npilk

Multimodal AI will lead to an interesting arms race in ad detection vs ad insertion. I played around with AI ad removal with older Gemini models, but it seems like this would be even more powerful to instantly identify ads (and potentially mute or strip them out).

https://notes.npilk.com/experiments-with-ai-adblock

Can you not make two user accounts on a Mac and sign them into different iCloud accounts? I know you're limited to one on iPhone and iPad, though.

How much longer will this be true, though? With improving computer use, it may be possible in the next ~year or so that agents will be able to wire up infrastructure and launch to production.

I believe EVs also wear tires out faster (because they are heavier), so they need more frequent replacement.

So they are comparing to the conversion rate of people who click on a link in the chat and go to Walmart's website to view the product? Wouldn't that be a really strong intent-to-buy signal?

The better comparison might be conversion rate for those who searched on Walmart.com vs those who searched within ChatGPT. Or maybe that is what they're comparing and I misunderstood?

Yes, spending time working with Claude Code leaves me feeling the same way I feel after a day scrolling Reddit and HN - a thin, jittery, frayed sort of weariness. It's almost like gambling, with inconsistent dopamine hits, but it adds an element of keeping track of an ever-increasing number of projects and to-dos.

Don't we already have these filters in place? I only saw this because it was highly-upvoted on HN, for example - I don't read every new submission. I also read things sent by friends and family, shared by curators I trust, etc.

Of course these systems may eventually break down, but for now they seem to work.

By the way, for anyone wondering/unfamiliar, the scale of this tournament is such that it's not realistically possible to enumerate all possible outcomes, let alone submit them to a site like this. With 63 games, there are 2^63 possible brackets, and it takes 63 bits to encode each possibility.

2^63 brackets * 8 bytes/bracket ~= 74 exabytes - just to list all possible combinations!

There are many combinations that are completely unlikely, but even if you could reduce this by 90% (I doubt it) it's still infeasible to even list all the combinations.

Someday, in another 20-30 years, this might be achievable. Somehow I feel like it will be a sad day when that happens. Of course the tournament will probably have expanded to 128 by then making it safely out of reach of computation.

For sure it was overkill/not the most efficient approach - really I was more just curious if it would work. The answer was "kind of", but even that is pretty amazing. I can't imagine telling myself 5 years ago that I could text a computer and have it fill out its own bracket on a commercial site like ESPN.

Love it! Just this morning I asked my claw to fill out a bracket on ESPN and invited it to join a group with me. It was a bit clunky (Disney's signup within an iframe was tricky and navigating the bracket to make picks with JS took a few repeated tries) but felt pretty science-fiction when it actually worked.

Fun project but I need a downvote button - Secret History is great! Incredibly rich and interesting characters.

I find this happens naturally on high-trust teams highly motivated by their work, and I doubt just asking for it like this will be effective.

The author is saying something different here - that in this mode, the speaker’s feelings about how the recipient will receive a blunt message are the speaker’s problem.

In this case the recipient has already reassured the speaker they can handle their own feelings, but still meets resistance from speakers who are guilty or worried about how they will come across if they are too direct.

What's odd is according to the article, this index estimated an ~8% default rate in 2024. So maybe the stress test was measuring something different? It's weird to think the stress test would find a lower loss rate during a severe recession than in the most recent year with data available.

It's some sort of Gell-Mann-Amnesia-like effect. I am accustomed to seeing thoughtful, informed discussion about technical topics on HN, so then it's jarring when something like this hits the front page and nobody seems to have any idea what they're talking about.

This doesn't seem too far off from ~80 characters per line, which I believe is best practice for readability. Though you could make the column wider and bump up the font size and it would be even more readable.

It feels like building humanoid robots so they can use tools built for human hands. Not clear if it will pay off, but if it does then you get a bunch of flexibility across any task "for free".

Of course APIs and CLIs also exist, but they don't necessarily have feature parity, so more development would be needed. Maybe that's the future though since code generation is so good - use AI to build scaffolding for agent interaction into every product.

With LLM-based tools that inherently rely so much on the semantics of language, I wonder if there will be differences in code generated for the "wanted board" in the "Wasteland", compared to the "task list" in the "public square" or the "wish list" in the "Utopia".

Infinite Jest describes a very similar (fictional) development, albeit with network TV. As viewers leave, content producers are ever-more desperate to monetize remaining traffic, which worsens the experience and drives more viewers away, creating even more desperation to monetize... a vicious cycle.

Just because G&A was up $68m doesn't mean it was all spent on that one party...

Edit: never mind, the report clarifies that without the party expense G&A would have been flat YoY.

@dang would there be any possibility of creating a view that hides posts and comments by accounts newer than, say, Jan 1 2026? Similar to how https://news.ycombinator.com/classic works (only showing votes from the oldest accounts)?

I know this is unfair to prospective new community members, but I'm unsure of other good methods to filter out AI bots at scale. Would certainly welcome other ideas.

For what it's worth, your original comment is a little hard to parse, particularly because you say "our pilot deployment" which makes it sound like you were involved in deploying the cameras. Combined with your realpolitik comment about knowing your neighbors want them, I think several people are confused about your opinions and what you ended up doing to fight the cameras.

This is cool, I love the concept.

I wonder how much our understanding of past language is affected by survivorship bias? Most text would have been written by a highly-educated elite, and most of what survives is what we have valued and prized over the centuries.

For instance, this line in the 1800s passage:

> Hunger, that great leveller, makes philosophers of us all, and renders even the meanest dish agreeable.

This definitely sounds like the 1800s to me, but part of that is the romance of the idea expressed. I wonder what Twitter would have been like back then, for instance, especially if the illiterate had speech-to-text.

So about 1 in 10? Doesn’t seem that terrible to me. Especially when many of them are in response to questions about his work, and he’s answering with a link to a different post.

I think 7 or 8 out of 10 would be a bad look.

Fair enough - I didn’t turn the sound on for the video. The premise of the sketch seemed to fit the rest of the slimy stuff in the feed, which is why I figured that video in particular got recommended.

I mostly included it because of the absurd question Meta suggested I ask their AI.

Interesting. I wonder if the phrasing as "do you ever use this platform" leads to this result. I could definitely believe more 30-49 year olds log in every so often for Marketplace etc., but would expect DAU to be lower than 50+. But maybe that's just more of the same bias you describe.

Well, I haven't really used any Meta platforms for at least 5 years, so I don't think that's how they're deciding what to serve me.

I could definitely believe that I used to click on more pictures of girls than boys back in high school and college when I actually used Facebook. But they would have been real pictures of people I was friends with.

To your point, I'm sure if I used the product more, the algorithm would get "better" according to what I engaged with.

Yeah, this makes sense. It does sort of imply that new users would just see a bunch of garbage, which you'd think isn't ideal. On the other hand, how many new users could possibly still be signing up for Facebook? So maybe it's not a problem as they just manage the decline.

The opposite issue exists in men's clothing - it seems to be harder to find clothes that fit shorter/smaller men. "Medium" sizes tend to be quite large.

I made a page about this myself - Model is 6'2" Wearing Size Medium - https://pilk.website/2/model-is-6-2-wearing-size-medium

Curious how you generate the maps. OpenStreetMap turns out to have most golf course features mapped by volunteers, so it's possible to pull that data (see here for a fairly rudimentary example: https://github.com/npilk/hacker-yardage). But you may have a better source.

I see you mention looking for topographical data at a good resolution. I'll follow that with interest. There are US public datasets (IIRC) at more like 3m or 5m resolution... not bad but not really good enough for golf, especially greenside. I couldn't figure out how to pull that info programmatically in the past, though.

Overall this is super cool work, will be watching for more!