redeyedtreefrog
redeyedtreefrog on Hacker News
https://vgcgroup.co.uk/news/signalbox-for-train-locations/
suggests the data mostly comes from railway signalling information, plus a bit of "AI" in some way. I wonder how far apart railway signals usually are, or what the AI is trained on, or anything really vs just looking at a map.
Meanwhile social games are fun even if they are extremely simple - rules-lite RPGs, Guess Who, Pictionary, etc.
This seems to exist at an intersection of the two. It's interesting that following a mindless algorithm with a group of people sounds so ridiculous and pointless in principle, but it's actually fun. Our brains are somehow wired to find it rewarding.
Extreme luck has the benefit of bringing everyone to the same level regardless of age or background, whereas e.g. chess, Counter-Strike or boxing is much less fun if one person is 10x better than the other. The more skill a game requires, the more it needs separate leagues for differing levels of ability.
Maybe there's also ironic enjoyment in playing a horse-race gambling game when most people playing have likely never watched a real horse race, let alone bet on one.
Personally I generally enjoy the blog and the writing, but not so much this post. It has a very clickbaity title for some results which aren't particularly impressive.
Previously: make a little bit of progress, realize I'd need a year's worth of weekends to complete it, give up after one day.
With vibe-coding assistance: make 10x+ progress, realize that with a few more weekends I could finish it, keep going.
I'm having fun with Claude Code and Vibe Kanban on personal projects, and before that I spent a lot of time with both the Windsurf and Cursor agents. It's making me literally 10x more productive on personal projects, maybe even 50x.
On personal projects:
- no-one but me needs to decide on requirements
- no-one but me needs to make decisions
- no-one but me needs to maintain the code going forward
- much of the time I'm intentionally using languages and frameworks that I am somewhat clueless about, and an LLM providing continuous ideas (even if sometimes entirely silly ones) stops me getting stuck
- I don't mind if there are large chunks of useless or half-working code
On commercial projects:
- every line of code is a massive liability. Every line needs to be reviewed by another engineer, and every developer who joins the project needs to be aware of it, take it into consideration when making changes elsewhere, and potentially debug it if something goes wrong
- senior engineers are almost always hired to work with languages and technologies they are already very familiar with, meaning for many tasks it's often quicker to write out the code by hand (or perhaps with Cursor's auto-complete) than guide an LLM to do it
- much of the time is spent in meetings trying to unearth the real product requirements or providing updates to stakeholders
- much of the time is spent reading old code and working out how to implement things in extremely large and complex systems in a minimally disruptive way
- a lot of time is spent reviewing other people's PRs, and getting infuriated when people (often either very junior or very senior) produce 1000 line PRs consisting of unnecessary changes, excessive boilerplate, half-finished experiments, and things that clearly haven't been tested properly. This was the case long before LLMs, AI just makes it ever more tempting for people to act this way.
- trying to avoid or gently negotiate political games over who is in control of the project, or who gets to makes technical decisions
www.linkedin.com##[aria-label="Main Feed"]
I go absolutely out of my way to avoid politics nowadays, which makes it all the more frustrating when this nonsense is shoved in my face by idiots on HackerNews or people dimwits sitting next to me on the plane.
https://www.alltext.nyc/panorama/z0SOvmU-5_yuspnsFvjVuA?o=16...
I don't believe the women entirely made it up, or that Jon Pretty is entirely guilt-free. Likely he is a narcissist who took advantage of his status to pursue sexual relationships where there was a huge imbalance of power. Maybe this strayed into manipulative gaslighting, I don't know. But it also seems entirely plausible that the women in question desired a relationship with a powerful older man and that the relationships were essentially consensual. It's a mess of power, sex, alcohol, a lack of shared social norms, and overlapping social and professional relationships. Quite where the truth lies between "totally non-consensual gaslighting" and "consensual relationship with large imbalance of power" I don't claim to know.
https://www.theguardian.com/world/2023/oct/31/from-the-river...
The people behind AI Dungeon have been making AID: Heroes for ages, maybe that will turn out to be great when it's finally released. There's also Friends and Fables that seems to be making rapid progress. And a thousand side projects on GitHub and Steam.
I briefly started my own, hoping to build something more like a traditional visual novel choose-your-own adventure with set paths, dice rolls, and limited AI elements. However, I soon realised that if you show people a large amount of text and only occasional user interaction they rapidly check out. The alpha version is here for the time being: https://prismaticnoun.xyz but I'm unlikely to do any more work on it.
It's meaningless, and I suspect intentionally so. It clearly implies that they were doing things in a more advanced way than a standard Indian outsourcing firm, an implication heightened by hiring people to build impressive tech demos. But the impressive tech demos and the work on real client projects were two entirely separate things.
As far as I know all the development work on client projects was done by developers in India. I was too far removed from the details to know quite how they were employed and paid - I literally never had any interaction with them - but I suspect this is the main source of the LinkedIn employees.
As far as I know customers were aware apps were being built by human developers, the "80% by AI" was prominent on the website and the sign up process but it was never claimed the whole thing was automated.
It looks like Craig Saunders joined only a year ago, in June 2024, which is long after I left. Given how much AI has advanced over the last few years, and given his LinkedIn shows he previously held a senior AI role at Amazon, I wouldn't be remotely surprised if he and other employees built some impressive tech demos over the last year or two.
Regarding Natasha assigning developers to projects, I don't know to what extent it was really automated, I wasn't involved in that.
In 2019 its former Chief Business Officer sued them for fraud, claiming that apps were built by Indian developers despite the claim that it was "80%" done by AI. There were detailed articles in the Wall Street Journal and The Verge at the time. I've found one reference saying it was settled out of court (the Telegraph), though I thought I'd previously read the case was dismissed.
When I was there 3 years later:
- the company had been renamed from Engineer.ai to Builder.ai
- the marketing materials still heavily pushed the claim that apps were 80% built by AI, curiously the exact same figure as it had been 3 years earlier
- there was a bunch of automation around small parts of the software development process. When customers went to create a new project there was an AI chatbot assistant (Natasha), which amongst other things asked the customer for their requirements, and created some estimates for how long things would take. There was also some automation for turning UI mockups into CSS styles and then merging the styles with templated React components. These various small bits of automation did have teams of real engineers working on it in the UK and the US. However, by and large it didn't really work. It seemed to me that the Indian outsourced programmers working on client projects totally ignored this technology anyway, and just went about their jobs as though it didn't exist. Despite being employed to work on this tech, I never had any interaction with the Indian developers building real client projects.
- a new team was spun up to use Generative AI to create frontend mock ups of an application from template components. This was integrated into the flow when potential clients were chatting with the Natasha AI chatbot. Some people worked genuinely hard on this project, and to some extent it did function. However, it didn't do much beyond the many other "create a frontend mockup from a single prompt" projects out there, other than having access to the company's internal React component templates. As far as I'm aware these frontend mockups were never used by the Indian developers who built the final projects.
In the tech world there is a very wide blurred zone between "outright fraud, which leads to a conviction in court" and "exagerrated claims". Given the extent to which Builder.ai committed literal fraud with their revenue figures and accounting, and given the hundreds of millions of dollars they raised on very limited real sales, I believe their claims about AI could plausibly also be literal fraud. However, the standards of a court case are much higher than my personal use of the word "fraud". I'm tempted to believe Musk is also bordering on fraud with e.g. the Boring Company, or his repeated claims about how close Tesla is to fully automated self-driving robo taxis. But given Tesla is building a genuine product with genuinely high sales and revenue, and given how much other crazy stuff Musk does that makes these exagerrated claims pale in comparison, it's not something that's ever going to lead to a court case.
a) the keyboard doesn't work until after I click to fire, and this state resets itself every time I press f5 to restart the game (presumably a browser focus issue)
b) both up/down and w/s work for back and forward, but a and d don't work for left and right, you need to use the left/right arrows.
The combination of the two had me scratching my head for a little while!
Their terms of service say "The DeepSeek Open Platform is jointly owned and operated by Hangzhou DeepSeek Artificial Intelligence Co., Ltd., Beijing DeepSeek Artificial Intelligence Co., Ltd. "
And they're funded by https://www.high-flyer.cn/en/fund/ which the FT did an article on: https://www.ft.com/content/357f3c68-b866-4c2e-b678-0d075051a...
In terms of the personal data you share when using their models, I can't see why they would be any more or less nefarious than big Western tech companies.
That said, if you're using a model based in China then by providing them with data and feedback you are in a very small way helping researchers in China catch up with/keep up with/overtake researchers in the West. Maybe in the long term that could end badly. And if you are concerned about the environment, it's entirely possible their training and inference is run using coal power stations.
However, AFAIK it's only ever at inference time, an interpreter isn't included during LLM training? I wonder if it would be possible to fine tune a model for coding with an interpreter. Though if noone has done it yet there is presumably a good reason why not.
> Job Openings vs. Interest Rates
Everyone already knows this
> Company Level to Compensation Scale
Everyone already knows that over the last 20 years employees at LASAGNAFAANG big tech companies in the US have made absurd amounts of money relative to how much skill is actually required. If you work in tech in San Francisco it must be easy to compare yourself to these people and be envious all the time. Outside of the US (or outside of tech) it's more normal to view these jobs as a weird aberration, not the norm. Why not compare yourself to public school teachers, lorry drivers, nurses, enlisted infantry soldiers, etc? Many of them could probably have earned absurd amounts in Silicon Valley too, had they been in exactly the right place at exactly the right time.
> Coding interviews
If you are failing leet code interviews and you're unemployed then why not just spend a week or two practicing leet code questions? There are some egregious examples of small companies using algorithms questions on graphs as a hazing exercise, but most places are just using easy or medium coding tests as a basic filter on people who are able to do arbitrary programming tasks whilst communicating their thought process. If you have US citizenship and enough experience to get an interview, you are already extremely lucky. Whining that getting a job also requires practicing a few coding questions because you see it as beneath you is silly.
> "We find the midwit problem in job interviews all the time where interviewers think they are “elite special evaluators” needing to gatekeep the unwashed hoards of desperate candidates, but interviewers often can’t reliably judge or measure people who have better answers than they expect"
Isn't that exactly what standard coding tests avoid? People who think they are elite special evaluators ask off-the-wall questions about particular technologies or processes, or abstract logic questions, thinking they are so clever they can infer the inner workings of the candidates mind. The low IQ and high IQ solution is to realise that as an interviewer you aren't very special, and that assessing candidates in the space of an hour or two can never be more than a very lossy, rough filter. Under these circumstances, generic coding tests aren't such a bad idea.
> Field Report: Job Experience Notes
My experience of work have been similar. That's just what it's like at the average mediocre tech job. Hence the The Daily WTF running since 2004
> "At some point, a switch flipped in the tech job market and “programmer jobs” just turned into zero-agency task-by-task roles working on other people’s ideas under other people’s priorities to accomplish other people’s goals"
Yup I agree it sucks, but it pays the rent.
wow the job market sucks right now