The trailer suggests that marine armor gets reused between marines. That explains why this mech suit only costs 50 minerals.
3
567
Trainium has by far the best profiler of any accelerator. As you can see we have nanosecond-accurate traces of our programs, allowing you to write a program that generates images reliably in the profiler.
10
23
371
32,961
Sasha Krassovsky retweeted
The VOD for my BSC2026 talk "The Root of The Root of All Evil" is now available! It's the spiritual successor to last year's "The Big OOPs", so if you enjoyed that one, join me for one more adventure into the history of computer science:
computerenhance.com/theroot
25
86
777
46,480
This morning my espresso scale wasn’t working so then I plugged it into my laptop and Claude debugged it over serial. It had me wiggling wires while it read the raw ADC values. The future is absolutely wild
13
524
When I was a kid and had heard of JIT compilation for the first time, I thought it worked was having a compiler thread and a runner thread, and the compiler would fill in the runner’s instruction stream WHILE the runner was executing. The mystery was in how the compiler would always finish compiling just in time for the runner to continue executing. “Wow a hard realtime compiler that predicts exactly how long a given chunk will execute!!”
Then when I found out the truth I was pretty disappointed.
12
8
436
50,058
I bought this and it’s honestly amazing how snappy it is. We need all software to be this fast. Who’s rewriting emacs to be this snappy?
1
5
1,533
Sasha Krassovsky retweeted
I did not sign this petition fwiw. Still not at all convinced that slower is safer than faster. With strong benevolent institutions slower is probably safer on net, but with our current crop it's a much more mixed bag.
We support this petition, signed by our CEO, several co-founders, and senior staff.
Our own research on recursive self-improvement, published last month, points to the need for tools to deliberately pace the frontier of AI development so society can prepare. We’re glad to see broad agreement across the field. pacingthefrontier.com/
40
38
652
73,625
Booted up Order of the Sinking Star to try it out for 15-20 minutes before bed and now it’s been 2.5 hours and I don’t know where the time went.
The puzzles are fantastic, they somehow entice you at first glance but then you have to spend time refining your idea.
1
457
It's a crazy good model - the first one to mostly replace handwritten code for me. That's how good it is.
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision.
The longer and more complex the task, the larger Fable 5’s lead over our other models.
ALT Benchmark table titled Mythos 5 & Fable 5, comparing Claude Mythos 5 and Fable 5 against Claude Mythos Preview, Claude Opus 4.8, GPT 5.5, and Gemini 3.1 Pro.
1
14
1,060
Opus 4.7 haters in shambles
Many (many) months later, Claude is finally challenging the Elite Four and will likely become a Pokemon Champion tonight 🥲
twitch.tv/claudeplayspokemon
1
517
This is the part of today’s announcement I’m most excited about 🙂
SpaceXAI and @AnthropicAI have also expressed interest in partnering to develop multiple gigawatts of orbital AI compute capacity
5
716
OK last post for the night: I tried all the fancy stuff they recommended in their GEMM doc: Z-curve, static extents, accumulation group synchronization. None of it seemed to make any performance improvement - I seem to be stuck at 40 TFLOPs in bf16 across a variety of shapes.
I got my M5 MacBook over the weekend and had some time to mess around with Metal 4 and the Neural Accelerators!
Wanted to document some of my first impressions below:
2
1
20
3,611
I got my M5 MacBook over the weekend and had some time to mess around with Metal 4 and the Neural Accelerators!
Wanted to document some of my first impressions below:
3
7
233
58,570
I was also expecting a much more dramatic speedup from the Neural Accelerator. It seemed that with my original tile size of 32x32, I was only getting 244 GB/s of memory bandwidth. Bumping it up to 64x64 gave me 740 GB/s, dropping the time to 3.36ms!
1
2
20
5,131
Overall had a fun time! To close off with some criticisms:
- it took me a long time to figure out how to enable Metal 4. I wish this were better-documented
- MPP seems a little boiler-platey. I wish there were a slightly more convenient syntax for this stuff, but not a dealbreaker.
Hope this was interesting!
2
3
35
4,413