I have been building a large system recently this has been the case. However when you are building tooling where running agents all the time makes no sense IaC is still useful. So is the case of monitoring, agent can monitor things on their own but do you want agent to access your systems direct or first check the monitoring systems and then if info is not found go down to find access to the system.
IaC is dead.
Sorry DevOps / SREs / Cloud Engineers but you're in for the same roller coaster as Devs. AI can manage infra change better than you.
Unfortunately for IaC, we don't need code because with the actual system source code and the infra, AIs have everything needed to determine change sets / approvals and dev env lensing.
btw, Effect Oriented Programming probably will help you get there; but that is inevitable for your codebase.
This is quite amazing. Great job @divagr1925 .
well said @OfficialLoganK
You can hide your stuff from AI but then you will never be recommended by AI. Business need to think long and hard about this.
This has been true for donkey years. Do not ship secrets inside any app, iOS/Android/Windows/Toaster whatever
iOS apps that have baked-in secrets are in trouble.
The right solution is using AppAttest attestations to authenticate to services instead of baked in API keys:
developer.apple.com/document…
What the hell is happening. I used to think FP8 is a hardware feature. @CommonFoundry1 Do the FP8 perf here scales without any issues?
Nvidia makes plenty of money in server and data centre GPUs there should be no reason for them to squeeze their Gaming customers but here we are. The thing that built Nvidia. Gaming Revenue doesn’t even amount to a single digit percentage of their overall revenue.
Recently implemented some GLSL for music visualization and this blows my mind. Downside they are super taxing to render but even that will be solved with time
I see that these days coding agents would just use Bash(python3 - <<'EOF' as the tool to write code, and I know why they do it but man it is so furiating to not see the code changes happening and only see Run xxx commands.
For those who are wondering heredoc-to-python allows the model to overcome limitations of existing tools such as doing bulk edits, no requirement of giving old_str, write plus verify in one step etc. So yes this a better option but I hate not being in the loop.
One of 3090 Asus Blower edition was running very high memory temperatures so did repasting of all thermal pads. GA102 chip look like it has a nice face, don’t worry chip is not becoming sentient, my phones cameras are showing in here
This reminds me of the question that @dwarkesh_sp asked Jensen, if Nvidia just designs stuff and all the things are manufactured by somebody else then isn’t Nvidia just a software company and by that logic why would AI not eat that business too. Seems some of it has started happening. Having said that, TSMC won’t give fabs to anybody and Nvidia has booked that in advance. But if we prorate it even further, may be somebody using AI can indeed replace TSMC, I am sure the recent success CXMT and Chinese fab would have AI role in it as well. Interesting times!
OPENAI<>NVIDIA: It is clearly a good idea for OpenAI to start an ASIC program, spend hundreds of millions on R&D, present at Hot Chips, deploy some stuff, scare NVIDIA, and get some more money and backstops from NVIDIA on the order of billions of dollars.
If Jalapeno does well at scale, OpenAI will increase its revenue per MW. If Jalapeno doesn't do well, they will still get a lot of financing and backstops from NVIDIA, as Jalapeno is scaring NVIDIA.
@sama wins even if their chip loses.
UTM folks have provided a lovely insight on how they have brought Directx 11 support on Mac. I love how it studies existing implementations in Qemu, Wine etc and come up with a nice way to support. I am sure omarchy folks are getting trippy over making Hyprland working nicely in it. @dhh FYI
UTM v5.0.5 beta is out with support for DirectX 12! UTM is the first virtualization software for macOS that can run Windows 11 with FL 12_1. Your move @parallels @VMware
Infrastructure as code but for physical hardware. This technology allows you to swap any PCIe device like GPUs via software. I like the way it works: you plug in their adapter card that goes into their case with multiple GPUs installed, and then there is a layer that manages the routing. youtube.com/watch?v=8I2tXHN6…
Kunal Deo retweeted
We’re hiring for various roles on the Model APIs team at @sarvamai.
Our API business has grown 100x in the last year, and we’re still very early.
We’re looking for people who, regardless of their role:
• Try every new model the day it drops
• Build agents to automate parts of their work or life
• Have experience playing around with voice APIs
• Are just genuinely obsessed with AI and can show us what you’ve built
Sarvam is at the forefront of AI research and large-scale model training in India. If you want to work alongside the people building these models from the ground up and help put them in the hands of millions of users, come join us.
Apply through the links below!
