Question for #adtech twitter. These two versions are refreshes of the same ad markup. One version clicks out to Walmart, the other to Amazon. Never seen this behavior before and curious. What's going on here? Seems it's just testing different channel performance but anything

Why you should to be sharing context with your team From “Turing’s Cathedral”

To solve birth rates someone just needs to create a car seat that takes up the entire back seat and fits 3

What a scam. Does anyone fall for these? Wish this sort of crap was just not allowed. Preying on the non tech savvy.

The TwingData team put together a guide to estimating big data workloads. It’s meant to be a quick back of the napkin exercise; use it as a starting point, then benchmark and test your workloads before committing to an approach. Some factors to consider: - Workload - Instance

I'd love to see the API usage charts of the AI models. When Claude Sonnet 4 ships do the other see as usage from Cursor et al shifts. Is it a sudden decline? Gradual decline? Bounce back? Would love to see that curve

Everyone’s talking about the liability of using the AI tools but what about the liability of not using them?

Dozens of teams have asked my advice on running LLMs. How fast is DeepSeek V3 with vLLM on 8 GPUs? What's the max throughput of Qwen 2.5 Coder with SGLang on one H100? Running & sharing benchmarks ad hoc was too slow So we built a tiny app, the LLM Engine Advisor

A few days I was able to have multiple Code Claude GitHub Actions running in parallel with no issues. Now it seems it's crashing from a single one from hitting token rate limits. Anyone else seeing this? Something change on their end?

Anyone else get the feeling that companies are just using AI as a way to jack up their prices? I'm an avid user of the AI tools but I want to choose what I pay for, not have Intuit tell me I get the privilege of paying them more for functionality I don't plan on using.

Every company past a certain size should be doing this. Not limited to software

Does anyone actually pay for annual AI subscriptions any more? The entire space moves so quick that my usage shifts from model to model and tool to tool nearly every month. Incredibly commoditized space and going after a picky, performance focused audience

Shipping the org chart, AI CLI edition:

Late night conversation made me realize we’re just going back to Unix. AI tools like Claude Code and Gemini shine when calling CLI tools: gh, ffmpeg, etc. The future? Expose everything via CLI. Let AI glue it all together. UX friction becomes an implementation detail.

As with most things it depends but generally you want the structured, repetitive data explicitly typed but also useful to have a JSON field for quickly evolving schemas, config options, and terminal nodes that don’t need to be joined to anything

We all know the one person who is responsible for this. If you don’t, it’s you

Man.. these AI app generators are supercharging phishing. I got a from email “Puma” sent from a Salesforce email address; the clickthrough landed on a Vercel app called “workwithpuma-remotely" and asked me to log in with Facebook. Reminder to stay vigilant, even more so as these

Every time I get an email from someone at a cloud provider offering help I want to respond with the Ron Swanson meme

Reminder that Nvidia and Google are publicly traded