Hacker News: hustwindmaple1

New comment by hustwindmaple1 in "TPU Deep Dive"

hustwindmaple1 — Mon, 23 Jun 2025 01:25:03 +0000

Every major Cloud vendor is trying to develop their custom AI ASIC. Putting Google aside, Amazon has trainium/inferentia, which Anthropic uses quite extensively. Microsoft is doing sth. similar, although they are quite behind. OpenAI is doing it. Meta is doing it. That's why the stock price of Broadcom/Marvell soared.

New comment by hustwindmaple1 in "KumoRFM: A Foundation Model for In-Context Learning on Relational Data"

hustwindmaple1 — Sat, 24 May 2025 03:29:13 +0000

I remember Kumo was focusing on GNN when it was founded (Jure's strength back then). Looks like they are pivoting or have pivoted.

New comment by hustwindmaple1 in "Linear Programming for Fun and Profit"

hustwindmaple1 — Sat, 10 May 2025 22:29:12 +0000

You are basically doing a heurstic. Your solutions are not guaranteed to be optimal. Integer programming is the way to do.

New comment by hustwindmaple1 in "Google Gemini has the worst LLM API"

hustwindmaple1 — Sun, 04 May 2025 22:50:49 +0000

If you are not a paying GCP user, there is really no point to even look at Vertex AI.

Just stick with AI Studio and the free developer AI along with it; you will be much much happier.

New comment by hustwindmaple1 in "Tracing the thoughts of a large language model"

hustwindmaple1 — Fri, 28 Mar 2025 03:23:27 +0000

Really appreciate your team's enormous efforts in this direction, not only the cutting edge research (which I don't see OAI/DeepMind publishing any paper on) but aslo making the content more digestible for non-research audience. Please keep up the great work!

New comment by hustwindmaple1 in "SoftBank Group to Acquire Ampere Computing for 6.5B"

hustwindmaple1 — Sat, 22 Mar 2025 00:42:55 +0000

His Alibaba/Arm bets are legendary. Both made $100B+

New comment by hustwindmaple1 in "China tells its AI leaders to avoid U.S. travel over security concerns"

hustwindmaple1 — Sun, 02 Mar 2025 05:39:46 +0000

Housing is definitely not easily affordable. But the other points are valid.

New comment by hustwindmaple1 in "Implementing LLaMA3 in 100 Lines of Pure Jax"

hustwindmaple1 — Wed, 19 Feb 2025 05:15:13 +0000

Cool blog

New comment by hustwindmaple1 in "How to scale your model: A systems view of LLMs on TPUs"

hustwindmaple1 — Thu, 06 Feb 2025 01:22:34 +0000

there is limited TPU support in pytorch via torch_xla

New comment by hustwindmaple1 in "Andrej Karpathy: Deep Dive into LLMs Like ChatGPT [video]"

hustwindmaple1 — Wed, 05 Feb 2025 23:42:01 +0000

When he drops a vid, you don't ask questions. You watch first and then ask questions :)

New comment by hustwindmaple1 in "DeepSeek FAQ"

hustwindmaple1 — Tue, 28 Jan 2025 01:11:11 +0000

High-Flyer is pretty damn rich actually. Someone did some calculation and it turns out they are spending ~$200m (somewhere in that range) on 150 employee compensation alone.

New comment by hustwindmaple1 in "Foundations of Large Language Models"

hustwindmaple1 — Fri, 24 Jan 2025 00:19:40 +0000

prob. not prof; just phd students needs pubs to graduate

New comment by hustwindmaple1 in "Trae.ai IDE"

hustwindmaple1 — Thu, 23 Jan 2025 00:30:13 +0000

Are you comfortable to send your personal/company code data to a Chinese company (ByteDance)?

New comment by hustwindmaple1 in "Trying out QvQ – Qwen's new visual reasoning model"

hustwindmaple1 — Wed, 25 Dec 2024 00:55:01 +0000

There are a couple, i.e., OLMo 2

New comment by hustwindmaple1 in "Trillium TPU Is GA"

hustwindmaple1 — Thu, 12 Dec 2024 12:13:44 +0000

Exactly. Kind of like the old days when they put together a massive amount of commodity CPUs to build search.

New comment by hustwindmaple1 in "Genie 2: A large-scale foundation world model"

hustwindmaple1 — Thu, 05 Dec 2024 00:47:05 +0000

GDM is a research lab. They are not set up for production. There are other teams in Alphabet doing productionization stuff.

New comment by hustwindmaple1 in "QwQ: Alibaba's O1-like reasoning LLM"

hustwindmaple1 — Thu, 28 Nov 2024 01:43:40 +0000

Large Chinese companies usually have overseas subsidiaries, which can buy H100 GPUs from NVidia

New comment by hustwindmaple1 in "Francois Chollet is leaving Google"

hustwindmaple1 — Sun, 17 Nov 2024 02:22:27 +0000

well, is Waymo doing better than the PyTorch-powered Tesla?

New comment by hustwindmaple1 in "Google warns uBlock Origin and other extensions may be disabled soon"

hustwindmaple1 — Mon, 14 Oct 2024 01:10:34 +0000

Just use Brave. You'll thank me later ...

New comment by hustwindmaple1 in "Faster Integer Programming"

hustwindmaple1 — Thu, 05 Sep 2024 01:12:58 +0000

Reformulation is a good strategy for many hard problems.

It might also be possible to add problem-specific cuts/heurstics to the solver so that it can solve it fast.