https://github.com/sapientinc/HRM

Hierarchical Reasoning Model is a new architecture that's inspired by neural computation principles observed in the brain, such as hierarchical processing, temporal separation of neural rhythms, and recurrent connectivity.

The bio-inspired design demonstrates significantly improved efficiency and accuracy on complex reasoning tasks compared with current LLMs.

The HRM architecture is designed to achieve significant computational depth while maintaining stability and efficiency during training. It consists of two interdependent recurrent modules operating at different speeds.

The High-Level module operates slowly and is responsible for abstract planning and deliberate reasoning. The Low-Level module functions rapidly, handling detailed computations.

A dual-module system allows the HRM to perform sequential reasoning tasks in a single forward pass without needing explicit supervision of intermediate steps. The model is also designed to be Turing-complete, meaning it can theoretically simulate any Turing machine, overcoming the computational limits of standard Transformer models.

Another interesting feature is the use of one-step gradient approximation, which improves efficiency by avoiding the computationally intensive backpropagation through time method typically used for recurrent networks. Avoiding backpropagation offers a constant memory footprint, making the model more scalable.

The model also incorporates an Adaptive Computation Time mechanism, inspired by the brain's ability to switch between fast, automatic thinking and slow, deliberate reasoning. The HRM is thus able to dynamically allocate computational resources based on the complexity of the task.

Despite having only 27 million parameters, the HRM achieves nearly perfect performance on difficult tasks like complex Sudoku puzzles and finding optimal paths in large mazes, areas where even advanced models using Chain-of-Thought (CoT) methods fail completely.

The HRM also outperforms much larger models on the Abstraction and Reasoning Corpus benchmark for artificial general intelligence. It achieved a 40.3% accuracy, surpassing models like 03-mini-high (34.5%) and Claude 3.7 8K (21.2%).

The model's design means that its training phase is much cheaper as well. It can be trained effectively with a small number of examples (around 1,000) and does not require pre-training or CoT data.

HRM conducts computations within its internal hidden state space which is more efficient than CoT where reasoning is externalized into token-level language. The externalization process can be brittle and requires extensive data to work.

At the high-profile summit on Tuesday—where, in addition to Sacks, panelists and attendees included Anthropic CEO Dario Amodei, Google president and chief investment officer Ruth Porat, and ExxonMobil CEO Darren Woods—companies announced $92 billion in investments across various energy and AI-related ventures. These are just the latest in recent breakneck rollouts in investment around AI and energy infrastructure. A day before the Pittsburgh meeting, Mark Zuckerberg shared on Threads that Meta would be building “titan clusters” of data centers to supercharge its AI efforts. The one closest to coming online, dubbed Prometheus, is located in Ohio and will be powered by onsite gas generation, SemiAnalysis reported last week.

For an administration committed to advancing the future of fossil fuels, the location of the event was significant. Pennsylvania sits on the Marcellus and Utica shale formations, which supercharged Pennsylvania’s fracking boom in the late 2000s and early 2010s. The state is still the country’s second-most prolific natural gas producer. Pennsylvania-based natural gas had a big role at the summit: The CEO of Pittsburgh-based natural gas company EQT, Toby Rice—who dubs himself the “people’s champion of natural gas”—moderated one of the panels and sat onstage with the president during his speech.

All this new demand from AI is welcome news for the natural gas industry in the US, the world’s top producer and exporter of liquefied natural gas. Global gas markets have been facing a mounting supply glut for years. Following a warm winter last year, Morgan Stanley predicted gas supply could reach “multi-decade highs” over the next few years. A jolt of new demand—like the demand represented by massive data centers—could revitalize the industry and help drive prices back up.

Full Article

Abridged version

This isn’t the first time I’ve blogged about the dearth of truly great PC laptops out there, and I suspect it won’t be the last.

Seven product lines (or is it eight; there’s an extra one in the sidebar not shown in the main view) and 330 distinct models! How can a normal person who isn’t a laptop enthusiast find anything in here? Even my eyes glaze over when I’m trying to distinguish the differences between the models and product lines.

HP further complicates things by having separate sites for consumer laptops and business laptops.

You might think that this level of choice should provide anything one could want, but that’s not true. Most of the models differ by like 1% and make all the same mistakes, copy-pasted across the while product line. Maintaining so many product lines at a reasonable level of development and quality is impossible, even for companies of their size with billions of dollars to throw at the problem.

These companies are clearly trying to micro-target specific market segments to match prices to buyers’ budgets, but offering so much choice is foolish. Most buyers — even big commercial buyers — are not informed enough to be able to pick the perfect device from among a massive blob of options presented at the same level, causing choice paralysis and lost sales, disappointing purchases that reduce brand loyalty, and expensive returns.

There has to be a better way!

The author is a high level contributor to KDE (read: nerd), and even they can't figure this shit out. Honestly the laptop market is getting more shit by the day with every company taking a leaf out of apple's book (no upgradeability, repairability, glossy design) but they don't have the credentials to back it up. You end up in a market where everyone's chasing the macbook pedestal but are shipping hot, plastic devices with no ports and soldered components.

Hey comrades, I've just purchased a new motherboard and an MP44L SSD to put my OS and my currently-playing games in. The thing is, I also thought that this would be the perfect time to finally make the switch into Linux and a more FOSS-based approach to the time I spend using my computer.

I tried Linux like twenty years ago and did not adapt to it at all. Nowadays I'm much more knowledgeable about computers in a general way but I have a massive blind spot when it comes to Linux. I want to ditch Windows but frankly don't even know where to start the switch.

So I have the following questions, I hope you can help me figure things out:

1 - Is dual-boot a plausible thing? Like, having a Linux distro installed for everyday usage, and Windows for gaming only?

2 - Speaking of which, I've heard good things about gaming on SteamOS. What's going on with that? Honestly, I'm completely clueless and I thought it was a proprietary OS for the Steam Deck. Is it already available for PCs? Also, is it safe? I don't want to just switch the company that has me under their thumb from MS to Valve.

3 - Are there any pages / youtube channels / other kinds of resources you would recommend, so that I can do some learning?

Thanks!

Context elsewhere shows that he was conscripted by Ukraine. In this message, they leave some plausible deniability.