junjihashimoto

[junjihashimoto @junjihashimoto3]

@scottnarmstrong It is not often discussed, but it's true. While the proof itself is easy, what matters is the formulation—including the theorems, conjectures, and the management of the interface with requirements engineering.

[junjihashimoto @junjihashimoto3]

As I recall, Linux was adopted for embedded systems using both 32bit CPU and MMC, because developers wanted to use TCP/IP. However, if AI becomes capable of generating TCP/IP libraries that operate without an OS, we might see a shift away from Linux in embedded applications.

[junjihashimoto @junjihashimoto3]

@srush_nlp You may have already experimented with this, but I am endlessly curious about things like attention performance and memory limits—as well as the performance and memory usage when combined—and whether quality is maintained in diffusion models.

[Sasha Rush @srush_nlp]

Lean Verified Transformers (https://t.co/ZdDzZwa4iO)

In which we prove a bunch of Transformer invariants from scratch in Lean, and speculate about how hard it would be to do that for the rest of the world's code.

[junjihashimoto @junjihashimoto3]

@AndrewCurran_ It's fun to make predictions. Here is a new one:
The mechanism of high-temperature superconductivity has been mathematically elucidated. (I’d like to see them compete on benchmarks involving physics proofs next, rather than mathematical ones.)

[junjihashimoto @junjihashimoto3]

eBPFでPHP APMを自作した話の発表資料とソースコードです。https://t.co/T9OBJC00m2 https://t.co/QtuifEUNfy #CNCJ #ebpfjp

[junjihashimoto @junjihashimoto3]

New "sparkle" tutorial (Lean 4 HDL framework)! We formulate a PID controller with reals & prove Lyapunov stability. Then, we convert it to fixed-point for hardware and formally verify the stability STILL holds! https://t.co/Wt90Wq2X2u #Lean4 #FormalVerification

[junjihashimoto @junjihashimoto3]

@satnam6502 @avi_press I suppose it’s a trade-off between exploring new capabilities and maintaining compatibility with existing ones, but personally, I’d like to use Lean by taking advantage of features that are unique to formal verification.

[junjihashimoto @junjihashimoto3]

@satnam6502 @avi_press I keenly realize the importance of TAT in my own development work as well. At the same time, why is it that the development of coding agents so often shifts from JavaScript to Rust?

[junjihashimoto @junjihashimoto3]

I believe this is the first time an Ethereum signing device has been developed using an FPGA (Tang Nano 20K) with a design created in Lean 4.
https://t.co/QnYqZ9w7XW

[junjihashimoto @junjihashimoto3]

@austinvhuang I hope the patent I proposed before making it is valid.

[junjihashimoto @junjihashimoto3]

@austinvhuang I just realized this.
https://t.co/SekZuwjGfi
It all started with a project I created to realize the same idea as taalas, but I was way behind the times.
https://t.co/wglzYt3HgW
https://t.co/UiYzaYggi7

[junjihashimoto @junjihashimoto3]

@AnthropicAI Next, please create Linux. Can you do it in a month?

[junjihashimoto @junjihashimoto3]

As a result, projects that use Hasktorch can now be built directly on Hackage. The first example is https://t.co/VXYn32scTr . Thanks again!

[junjihashimoto @junjihashimoto3]

@hasktorch Good news for folks who thought setting up Hasktorch was tough — it now auto-downloads LibTorch. No more env setup. Big thanks to the Hasktorch team!

[junjihashimoto @junjihashimoto3]

https://t.co/sod8xeuAru

[junjihashimoto @junjihashimoto3]

@i2cjak It's partly a technical issue, but it's also important to use a tester to test whether the parts you want to connect are connected and whether there are any short circuits with the ones next to them.

[junjihashimoto @junjihashimoto3]

次は心、個性、信念、正義感の制御？　https://t.co/i3kstG3A1L

[junjihashimoto @junjihashimoto3]

@ncaq https://t.co/bxEfj9LfZT
このベンチマークと同じものをhasktorch でも提供してます。
小さなサイズであればmassivが速く、大きなサイズであればhasktorchの方が速い方が出てきます。

[junjihashimoto @junjihashimoto3]

@zyklotomic @hypernoided It is already done.
https://t.co/VsdoHrOjQ8