Ricardo Tavares

It's reasonable to expect productivity gains in open source projects using coding agents. It's always hard to measure, but we can say by number of issues getting closed and staying closed.

The simple reason is that, if you have a project that for many years maintained code quality, relevant test coverage and helpful documentation, you have already been accepting contributions from random people of the internet. Getting value out of questionable code is something that open source already does.

🍏🐛 Bad things happen when you stop caring and profits increase:
https://www.bugsappleloves.com/

#apple #ios #macos

With its 100 billion neurons, the human brain makes 100 trillion synaptic connections. All of that for one lifetime of many experiences.

Neural networks aren't making anything close to that number of connections. But they can run those over many lifetimes in just a few weeks. An almost infinite amount of experiences. Or should we say experiments?

Maybe send this to friends and family: https://world.hey.com/ricardo.tavares/stay-safe-online-pull-don-t-get-pushed-bf7a90e1
Maybe don't let people on the internet tell you what to do 🤭

#cybersecurity #phishing #infosec

Can thinking really be defined as a flat sequence of conversational tokens?

Feels like LLMs are missing something more chaotic where different principles could be pulling the chain-of-thought in different directions which could even be contradictory.

Not sure if this can recreated with several LLMs playing different roles for that internal "monologue". How does that conversation get resolved? What hierarchies can emerge and be dissolved?

The specs for a nice app launcher aren't complicated. Index files, fuzzy find, remember frequent matches.

Spotlight on MacOS: no, can't be done.

Every other third-party launcher: sure, of course.

checking on my agents walking the dog checking on my agents catching some sun checking on my agents hugging my child checking on my agents checking on my agents

For the good folks at Wired magazine wondering if the janky Condé Nast subscription platform is harming their bottom line, my answer is subscription cancelled. The experience is so bad that I'm not even sure whether I'm still subscribed or not.

Remember when we still could say that a web page is just something you can write down on notepad? Yeah, microsoft killed that.

If QA and testing has never been a part of your company, do you think bringing in AI can magically change that or mostly highlight that weakness?

All the cool AI applications can't get away from hitting the lethal trifecta: - some way to get untrusted input - some access to sensitive data - some way to send that data away

In the midst of another weather calamity, the parliament in #Portugal may be deciding if core legislation should have terms like "climate emergency". The emergency is real, the question is if those are just empty words or if we have laws that protect our future in this planet. Environmental damage that happens now needs to be prevented, not merely compensated for eventually in the future. Line can't go up and up forever. #ptpol