At Big Tech Company I Work At the LLM is quite happy to make raw API calls. If it thinks the data is big, then it'll write a Python tool to do it.
The reason crafted backing CLIs are useful is you can guide the LLM towards stuff that is immediately useful rather than hoping the nondetermism can separate the wheat from the chaff.
Take CI: is it interesting to know which tests passed? Maybe, but probably not. What is really interesting is what failed. Instead of having the LLM go out and talk directly to the CI system, write an intermediate CLI that filters out less actionable stuff by default, and have a flag that'll deliver the full dump if necessary.
It's a skill to do this stuff, and it's a lot of hard won experience than something I think is easily teachable. You kind of have to feel out your model and how it "thinks" about solving problems.
And then a new model version comes out and you have to learn it all again!
I think having highways turned off has really hurt pickup times and availability in South Bay right now. It pays to always book as far ahead in advance as you can.
There are places with construction where Waymo just seems to turn off the road and makes the car go around. But then taking away freeway access has made it more likely for vehicles to have to navigate these.
I worked for Uber about a decade ago. I was aware of the stories as they'd circulate internally. It's fine, until it isn't. Calling it hilarious is a lack of imagination or critical thinking at best.
They presumably know their wife better than you do, internet stranger. They did not hide that they were talking from their experience, and not from a published paper on Uber statistics.
I switched from iPhone to Pixel after I couldn't stand Liquid Glass and found myself using Gemini more than I expected.
If you're in the Google ecosystem like Gmail and Calendar, it is exceptionally refreshing to be able to use an assistant that uses that ecosystem, instead of iOS requiring you to use Mail or its own Calendar app.
I don't think there's any real gap between Pixel and iPhone on the things that matter: UX jank, battery life, camera. Even the messaging issue in the US has closed with encrytped RCS support between them launching. So now it's just an ecosystem question, which might be why Gemini is mentioned so much with Pixel.
I would love for this to be more public knowledge. I think the general public (and myself for a long time) believes the AI people know how this stuff works end to end, and so it must be trustworthy. But if we told the public "Look, we know if you put this thing in one end, you'll get something that looks similar to this out the other, but we don't really know what happens inbetween" I think we'd be able to have a more honest discussion about the relationship between AI, productivity and ongoing employment.
The majority of people in the world cannot complete a half marathon, let alone under two hours. I was pleased to train enough that I managed under three. You're doing great!
A majority of people won't run 1 km without needing a rest afterwards.
There were recent tests (in France I think) in schools where 50% or something could not run 1 km (sorry I don't have the details on mobile). These are children who have infinity energy (source: parent).
A typical adult won't make it to 1km (source: going back to sport and dying on a 2.5 km run)
> A typical adult won't make it to 1km (source: going back to sport and dying on a 2.5 km run)
I'm... dubious of this. At 39 I started running, for the first time in my life, ran 5km, didn't actually die, and went from there. I'd think most adults could manage 1km, anyway.
This really depends on people and your surroundings (I mean the kind of environment you live in, including the country)
I used to play volleyball a lot at the university, and then at a business league (we were horrendous, but anyway :)). Then I travelled a lot and let myself go.
So I started to commute to the office by bike, about 15 km each way, 5 days a week for two years. I had a puncture and decided to run the remaining distance (about 2 km) because I was late. After 500 m my lungs were on fire.
55 now, I restarted this year sport to do the smallest existing triathlon. I was already biking a bit and "just" have to do the swimming and running. I am at my 10th or so run of 2.5 km (I try to do it 2 times a week) and I am dying several times on the way. The fact I make it back home is religiously miraculous.
I am not an athlete, far from this. I am not the extreme obese either. I am somewhere in between, like are most of the people I know (of different ages). I do not see them running 1 km strong (let's say right above trotting)
Ah - and I did the whole battery of medical tests to make sure that I will not literally die when training. I am in a very good shape, heart wise.
half marathon is ~20 KM its pretty sad that a large percentage of people cant do that. One can simply walk that much distance wihtout any prior training.
You can walk it in under 4 hours. I can do it, but I can't run that much, my calves are just not trained for aerobic workloads. I don't run normally, I walk or cycle, but do not run.
I'm going to be charitable and say that the papers from prestigious universities were honest mistakes rather than paper mill university fabrications.
One thing that has bothered me for a very long time is that computer science (and I assume other scientific fields) has long since decided that English is the lingua franca, and if you don't speak it you can't be part of it. Can you imagine if being told that you could only do your research if you were able to write technical papers in a language you didn't speak, maybe even using glyphs you didn't know? It's crazy when you think about it even a little bit, but we ask it of so many. Let's not include the fact that 90% of the English-speaking population couldn't crank out a paper to the required vocabulary level anyway.
A very legitimate, not trying to cheat, use for LLMs is translation. While it would be an extremely broad and dangerous brush to paint with, I wonder if there is a correlation between English-as-a-Second (or even third)-Language authors and the hallucinations. That would indicate that they were trying to use LLMs to help craft the paper to the expected writing level. The only problem being that it sometimes mangles citations, and if you've done good work and got 25+ citations, it's easy for those errors to slip through.
I can't speak for the American universities, but remember there is no entrance exam for UK PhDs, you just require a 2:1 or 1st class bachelor's degree/masters (going straight without a masters is becoming more common) usually, which is trivial to obtain. The hard part is usually getting funding, but if you provide your own funding you can go to any university you want. They are only really hard universities to get into for a bachelors, not for masters or PhD where you are more of a money/labour source than anything else.
Yeah in principle funded PhD positions are quite competitive and as I understand it you tend to be interviewed and essentially ranked against other candidates. But I guess if you're paying for yourself to be there you'll face lower scrutiny
I see at least some of the patterns we came up with appear on the site. Happy to answer any questions about it all, I think we were the first to write about dark patterns in games, at least academically. It was 2013 so predated Overwatch loot boxes, which I am sure I would have put in there, but now they seem quite tame.
I do want to get ahead of something many of the comments here made: we were very aware that one person's dark pattern was another's benefit eg Animal Crossing's appointment mechanics make it easy to just play for a bit then put it down for the day and come back tomorrow. We went back and forth a lot about how to phrase this dichotomy, as we knew it was the stickest point of the whole plan. That's why the paper's Abstract immediately addresses it: "Game designers are typically regarded as advocates for players. However, a game creator’s interests may not align with the players’." Alignment was the key: are the players and designers in agreement, or is there tension where the designer (or, more usually nowadays, bean counters) is trying to exploit the players in some dimension?
I enjoy following academic discourse, review and collaboration give me the feeling that actual progress is being made.
So I love that you linked the rebuttal paper. In the last paragraph the authors mention that some ideas could lead to "fruitful analytic or empirical starting points" - did anyone follow up on these? From your perspective, what are the most interesting directions in this area of research today?
I honestly have no idea; I left academia 12 years ago now. I do know that game research continued (e.g. the conference I published that paper in continues: http://fdg2025.org/ and the workshop I started at ICSE continues on as well: https://sites.google.com/view/icsegasworkshop2025/home), but I'm not aware of anyone working in the patterns work right now.
My read from the paper was that Deturding was getting at in his rebuttal was my paper that was getting really popular for citing (now over 500) when really it was some Stuff Made Up By Some Guys. And it was! We all had backgrounds in pattern research, but even things like the Gang of Four are just Stuff Made Up By Some Guys. He reviewed my book that I span off from my thesis which contained the patterns so he was intimately aware of it all. We were friendly, if not capital-F friends, and I was interested in what he wrote for my academic career. He's a smart guy.
My co-authors and I never intended for the paper to be a be-all-and-end-all at 2013. Much of the non-AI research work in games at that time was "well, what if we poked at this avenue of research? what if we poked at that avenue?" And we did that by coming up with papers that were supposed to trigger conversation. It was not a good idea to go down a research avenue for 5 years only to find out no-one cared or someone had an idea that would have changed the direction dramatically had you just gotten something out there in year 1. So we thought hard about what we wrote, but we didn't do legwork tying it back to behavioral economics or something like that (my thesis attempted that to varying degrees of success).
I gave up some time ago trying to track where all the citations were coming from, but it did seem it was being cited because other people cited it. It wasn't really related to many of the papers, and certainly I didn't see anything directly building from it. And that's really what the rebuttal was saying: stop citing this paper unless you're building from it and making it more rigid in its foundations. It's not got the strong analytic/empirical basis that science is about. Which is 100% true, but was 100% known and somewhat by design.
Thanks for the insights! A bit disappointing that this avenue didn't turn out to be the one worth pursuing at the time, although I don't think the ball was completely dropped. Some light prodding surfaces recent research into dark patterns with empirical data based on player perception [1] and attempts to create frameworks for categorization [2].
I tried to report this and Uber does not make it easy. It does not fit in any of the multiple choice categories and there's no freeform. At least there was none back then, started to use waymo.
At Big Tech Company I Work At the LLM is quite happy to make raw API calls. If it thinks the data is big, then it'll write a Python tool to do it.
The reason crafted backing CLIs are useful is you can guide the LLM towards stuff that is immediately useful rather than hoping the nondetermism can separate the wheat from the chaff.
Take CI: is it interesting to know which tests passed? Maybe, but probably not. What is really interesting is what failed. Instead of having the LLM go out and talk directly to the CI system, write an intermediate CLI that filters out less actionable stuff by default, and have a flag that'll deliver the full dump if necessary.
It's a skill to do this stuff, and it's a lot of hard won experience than something I think is easily teachable. You kind of have to feel out your model and how it "thinks" about solving problems.
And then a new model version comes out and you have to learn it all again!
reply