Since such a database should evolve continuously, I wouldn't see that as a problem. The important thing is, that each example is somehow verifiable, in the form of a unmodifiable test setup. So the LLM provides a solution, which is executed against the test to verify. Something like ACID3 Tests... But sure it can be gamed somehow in probably all setups...
At this point it would be an interesting idea, to collect examples, in a form of a community database, were LLMs miserably fail. I have examples myself...
I think HN might be one of the few communities where people have been running extensive experiments with LLMs since their inception. Most here take a realistic view of their capabilities. There are certainly proven use cases where LLMs provide clear productivity gains—for example, copying an error message and retrieving potential solutions. At the same time, many recognize that marketing fantasies, such as the idea of having a "PhD in your pocket," are far beyond what this technology can deliver.
In general I would agree, however the resulting systems of such an approach tend to be "just" expensive workflow systems, which could be done with old tech as well... Where is the real need for anything LLM here?
When you argued that it saved you time by a factor of 10, have you even measured that properly? I initially also had the feeling that LLMs save me time, but in the end it didn't. I roughly compared my performance to past performance by the amount of stories done and LLMs made me slower even if I thought I am saving time...
From several month of deep work with LLMs I think they are amazing pattern matchers, but not problem solvers. They suggest a solution pattern based on their trained weights. This even could result in real solutions, e.g., when programming Tetris or so, but not when working on somewhat unique problems...
For customer projects I am happily using paid "text to speech" APIs from several vendors.
Privately, I would like to use an API, which gets my Lego bricks inventory and returns all sets which I could build. Further it should show me sets, which I almost can build and show me missing pieces. Also e.g. it should show me sets with slight color differences, e.g. I have bricks in yellow but originally it requires orange. Things like that...
I would propose that you built it as an open source software with permissive license. This can be done during work, later you could add paid extensions etc, when not working for the company anymore.
Initially, Bitcoin light wallets were built with bloom filters. So a full node would only propagate transactions, which satisfy a bloom filter, given by light client to that light client. The bloom filter seems not to be privacy preserving, that was one reason why this was abondend by some wallets. However, bitcoinj and wallets built on top of it, might still use this...
I often have the requirement, during development cycles, to bring up a static Webserver. After trying several options I always happily come back to the PHP built-in Webserver:
php -S localhost:8080
Many helpful options are available, just check the docs...
Thank you for that insightful comment! "Starting small, research and scale then" is really a pattern often overlooked these days. I wish you all the best for your future endeavours.
I also wonder about that, e.g., considering the team behind deepseek, was it more important for them to have great engineering skills vs strong math backgrounds to achieve this success?
The UI looks quite nice. I am heavily using DBeaver with various different analytical DBs. Right now I am not sure though what the built-in UI offers, which is not in DBeaver...
I'm not an expert on the economics, but as of now, Bitcoin hasn't been hacked, which suggests it lives up to its promises. Personally, I find it incredibly useful to be able to carry my wealth anywhere on Earth, just by remembering a few words.
In Germany you would mostly be forced to fill forms on paper and send it back and forth by mail, do payments in cash or use support hotlines on the phone, which take a long waiting time to get to somebody... Apps would be fine :-)
I have my own side projects, which I hoped I could make much faster progress with LLMs. Didn't work... Actually, it decreased progress. LLMs don't code themselves and explaining things to them takes an enormous amount of time and they misunderstand things so much. How could somebody think a statistical parrot can replace any kind of knowledge work?
More often I see two or maybe even more software products with the same name. In this case Apache Airflow. The two products serve a totally different purpose. I guess for marketing it would be better to have a unique name.
Actually, I thought the same about flutter. As it seem to be a huge effort to maintain and further develop it. Even Dart seems to be only used for flutter development. The technology behind flutter is complex, similar to react native. If Google or Facebook would stop support for these tools, I think, it would be highly challenging for the open source community alone to maintain it. Facebook uses RN as part of their core products. However, I don't see the incentive for Google to maintain Flutter...
Ok, but if you look into physical products. There were many successful car manufacturers since decades, still new companies like Tesla or Chinese ones enter the market... And seemingly there is money to buy these products.
There are definitely apps, which could be websites, e.g., using the coinmarkecap app doesn't make any sense in comparison to the website.
But there are many things which cannot be done sufficiently on the web, e.g., augmented reality, tracking gps in the background, accessing a users calendar or phonebook, Bluetooth, scheduled notifications, etc...
In general there is something called app fatigue. Convincing somebody to install an app is difficult these days, even if it's free... Let's see what the future brings for app development.
Yes, but native Apps have adds, which at least I don't know how to block. On Firefox mobile I can just use ublock origin to watch YouTube Videos without ads, as one example...
I am using markdown files in a private GitHub repo for personal knowledge management. For me this works just fine... Is there any feature you think I might miss from tools like these?