Father/husband. Builder/Coder/Lifter/Cooker of things. Founder bitflux.ai. 5th Gen NorCal. Both grandpas irradiated by US govt - Mutant Powers: TBD
Joined July 2014
- Tweets 2,525
- Following 810
- Followers 270
- Likes 2,068
177 Photos and videos
Which is faster? ghostty WASM vs xterm.js
It's complicated.
I vibed a prototype and ran vtebench on it. My conclusion is I'm gonna use term.js but I'm looking forward to ghostty WASM maturing.
@ibuildthecloud commented that ghostty felt less responsive, it depends on which package wrapping it I used. Maybe there is some deeper build issues here too. Let me know if you see my methodology or build is flawed.
github.com/jaredeh/webtermte…
1
23
Confirmed, sort of. ghostty-web definitely has a little delay when typing compared with xterm.js, but wterm/ghostty seems a little faster than xterm.js. But my p10k shell looks worse on the ghostty versions I tested. wterm would be the win but the color isn't clearing from the status bar and there is a little glitch with the height of the colored cells. I'm just gonna use xterm.js for now in my project. I mostly just let agents put a test together and my React-fu is weak, so maybe there is something easy I'm overlooking let me know if you figure something out. github.com/jaredeh/webtermte…
21
Jun 13
"We should change the system so that Social Security trust fund can own equities. That should then turn into an account-based system so every individual has an account rather than a benefit. And we should use that capital to go and buy stocks. We should buy great stocks in great American companies that are going to reinvent the world." - @friedberg
I've been skeptical of the sovereign wealth fund because it's another way for the government to pick winners and losers in the market for political or cronyism reasons and there is no guarantee the wealth helps citizens.
This is the first vision of doing it that makes sense to me. I still think there is a risk of government putting their thumb on scale too much. But there could be ways around that. Can citizens have voting rights for the stock in their accounts? Can citizens choose which stocks?
23
Jun 11
My customer service calls to @fastrakbayarea have always been fast, friendly, and effective. Even though it's always been my screw up, they make me feel like I'm doing them a favor. Makes me have warm feelings for the agency taxing me $8.50 to see my daughter via public roads.
23
Jun 10
Illegible vs legible is the current SOTA framing for where value is/isn’t. We’ve long had a problem measuring white collar work, now solving that is existential.
Also illegible: Territory where the consensus is wrong. Current tech models are blind here, they need hand holding.
69
Jun 7
t was corny, ultra-formulaic, oddly preachy, and didn’t try to explain how weird and stupid it was. Everything you could want in HeMan. I had fun.
Alison Brie played Annie playing Evil-Lyn. And Skeletor is the role Jared Leto came to Earth to play.
I have notes where they could have better, of course. But good job. 👏
Jun 2
Okay. I'm gonna see it.
Look, I'll admit was nothing but a super corny, ultra-formulaic, and oddly preachy weekly ad for really strange toys. But I loved it and so did my older kids.
Hollywood, come up with new IP or respect the source material if want us to show up.
154
Jun 5
Was not on my bingo card
Jun 5
Gwyneth Paltrow just invited Anduril cofounder @traestephens onto the Goop podcast — and, in a wide-ranging conversation about love, war, God, and Gwyneth’s leftist husband who thinks she’s “becoming a Republican”… Trae explains Anduril’s core goal: to engineer lasting peace.
America, Trae explains, needs AI-assisted weapons that can execute missions without endangering American lives.
Sometimes conflict is necessary. Gwyneth gets it.
Full breakdown from @harrissockel 👇
60
Jun 2
Okay. I'm gonna see it.
Look, I'll admit was nothing but a super corny, ultra-formulaic, and oddly preachy weekly ad for really strange toys. But I loved it and so did my older kids.
Hollywood, come up with new IP or respect the source material if want us to show up.
Exclusive clip of Skeletor and Evil-Lyn in the new ‘MASTERS OF THE UNIVERSE’ movie.
In theaters on June 5.
1
215
May 28
Yes.
It’s not that we’ve discovered some magic bullet, but rather that JAX, or at least the open source version of it, is mostly optimized for small to medium-sized training runs on Google TPUs, whereas we need to massive training runs on Nvidia GPUs.
Pipeline parallelism is essential and crushes fully-sharded data parallelism at scale.
And C will compile to the most efficient binary short of assembly. Maybe we will do a little assembly too.
52
May 28
love this.
SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible.
The potential speed improvement vs JAX for large training runs is over an order of magnitude.
31
May 16
If you're a zealot about one programming language. You've lost the plot. They are increasing fungible.
30
May 15
This thread made me smile.
Musk's lawyer brought a big monitor (maybe 36 inches) into the courtroom. OpenAI's lawyers asked to use it. Musk's lawyers said no. The judge told Musk's lawyers they have to let OpenAi use it. Then OpenAI said it might not be possible to connect their laptops to it.
61
May 15
Training for professions will be harder going forward.
You need to UNDERSTAND the first principles of the tool/material/domain working with.
If not you're just slowing down the AI, you're a fleshy Ralph loop waiting to be automated away.
2
1
27
May 13
You can reply to AI with memes. And it gets it.
Intellectually I knew this was possible, it's still very weird.
20
May 11
AI possesses incredible detailed knowledge about computer hardware, yet it displays no intuition about how use that in practice when coding. It's a very human like tendency.
Reminds me of many distributed systems experts.
18
May 1
You could pave 6 lanes of I-5 with G-wagons and still have the original estimate in cash left over.
In a newly proposed business plan, project leaders estimate the Los Angeles-to-San Francisco segment will cost about $126 billion, with service beginning around 2040. ktla.com/news/california/cal…
67
Apr 16
New @a16z branding has grown on me. When I just saw it in a dinky logo image, I kinda hated it. But now that I've seen it in context. I like the IDK Art Deco vibe.
24