rapidly evolved to address key challenges like computational efficiency and real-time…
5811
Reverse-Engineering Claude Web's MicroVM: Uncovering Anthropic's Hidden Antspace
Full stack developer, open sourceror, serial entrepreneur.
5812
Continuous batching to increase LLM inference throughput and reduce p50 latency
In this blog, we discuss continuous batching, a critical systems-level optimization that improves both throughput and latency under load for LLMs.
5813
KumoRFM: A Foundation Model for In-Context Learning on Relational Data
Introduction to Kumo Relational — the relational foundation model for predictive AI on relational data
5814
Deal or no deal? Training AI bots to negotiate
Visit the post for more.
5815
Natural Language Corpus Data: Beautiful Data
Data files are derived from the Google Web Trillion Word Corpus , as described by Thorsten Brants and Alex Franz, and distributed by the Linguistic Data Consortium.
5816
5817
5818
TensorFlow Feature Columns
Welcome to Part 2 of a blog series that introduces TensorFlow Datasets and Estimators. We're devoting this article to —a data structure describing the features that an Estimator requires for training and inference. As you'll see, feature columns are very rich, enabling you to represent a diverse range of data.
5819
Deep Learning from Scratch to GPU: CUDA and OpenCL
We generalize the network code and run it on the GPU. On an Nvidia GPU with CUDA, and on an AMD GPU with OpenCL. Even more - we mix both CUDA and OpenCL, jus...
5820