Generations X,Y and Z are faced with a number of existential economic questions for the rest of the century. From the looming demographic crisis in developed nations to the precarious state of sovereign debt, and the escalating economic warfare between the US and China over AI and space technology, here is a sobering picture ...

David Berenstein explains how to finetune a ModernBERT model for text classification on a synthetic dataset generated from argilla's synthetic-data-generator.

In this blog post Philipp Schmid explains how to fine-tune ModernBERT, a refreshed version of BERT models, with 8192 token context length, for classifying user prompts to implement an intelligent LLM router.

6 years after the release of BERT, answer.ai introduce ModernBERT, bringing modern model optimizations to encoder-only models and representing a major Pareto improvement over older encoders.

What happens when a "normal" person tries to wrap their head around one of humanity's most elusive concepts? This article is as much about overcoming imposter syndrome as it is about the metaphysics of time.

Bubble or true technological revolution? While the path forward isn't without obstacles, the value being created by AI extends far beyond speculative investments, touching every sector of the economy and fundamentally altering the nature of work itself.

Artificial Intelligence is a divisive subject that sparks numerous debates about both its potential and its limitations. However, there's one aspect everyone can agree on: current commercial practices demonstrate that the sector is still far from maturity.

The Verge - TikTok, owned by Chinese company ByteDance, sent users in the US a push notification on Wednesday, warning that 'Congress is planning a total ban of TikTok' that would 'strip 170 million Americans of their Constitutional right to free expression.'

Tom's Hardware - Nvidia has banned running CUDA-based software on other hardware platforms using translation layers in its licensing terms, which previously wasn't included in the documentation placed on a host system during the installation process. The restriction appears to be designed to prevent initiatives like ZLUDA and...

Quantamagazine - Physicists have discovered that quantum phenomena may not be as fragile as previously thought, with certain quantum systems appearing to remain stable for eons. However, recent research suggests that these systems may not be eternal after all, challenging the foundations of thermodynamics and statistical mech...

Retell tackle the challenge of real time conversations with voice AI.

Semianalysis - Groq, an AI hardware startup, has been making waves with their impressive demos showcasing Mistral Mixtral 8x7b on their inference API. They are achieving up to 4x the throughput of other inference services while also charging less than 1/3 that of Mistral themselves. However, speed is only one part of the equa...

VIDEO | Andrej Karpathy builds from scratch the Tokenizer used in the GPT series from OpenAI, showing that a lot of weird behaviors and problems of LLMs actually trace back to tokenization and discusseing why tokenization is at fault

The Major-TOM Core-S2L2A dataset contains a global coverage of Sentinel-2 (Level 1C) patches, each of size 1,068 x 1,068 pixels. The dataset includes 2,245,886 patches and over 2.564 trillion pixels.

Nicholas Carlini introduces his new benchmark for large language models, which includes nearly 100 tests extracted from actual conversation history with various LLMs.

Minghao Shao, Boyuan Chen, Sofija Jancheska, Brendan Dolan-Gavitt, Siddharth Garg, Ramesh Karri and Muhammad Shafique found that Large Language Models (LLMs) achieved a higher success rate than an average human participant in solving Capture The Flag (CTF) challenges. This research highlights the potential of LLMs in cyberse...

Freeman Jiang introduces Nexus, a 3D data visualization of hacker interests

Chiyu Zhang, Yifei Sun, Jun Chen, Jie Lei, Muhammad Abdul-Mageed, Sinong Wang, Rong Jin, Sem Park and Ning Yao and Bo Long use pretrained language models (PLMs) to encode user histories and candidate items, treating the task as textual semantic matching. SPAR addresses long user engagement history and insufficient user-item i...

Niklas Muennighoff, Hongjin Su, Liang Wang, Nan Yang, Furu Wei, Tao Yu, Amanpreet Singh and Douwe Kiela introduce generative representational instruction tuning (GRIT) whereby a large language model is trained to handle both generative and embedding tasks by distinguishing between them through instructions.

In RealClimate, Stefan reports on a new study published in Science Advances suggests that the Atlantic overturning circulation (AMOC) is on a 'tipping course' and could collapse if the northern Atlantic Ocean is diluted with freshwater, reducing its salinity and density. The study confirms past concerns that climate models sy...

Cory Doctorow examines the CHIPS Act and its implications for America's high-tech industry. He argues that while the Act may address some of the issues caused by monopolies in the sector, it fails to tackle the underlying problems.

Quantamagazine - The discovery of the default mode network, a collection of seemingly unrelated areas of the brain that activate when you’re not doing much at all, has offered insights into how the brain functions outside of well-defined tasks. It is active during rest, when we turn mentally inward, and uses more energy than ...

In 404Media, Joseph Cox reports on an underground website, OnlyFake, is using 'neural networks' to generate realistic looking fake IDs for just $15. This technology threatens to streamline everything from bank fraud to money laundering and has implications for cybersecurity.

Liangsheng Yin, Ying Sheng, and Lianmin Zheng introduce an optimization for constrained decoding of JSON or YAML in LLMs using a compressed finite state machine. This method reduces latency by up to 2x and boosts throughput by up to 2.5x compared to state-of-the-art systems.

Nostalgebraist argues that while few people predicted the current revolution in AI, it is actually a 'revolution of predictability', with companies spending huge amounts of money on training runs and feeling secure due to the predictable input-output relations.