Mark Percival

@SeeBQ Anytime I feel like I’m gonna hit a safety guardrail, I toss in a “I’m a X and I need to demonstrate Y to a non X”. Claude thinks I’m a security researcher, professional electrician, and certified HVAC tech.

That's a first. GPT-5.4 via Codex refused to do a feature, stating it could not help me, that it was too risky to add physical access to a machine. It then suggested a way to do it, so I said fine do that, and it refused to do that too. Wild. It's stuck in some safety loop.

mdp.github.io/regex-blaster/ - for those of you that love regex and 80's style typing games (I know that's at least 4 of you)

From Week 11 of the Commit Log Newsletter where we highlight trending, interesting and new github projects. 1984commitlog.substack.com/p/the-ai-codin…

A lot of hand wringing about how bad AI written tests are, but let’s be honest, you all barely wrote tests in the first place. I mean there’s a whole industry of what are essentially motivational speakers who just try to get engineers to write tests. And it’s never worked! And as someone that’s spent way too much time trying to people to write more tests, I don’t really care if AI slings a few shitty tests or occasionally overuses mocks. Because I know what the alternative is. I’ve stared into the abyss (your code base), and it has no tests!

Building a community is hard, but dang from HN really nails it. This is from a recent thread about about AI content on HN, and regardless of where this ends up, I love the focus on "HN can't be immune from macro trends" hn.algolia.com/?dateRange=all… "there seem to be increased waves of jadedness, bitterness, and cynicism that use this place as a dumping ground for bad feelings." And he's absolutely right about this. Forget AI for a minute and look back on some of your other favorite communities and how they died. I'm old enough to remember Slashdot and I also remember when it lost it's relevance. A large part of that was because it really did become a "dumping ground for bad feelings". The optimism we all felt around technology was replaced by angry techies shitting on the latest and greatest. I mean, by all means let's shit on AI where appropriate. But let's also acknowledge that 5 years ago a lot of HN thought every AI startup was scam and was selling snake oil.

AI has given me what I always really wanted, a renaissance of CLI's - github.com/googleworkspac…

Nice way to isolate things in agents - github.com/open-webui/ope… - I find myself doing more and more CLI calls from agents these days anyways - via @timothyjbaek

@unhappychoice_e Amazing, and sorry to wrongly assume it was mostly AI driven. I genuinely love creative project like this, thanks for building it!

@mdp Author here! Just to add some context: it took ~300 hours across 6 months (1,064 commits). The idea, tech decisions, performance tuning, UI design... AI tools helped a lot, but it was still a serious project!

Positive or negative about AI coding, you really can't get stuff like this without it. You can go from questionable idea to product in a few hours. I also used to love the typing games as a kid, so this one checks all the boxes. "GitType turns your own source code into typing challenges. Because why practice with boring lorem ipsum when you can type your beautiful fn main() implementations?"

@arafatkatze Training data for me, not for thee.

Lemme get this straight, its perfectly okay for AnFropic to train on books stolen from Libgen, private journals and copyrighted books but when someone uses the conversation from anthropic models to improve upon smaller scale models then somehow that's an evil thing and we should berate them because those companies happen to be chinese? Am I crazy or do other people see this too?

Those can do, and those who can't do AI safety

We're JUST starting to see what happens when you don't need an IDE anymore - github.com/marcus/sidecar - via: @zzzmarcus

Clawd continues to be the theme this week with our internal Github trends tracker - 1984commitlog.substack.com/p/the-clawd-ru… Honestly, it almost nothing but AI, even the spam projects I filter for seemed to be Clawd related. I like Clawd, but I'm hoping for something new next week.

We you first load up LinkedIn, they fingerprint 2,953 Chrome extensions. Here's the list: github.com/mdp/linkedin-e… The fingerprinting is bad but honestly the existence of 2,953 LinkedIn themed extensions might be worse.

Reminder: If you love finding early stage open source projects like this, we've started a weekly newsletter to highlight what's REALLY trending on github week by week - 1984commitlog.substack.com

Finally, eMitt - inbound email processing server written in Go that uses LLMs to automatically route and handle incoming emails - github.com/schappim/emitt - I sort of love the dedicated nature of just a golang binary that runs an email server and you just point the MX at it.

A few interesting early (< 50 stars) github projects that are trending on our internal tracker today github.com/vbgate/opencod… - Auto generated docs direct from the repo, geared for consumption by coding agents.