kuil009

This feels a lot like the old RPA hype cycle to me — more sales narrative than structural change.

Most companies are not going to replace stable SaaS with a pile of AI-generated internal tools. They don’t want the maintenance or the risk.

If there’s a real B2B game changer, it’s Microsoft.

The day Excel gets a serious, domain-aware AI that can actually model workflows, clean data, and automate logic properly, half of these “build vs buy” debates disappear. People will just solve problems where they already work.

Excel has always been the real business platform. AI will just double down on that, not kill SaaS.

Thanks for this. It put into words a lot of the discomfort I’ve had with the current AI economics.

The positioning makes sense, but I’m still somewhat skeptical.

Targeting power, cooling, and TCO limits for inference is real, especially in air-cooled data centers.

But the benchmarks shown are narrow, and it’s unclear how well this generalizes across models and mixed production workloads. GPUs are inefficient here, but their flexibility still matters.

As someone who enjoyed 'Dilbert' at times long ago, I offer my condolences with a sense of appreciation for the work itself

Rather than treating SRS as a learning tool for facts, I find it far more valuable as a system for recording and periodically revisiting past judgments, especially to reflect on whether a decision made in context was actually a good one.

Thank you for your service

I use LLMs mainly as a mirror for my own thinking, not as a source of authority.

When I explain my ideas to the model during development, I often see flaws or confusion in my own words. This is where I learn the most. The author talks about people who rely on AI for arguments or research. They let the model's smooth, but statistical, language replace their own thinking. Language is naturally uncertain. LLMs just show this uncertainty using statistics. If you understand this, LLMs are no longer a "confidence engine." Instead, they become a tool to fix and improve your thoughts.

A key point is that even if we try hard, we cannot help but react to what the AI says. We must remember that neither AI nor humans are perfect. I believe we should accept AI responses critically and always be skeptical, just like when meeting a stranger.

see also: https://cacm.acm.org/research/automatically-translating-c-to...

눈물나게 감동적이었습니다.

OMG

I love this idea. :)

I enjoyed reading this. Among the points discussed, the one about structures that can be read by machines (machine-readable formats) seems to hold important implications for the future of data structures and representation.

While there are many attempts being made today, at the current pace of change I can imagine us returning to a clean, well-structured, text-based origin. As for UI representation, I believe AI will likely handle that on its own.

Slack is experiencing a widespread outage affecting message delivery, thread loading, and app functionality. Issue started around 3:36PM PT. Slack has identified the cause as a backend database routing issue. Status updates: https://slack-status.com/2025-05/7b32241eb41a54aa

Garbage in, garbage out. Like it or not, even someone’s trashy intentions can run exactly as designed — so I guess we’ve had the same experience.

It's natural to expect reliability from AI agents — but I don't think Cursor is a fair example. It's a developer tool deeply integrated with git, where every action can have serious consequences, as in any software development context.

Rather than blaming the agent, we should recognize that this behavior is expected. It’s not that AI is uniquely flawed — it's that we're automating a class of human communication problems that already exist.

This is less about broken tools and more about adjusting our expectations. Just like hunters had to learn how to manage gunpowder weapons after using bows, we’re now figuring out how to responsibly wield this new power.

After all, when something works exactly as intended, we already have a word for that: software.

The GitHub repository line/abc-virtual-background-maker is a standalone React application designed for creating and downloading virtual background images enriched with customizable text elements1. It has the following features:

Customizable text overlay: You can add your name, company name, or other information for video calls.

Adjustable fonts and text placement: You can customize fonts and text field positions according to your preferences.

Organizable themes: You can create collections of background images, similar to folders, for easy management.

Image storage option: You can manage your background images on the GitHub repo or load them from a local directory.

Mirror image option: You can download a reversed version of the image to accommodate video feeds that flip your background horizontally.

Image upload capability: You can also upload your own picture to create personalized virtual backgrounds.

Customizable appearance: You can change the UI with your organization’s key color and name for a consistent look.

The app is easy to use and does not require any backend setup. You can find more details on building and running the app in the README.md file.