ronbenton
ronbenton on Hacker News
For what it’s worth, I’m not so sure there’s much of a difference in delivered value between the hours truthfully billed vs. fraudulently billed
Anyone else romanticize going to work at some remote location like the IceCube? Probably some escapism going on
I had a similar experience with this that actually concerned me. I can’t remember the quote I was googling, but it was verbatim a quote that someone said. The AI interpreted it as my quote and furthermore interpreted it as me intending harm to someone else. Now I’m worried I’m flagged in a system and waiting for the police to show at my door next time google hands over data to the government.
A bunch of these patterns are things that real human designers have been using for a while (with the exception of emojis everywhere). This is like how some people assume an em dash automatically means AI generated text.
This is uninteresting to me given he has a significant financial interest in keeping AI research going full steam ahead.
Posted by
ronbenton on Sunday September 20, 2026 @09:02PM
from the show-hn dept.
https://pollresults.org
Not sure if this is even allowed here because of politics, but I made an API for US election poll results. NYT publishes cumulative CSV files with results on their site here (https://www.nytimes.com/interactive/polls/latest-polls.html) under a CC by 4.0 license, so my API parses those files every 10 mins and serves an API from cloudflare. I have done a bunch of random election analytics...
Inertia. It's extremely hard to build a network product and even moreso to dethrone the dominant player in the space
If an agent under a company’s control commits a crime, the company has committed that crime. If we hold companies directly responsible for the actions of their agents, we may end up seeing companies being more secure with their testing and model development.
Best case scenario they’ll try to monetize the site. Worse case it’s a malware distribution play (can eventually change legit repo link to malicious repo)
So this is a landing page made by someone who doesn’t maintain the actual tool? This strikes me as a potential scam—build some domain rep off someone else’s tool then capitalize somehow
So you’re giving it about a 10% chance this destroys us in the relatively near term?
Posted by
ronbenton on Saturday September 05, 2026 @10:07AM
from the ask-hn dept.
Or can anyone talk me down from being concerned that “goal seeking behavior at all costs” will not spiral catastrophically?
I had to programmatically generate PPT files a while back. Don’t recommend. Maybe better with LLMs now
Verbatim AI responses to my PR review comments drive me up a wall
> I dreamed of doing something significant and never wanted kids
This wasn’t a sentiment I took away from the post
Hits hard. I know this is wrong, but I often find myself resentful when I see colleagues burning the midnight oil (or otherwise putting in time that I no longer have) in service of career progression.
I think the author is spot on about ruthlessly optimizing the time you do have, though. That seems to be the only way to keep up.
> 600,000 ServiceNow tickets are another element of the collection, along with 13.7 million active emails addresses from Oracle’s Responsys marketing application, and details of 11 million sales of in-flight Wi-Fi services.
I really doubt all this stuff was “de-identified”
Anyone else have coworkers responding to your human PR review comments with AI? I feel like I'm taking crazy pills, how can anyone find this socially acceptable?
I think this is well said but also would have appreciated if you disclosed that you are a government vendor and are susceptible to your own biases. Would be only reasonable given that you call the author out on this. But the difference is she actually discloses her interest.
It was a pilot and was open to a small subset of taxpayers. It was intentionally not benefiting from economy of scale to incrementally deliver value rather than probably badly trying to hit every use case the first go around.
The cost per filing criticism is generally posited in bad faith, by people who don’t understand incremental software delivery, or both
I disagree wholeheartedly that there isn’t much to study. Direct File was actually a resounding success, especially when you consider how much difficulty governments tend to have shipping good software products. There is a lot of good information in here that government entities can learn from.
Posted by
ronbenton on Sunday August 16, 2026 @08:17PM
from the ischool.berkeley.edu dept.
Posted by
ronbenton on Sunday August 16, 2026 @01:38PM
from the ischool.berkeley.edu dept.
Ok I was confused about this because LLM seems like an ill-fitting solution for a recommendation engine. This is much more classic machine learning stuff right
Bogus query params could work maybe
Meta is extremely net evil though
So the people lagging in the commercial model game are issuing this warning? Got it.
I think a lot of people in this thread are missing the point that CEOs or other execs can’t be replaced simply because AI can do their jobs. C-suite job security comes from power structures, not task execution.
> With AI we do have the chance to return to meritocracy
I think I am more inclined to believe the opposite of this