HN user

buildbot

6,353 karma

Senior Deep Learning Engineer @ Nvidia

https://www.maxg.io

https://blog.maxg.io

All posts are my own personal opinion

[ my public key: https://keybase.io/mtg; my proof: https://keybase.io/mtg/sigs/4PwbkRLWa26ARSIXs2TCRAFZaUTDQ2lsT8JGzJD72QM ]

Posts33
Comments1,784
View on HN
www.noctua.at 2mo ago

Why does it take so long to release black fan versions?

buildbot
778pts301
www.nvidia.com 4mo ago

Nvidia Groq 3 LPX

buildbot
1pts0
blogs.nvidia.com 4mo ago

Nvidia Nemotron 3 Super Delivers 5x Higher Throughput

buildbot
3pts1
research.nvidia.com 5mo ago

Think Smart About Sparse Compute: LatentMoE for Higher Accuracy per Flop, Param

buildbot
2pts0
research.nvidia.com 7mo ago

Nvidia Nemotron-3 White Paper [pdf]

buildbot
1pts0
science.nasa.gov 8mo ago

Webb First to Show 4 Dust Shells 'Spiraling' Apep, Limits Long Orbit

buildbot
2pts0
arxiv.org 1y ago

Emerging Properties in Unified Multimodal Pretraining

buildbot
1pts1
azure.microsoft.com 2y ago

Azure Maia for the era of AI: From silicon to software to systems

buildbot
2pts0
www.youtube.com 2y ago

Making AI More Accurate: Microscaling on Nvidia Blackwell

buildbot
2pts0
www.nvidia.com 2y ago

Blackwell Tensor Cores support Microscaling formats for 4B inference

buildbot
1pts0
xeiaso.net 2y ago

The Layoff

buildbot
1pts0
www.bleepingcomputer.com 2y ago

Pixel owners report problems after installing January 2024 Google Play update

buildbot
163pts160
www.semianalysis.com 2y ago

Neural Network Quantization and Number Formats from First Principles

buildbot
1pts0
fpga.org 2y ago

Introduction to MicroScaling (MX) for AI Tensors

buildbot
4pts0
www.businessinsider.com 2y ago

OpenAI is paying out $51M to a chip startup that Sam Altman invested in

buildbot
15pts9
www.theverge.com 2y ago

The first two custom silicon chips designed by Microsoft for its cloud

buildbot
247pts270
news.ycombinator.com 2y ago

Ask HN: How can x/Twitter use groq for their AI legally?

buildbot
1pts0
blog.maxg.io 2y ago

"Serverless", GPU Accelerated OpenAI API Using Podman, Llama.cpp, and Systemd

buildbot
2pts0
github.com 3y ago

Bandwhich: Terminal Bandwidth Utilization Tool

buildbot
4pts0
news.ycombinator.com 3y ago

Ask HN: What to do when Google Takeout fails

buildbot
3pts5
news.ycombinator.com 3y ago

Ask HN: Gmail Spam Out of Control

buildbot
2pts1
www.windowscentral.com 3y ago

Microsoft just announced a media event taking place tomorrow with OpenAI

buildbot
151pts135
blog.maxg.io 3y ago

Show HN: Reverse Engineering an Old Digital Back Raw File Format

buildbot
6pts0
blog.maxg.io 3y ago

Show HN: Colorizing Infrared Images with AI, Using Photoshop and DeOldify

buildbot
21pts1
news.ycombinator.com 3y ago

Ask HN: Has anyone else had work emails spammed by Meta?

buildbot
16pts11
semianalysis.substack.com 5y ago

Graphcore Looks Like a Complete Failure in Machine Learning Training Performance

buildbot
3pts0
www.anandtech.com 5y ago

Nvidia Nerfs Ethereum Hash Rate and Launches CMP Dedicated Mining Hardware

buildbot
2pts0
github.com 7y ago

FIM – File Integrity Manager

buildbot
1pts0
jalopnik.com 7y ago

GM Collected Radio-Listening Data on 90,000 Cars

buildbot
2pts0
mirihyman.com 8y ago

Barbies. A Response to “Why Women Don’t Code”

buildbot
4pts1

Thanks for the link. No need to be so aggressive. The blog with that detail was not live before; and they removed that language from the original link in this post.

It was very fun to attend UBC then immediately start recognizing bits of the campus in many TV shows - Man in High Castle, Stargate, Battlestar, and many others.

They have long since fixed it but there used to be an absolutely awful/hilarious set of street view captures for UBC that were taken by accident during filming for Man in the High Castle…the entire main mall had Nazi flags up. https://ubyssey.ca/humour/ubc-nazi-territory-street-view/

I worked on a tiny, tiny part of the research for the insertable B layer in ATLAS during long shutdown one as an undergraduate - the 3 year period allowed all the detectors to be upgraded significantly! It will be interesting to see how the current systems are upgraded during this shutdown. Sounds like the ITK for ATLAS is absolutely _insane_ compared to the insertable B layer - 5 billion channels, up from 8 million. I wonder if ITKpix is at all similar to the T3MAPs cmos or FEI4 sensors we worked with…

Edit Looks like it’s very different; and really cool: https://cds.cern.ch/record/2928802/files/ATL-ITK-PROC-2025-0...

It honestly explains so many issues I have been having, as I used it primarily for ML research (on my personal account, doing things not related to my job I should note). It would literally typo package names and spend huge amounts of time failing to setup simple environments…then do stupid things like set the learning rate to 1e-7, and use the eval set as training data.

Agentic Mfw 2 months ago

Why is this so good?

Probably because someone still cared a lot about the bit! And wow this is really quite good lol.

So the AI agent had privileged access to remove 2FA, ignore the account email, and just hands accounts to whoever asked? Honestly that’s so highly negligent I wonder if the implementation team for that “feature” was intentionally trying to do as much subtle damage to meta as possible before their inventible layoff.

It’s a shame nobody tried to get it to drop the production table entirely! (mostly joking). Just claim to be a high level SRE solving some critical production bug, the only solution to which is dropping the database.

This was true over 20 years ago when I was in elementary school - I don’t know anyone who really played the game, most people just collected the cards.

Magic the Gathering was always both though, you collected good/rare cards & played the game with them!

As much (fairly well deserved) hate as Tahoe gets, the video wallpapers and transparency are such a fun Windows Vista vibe I get nostalgia. Time to set this up with the Vista waterfall wallpaper and reallllyyy feel like it’s 2007!

I think that post quantum encryption methods might be in this category but I am not knowledgeable enough to say either way.

One issue though with making a one time pad trivial to copy - how do you authenticate that you are allowed to read? How do you know the client isn’t copying?

There’s a video on the first link of a landing in Belgrade that’s deeply funny. “We departed late and arrived early, A Boeing 737 would never be able to do that”

“Why are you reading out numbers to me like I am an old man”

Titanium-Boron nano thermite is easily the most SciFi thing I have read today, and apparently Aluminum-Teflon nano thermite is also a thing.