gentleman sysadmin, aspiring post-economist

Menlo Park, CA
Joined July 2007
Effective Immediately, all organizations using PHP for their systems are advised to be prepared for an update addressing two critical vulnerabilities. PHP is widely used for web applications. These vulnerabilities often provide a direct path to the server's file system or database. If an immediate update is not feasible, it is recommend that organizations take the system offline until it can be fully mitigated. NSA will follow up shortly with more specific guidance.
1,297
10,837
2,839
70,331
12,576,817
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable. 🔗 github.com/deepseek-ai/DeepI… 1/4
1,652
4,901
2,338
43,370
6,811,898
Also another big thing that's surprising: all the "rent versus buy GPU" charts, the x-axis is always the GPU utilization rate. makes sense rationally, but in a way I think owning compute capacity should be like holding capital you're always getting interest on your capital
1
88
Also, are there well-known evals out there for multi-agent coordination?
1
31
use.sh: provision dedicated cloud resources on the public internet, in under 1 second:
2
11
183,018
Looking for an XML schema / standard to represent projects: vision, collaboration, metrics, review. Does something like this exist yet?
3
1
241
If you take a risk where there's a 10% chance of winning -> there's a 90% chance you won't make it. but if you try twice? 19% chance you'll make it. seven times? 52%
1
3
393
Median IPv4 sales: Jan 2020: $20.50 (139k addresses, $2.8M) Jan 2021: $25.00 (93k, $2.2M) Jan 2022: $51.00 (59k, $3.1M) Jan 2023: $44.50 (64k, $2.7M) Jan 2024: $34.50 (74k, $2.5M) Jan 2025: $34.00 (84k, $2.7M) Curious if anyone has looked into this market?
4
5
393
Top buyers: Amazon (190M addresses) Microsoft (75M addresses) Google (25M) Alibaba Cloud (14M) Sellers: DuPont Inc (22M) T-Mobile (20M) GE (18M) Haliburton (17M) ipv4.omarish.com/parties/
59
We're in the largest capex cycle in human history and people are beginning to wonder if there should be cause for concern, and if we should worry about a "2008 for data center buildout." Some notes on 1) why we should be cautious, but also, 2) why this time it might be different: Reasons to be cautious: - Hyperscalers are financing using debt instead of free cash flow and some question marks 1) future demand, 2) power constraints, the capital markets because 3) the 30-year is above 5.5% - "is this the opening theme to black monday?" If someone can make 5% clipping coupons, why take on equity exposure with multiples already has high as they are? - Should we be worried about 2008 for data center buildouts? One argument why "this time it's different," or at least, some reasons why depending on deal structure, distressed compute assets might be different than distressed real estate: - In 2008, subprime housing + the surrounding contagion continued because there is only so much demand for housing in a given city. a house in one city is fundamentally different than a house in another city. - And housing -> jobs -> local economy, but with compute, it might be easier to recoup value. - Deal structure and underwriting. it's hard to package and ship a foreclosed home in a depressed market. and a home takes a long time to pay for itself. but compute, not so much. this doesn't apply to assets that can't move - A house is a house - you can re-zone but the market for RRE in a given market doesn't change that much - people generally only need one home - You can do other things with compute assets acquired at a discount. - Compute is more fungible than housing: core chips and components can be repossessed, repurposed, or shipped anywhere. demand for rare earth metals, components, parts, etc. - Compute has more liquid, global markets, we're beginning to see spot markets and curves for future compute capacity - Compute assets can be broken down into smaller pieces: i.e whereas there isn't demand for 1/1000th of a repossessed home, I and others probably have a bid for fractional components of a data center build-out. - If tomorrow someone told me that a lender is selling distressed compute assets at foreclosure/discount, my due diligence would largely consist of getting wire instructions. A home is a home, a data center is a lot of lego pieces, and lego pieces could carry stronger residual value because 1) they require more specialization to build, 2) supply is controlled by only a few companies, and 3) they're useful elsewhere.
3
271
3 minute time drift: Just enabled NTP on one of the gateways, and wow, it's easy to forget how hard it is to keep track of time accurately and independently. uptime on this router is about ~315 days.
4
316
Omar retweeted
tokdump: hexdump for tokens Example: how o200k_base splits your text github.com/omarish/tokdump brew install omarish/tap/tokdump
3
1
9
551