HN user
https://www.linkedin.com/in/ivan-krasin-0ba73ba/
I'd like to see a system which attempts to reward people for the full transitive value of their work as long as the work remains valuable.
To a degree, this is what copyright was supposed to do.
U-234 is ~3000x less radioactive than Pu-238, so having ~120g of U-234 is negligible.
I really fail to see a problem with these tiny amounts of non-brittle material embedded into a solid case. It's still very dangerous, but it's locally dangerous (meters away), not at the scale of whole countries.
I found the specs for the fuel source: https://upload.wikimedia.org/wikipedia/en/e/e9/SNAP-19C_Moun...
The high-power unit had 300 grams of Pu-238 in 1965. Given its 87.7 years half-life, only 187g of Pu-238 remaining. It's very hard to do much damage with this amount of radioactive material.
that (checks notes) has the potential to poison most of North India.
How large is the amount of plutonium in there? I highly doubt that it has the claimed potential.
I am usually running Claude Code in tmux on a Mac. Absolutely zero scroll issues (but I use tmux scroll for that, naturally). It's a delight.
Apple raised iPad Pro prices in 2017 by $50 to address NAND memory chips price hikes from the vendors: https://www.cnbc.com/2017/09/13/ipad-pro-price-increase.html
Quilter is a real deal. Hopefully, open source catches up at some point. But for now, it's Quilter or a lot of manual work & more design iterations.
Yeah, this company (GeneralistAI) is, in my opinion, the most advanced robotics+AI company in the world. Slightly behind them Google DeepMind Robotics and Physical Intelligence, and then the rest.
In the meantime, China is constructing nuclear cargo ships ([1], [2]) that will be able to transport 14,000 containers at full throttle (200MW) without a need to refuel for years.
Obviously, it's still not done, and yet to prove to be profitable, but their reactor design does suggest that they have a chance to make it work and replace a lot of CO2 emissions.
1. Original article, but paywall: https://www.scmp.com/news/china/science/article/3331031/chin...
2. Derivative (+based commentary), but no paywall: https://www.chosun.com/english/world-en/2025/11/07/MND3QUGUT...
I refer to the RDNA4 instruction set manual ([1]), page 90, Table 41. WMMA Instructions.
They support FP8/BF8 with F32 accumulate and also IU4 with I32 accumulate. The max matrix size is 16x16. For comparison, NVIDIA Blackwell GB200 supports matrices up to 256x32 for FP8 and 256x96 for NVFP4.
This matters for overall throughput, as feeding a bigger matrix unit is actually cheaper in terms of memory bandwidth, as the number of FLOPs grows O(n^2) when increasing the size of a systolic array, while the number of inputs/outputs as O(n).
1. https://www.amd.com/content/dam/amd/en/documents/radeon-tech...
2. https://semianalysis.com/2025/06/23/nvidia-tensor-core-evolu...
This sounds like a happy ending for the employees of Windsurf and a good deal for Cognition
The employees were robbed from having a big cash exit. Illiquid stock options from Windsurf were converted to illiquid stock options of Devin.
What's worse is that the well is now poisoned. I would advise against joining startups from now on, because I think that there's no upside for employees anymore.
Not only they do not release models/weights. They don't even tell the size of the models!
The linked whitepaper is pretty useless, and I am saying as a big fan of diffusion-transformers-for-not-just-images-or-videos approach.
Also, Gemini Diffusion ([1]) is way better at coding than Mercury offering.
The story is amazing and officially available online: https://marshallbrain.com/manna1
This is just a fake product page to gauge interest.
From their "order" page:
UPERFECT Omega is currently in the concept stage and will soon be put into research and development.
Quote from the article:
With these innovations, the team fabricated a 1.09 cm2 flexible monolithic perovskite/CIGS tandem solar cell. Competing with top rigid counterparts, the device achieved an impressive stabilized efficiency of 24.6% (certified at 23.8%), one of the highest reported values for flexible thin-film solar cells to date.
After 320 hours of operation and 3,000 bending cycles at a radius of 1 cm, the device retained over 90% of its initial efficiency, demonstrating exceptional mechanical durability and long-term stability.
Slightly tangential, but what's the state of the art in auto routing and placement these days?
I always recommend quilter.ai: https://www.quilter.ai/ - they "eat" Kicad or Altium files, and produce autorouting and/or autoplacement. I tried them about 9 months ago and made a couple of boards that worked well.
It's 32GB for $1300 for Blackhole p150a([1]). The rest of your point is very true.
They seem to be showing very decent performance results for diffusion transformers. Not so much for the autoregressive transformers (the "regular" ones).
DeepSeek R1 has 671B parameters. If you want to run at 8 bits/parameter (the native format for DeepSeek R1, in which it was trained, so the top performance possible for the model), one would need about 32 Blackhole p150a ([1]), which is around $42k and 10kW power consumption.
So, yes, you can run DeepSeek R1 on it, but there are cheaper options (if we only talk about inference).
It's far from the first generation. Unitree has been selling robot dogs for over 6 years now and I remember them being a lot less refined. See, for example, what Unitree was selling in 2019: https://www.youtube.com/watch?v=FFJUkrCQzx8
They just need a few more iterations.
Somewhat related: a teardown of a Unitree Go2 robot dog by iFixit: https://www.youtube.com/watch?v=YjVbW6Fc11Y
TL,DR: Unitree is making a real attempt to make their robots to be at least somewhat repairable, but these robots are still very difficult to work with due to glued connectors, tight cable managements, hundreds of screws, lack of availability of spare parts, lack of documentation, etc.
Is this some joke? They use Llama 2 7B? What year is it?
They use llama2 to demonstrate that their compression method works. There are potential cases:
1. The method works on all / most LLMs. In this case, it does not matter on which model they demonstrated the effect.
2. The method only works on llama2, but not on other models. Given that they published the code, I expect that people will quickly test the method on many other models, so we will know that soon. And yet - there would be a scientific significance even if it works only on llama2, as it would mean that there's some special and good in that architecture.
But I would bet it's #1 - the method works on most of the models and they just picked whatever they had already had code bindings to, to save the effort.
RealSense is a golden standard in Robotics research. ALOHA, which was one of the first swallows of the new AI+Robotics wave, was enabled by RealSense: https://aloha-2.github.io/
Relevant (upcoming) book, "AI for Robotics: Toward Embodied and General Intelligence in the Physical World" from one of the authors of RT-1 (Robotic Transformer) from Google DeepMind Robotics group: https://www.amazon.com/dp/B0DFLTZ4SQ?ref=ppx_yo2ov_dt_b_fed_...
I have not read it yet (it's being released in May), so can't give a true recommendation, but I did preorder it. The state of art is changing fast, and the book is expected to capture at least some snapshot of it.
With siblings of similar age, it's much harder in the first 2-3 years (external help is essential), but then the kids considerably shortcircuit to each other, so it becomes easier than having just one, constantly bored child.
Source: I am a father of two daughters with less than 2 years difference; and my brother is also less than 2 years apart from me. So, I lived my advice in both aspects: as a parent and as a child.
Yes and no. Jetson line (which Jetson AGX Orin is a part of) is also providing multi-camera support (with MIPI CSI-2 connectors) and other real-time / microcontroller stuff, as well as rugged options via partners.
I hope to see new Jetsons based on Blackwell sometime in 2026 (they tend to be slow to release those).
I generally agree with your analysis of the current state of art but strongly disagree with the overall conclusion of where it leads us.
I believe it will take on the order of 100M hours of training data of doing tasks in real world (so, not just Youtube videos), and much larger models than we have now to make general-purpose robotics working, but I also believe that this will happen.
I've saved your comment to my favorites and hope to revisit it in 10 years.
What is the new breakthrough in robotics that is gpu driven ? There are subsets of the overall problem that can be solved by a gpu (eg object detection) but the whole planning and control algo scheme seems to be more or less the same as it has been for the past decades. These typically involve non-convex optimization so not much gpu benefit.
In the past two years two very important developments appeared around imitation learning and LLMs. Some starting points for this rabbit hole:
1. HuggingFace LeRobot: https://github.com/huggingface/lerobot
2. ALOHA: https://aloha-2.github.io/
You're correct on $/bandwidth. The point about low latency continues to be ignored, though.
If models keep lasting ~year timescales could we ever see people going with ROM chips for the weights instead of memory?
Before ROM, there's a step where HBM for weights is replaced with Flash or Optane (but still high bandwidth, on top of the chip) and KV cache lives in SRAM - for small batch sizes, that would actually be decently cheap. In this case, even if weights change weekly, it's not a big deal at all.