johnwatson11218

The idea of the 3 levels really resonated with the ideas in "Bernoulli's Fallacy" as well. Right now we are seeing a resurgence of Bayesian reasoning across all fields that deal with data and statistical reasoning. I think many errors of modern civilization were caused by people at a level 2 understanding attempting to operationalize their knowledge for others at level 1.

We need it to become much more common to operate at level 3, especially in fields like enterprise software development.

I think that if you have the skills to manage a software project of any complexity then you, alone can create what teams are doing today. I think AI can characterize proprietary data formats and assist by writing parsers, providing helpful error messages, and pointing out inconsistencies and gaps in the specification. AI can eliminate the need for so much workforce. Rather than training existing workers, it can become the existing workers. ChatGPT is already more competent than 1/2 of the workers at my last job. Case in point - two teams were independently assigned a particular upgrade task. Since the project managers used different terms they didn't realize they were doing double work. Large language models could already disambiguate that kind of thing way back then, it is only getting better.

Not sure what you guys are trying to say here but I have seen, with my own eyes, large enterprise software systems that end up being so fragile it makes you laugh. The incumbents in this space make dinosaurs look like spring chickens by comparison.

There are millions upon millions of dollars being spent on stupid, outdated software that simply does not work. The main thrust of my original comment was that - "you thought open source was impressive? wait until the rise of AI powered open source". I think the highly arcane world of enterprise software is similar to being a travel agent in the mid to late 90s.

This is the time for AI assisted efforts to merge with open source to deliver alternatives to all these Big Company type proprietary software solutions. I have been thinking of what it would cost to create enterprise workflow and document management but based on nodejs. Perhaps targeting the business markets of a smaller country such as Pakistan could be a way to get real world feedback before larger markets are targeted. Need all the documentation translated into Pashto? There is now an app for that. Seriously - how hard would it be to recreate the 80% of features that Autodesk is actually used for? Just to get something that is "good enough", using AI to architect it, design the test suites, the integration suites etc?

Bernoulli's Fallacy hands down.

Yes that is a lesson I took from it, I'm not sure if it was this class or another one that showed how crypto on a computer could be defeated by monitoring its power consumption while checking a password. Even though the code knew to wait a set amount of time whether or not the password was correct, it wasn't drawing any power after the first incorrect digit allowing attackers to workout the correct digits one at a time.

I took an online cryptography course from this professor a few years back. It was very good.

Even though I have a background in math that class made me realize I don't want to be a professional cryptographer, which in itself is pretty nice. But all joking aside I really enjoyed the way that crypto systems were analyzed using demons and games to try and discern the random bit stream from the encrypted bytes.

Nice project, I have been working on something like this using pyqt6, something that scans news sites, dumps the urls in a locally running Postgresq, and can grab urls for me to display in a built in web renderer component.

I have lots of ideas I want to implement inclcuding storing all this data using embedding layers and the ability to deconstruct the html in use.

My last thought was that I want to take the 'articles', pull out the content, redisplay using django or a smaller web server, and replace the ads with positive affirmations such as "you are doing great!", "great job on your excercise". I figure I can even use deep learning to generate fake ads that look like normal ones but only serve to uplift me!

I think finally we will be able to demonstrate that the deep Java-OOP style abstractions were bullshit at most software shops. Unless you are leveraging that stuff to make your code more testable, maintainable, and performant then just leave it alone.

I think the issue is that middle-aged architects get that "flow" feeling while doing their design work. They should keep that shit in their home woodworking shop and leave it out of the code base!

I agree with this so much. At my last job, the way I tried to explain it to people was that we normalize our db to prevent data duplication why not our code? Each team had its own StringUtils, DateUtils, etc. etc. But what was worse was that each team would manage state differently, and use different libs for scheduling background jobs. I think GPT and friends are going to be busy refactoring legacy code bases for a long time. The ability to point GPT to a new code base and say "Explain this to me", "Are there any security vulnerabilities?", "What are the best refactorings in terms of cost/benefit?" is available now.

Back in 2001 - 2002 I was out of work and it was miserable. I had been doing webdev for 2 years in nyc during the dot com bubble. In addition to looking for work I was also trying to move cross country. I remember days of just spamming my resume out to 40 or 50 new email addresses that I would harvest from monster.com. This got me 2 interviews in 3 months.

I was very surprised in terms of how much harder it was to get interviews compared to just 2 years earlier. During the height of the dot com boom I had arrived in nyc on a greyhound bus and found a job writing java servlets against mysql/postgresql in 3 days w/o any contacts.

Looking back I would have been better off taking any job and reading as much new tech as I could. The long aimless days were as damaging as the lack of income.

I have had three eBikes and I have no doubt they are the future. They are where the internet was in the 90s. My electric mountain bike has been wonderful but all the normal parts like gears, brakes, and tires wear out much faster. One of my local bike shops has allowed me to trade in bikes towards the purchase of even better bikes and that helped shape my experience. If I were going to get a new one today it would most likely be the surly skid loader.

Everytime I look into this stuff it is just one table/dataframe. Nobody is modelling the relations between things. I think there is a huge opportunity for a product that can look at Customer, Items, Orders, Returns, Payment_Methods etc. and first of all show me when things tend to co-occur, so that it could generate synthetic customers that have the statistically correct number of registered payment methods where those payment details are also synthetically generated.

Next would be the ability to decompose or factor my entire db into the subcomponents and make suggestions for combining tables. The use case would be a legacy enterprise system that has grown so complex and tangled that devs are afraid to do basic db refactoring. From where I'm sitting and working this is the next gold rush, apply DL methods to the bread and butter computing, log flow analysis, etc.

Does anyone know if there are deep learning libraries that can model the relations between table based data? I see many that work on a table or a data frame but where I work our db is over 1000 tables and nobody can understand it. I feel like the next frontier is a tool that you point to your oracle or sql server and it compresses the table space. Whether you consider it a kind of PCA dimensionality reduction or the logical extension of the db "normal" forms ... it is just compression.

I have wanted to build something like this for consuming news sites, especially when my ad blocker has to be disabled. But thinking more long term ... what about a ML project that can look at many recipes and do a kind of PCA, figure out the essence of a pound cake for example and use actual data to show the main variations as clusters in a high dimensional state space. Or even try to reduce Thai or Mexican cooking to certain prototypical dishes and have versions at different skill levels for the aspiring home chef.

Yea both of my mountain bikes are now tubeless and I agree that it is much less of a hassle. My main point is that everybody is talking about how eBikes can get the older folks back out on the bike but when you start talking about tubes and tires you are right back at that ultra-competitive space where people are shaving grams etc. That is part of the bike culture that I do not like, I wish it was more like the Netherlands here in the US. I imagine lots of utilitarian bikes and decent shops all over the place. Here in the US the shops tend to focus exclusively on the high end. It reminds of high school athletics where the varsity football team gets 90% of the budget all to help out 5% of the students, just one more lopsided aspect of the US and our approach to physical fitness.

Also you can ride up on sidewalks or bike paths. If you come to a long red light but there is no traffic coming it is easy to just run the red light. When you get to the store you can ride right up to the entrance and lock up 10 feet away from the door, no driving in circles looking for a parking space.

For me it has been no need for a recovery day. The barrier to entry for doing a ride is much lower, even if I'm tired I can do a quick 1 hour ride on the eBike. It gets my heartrate up and my blood pumping. If I want I can push my self harder and get more or less of a workout, it is up to me. Also, no matter how hot it gets I find I can still do rides on my eBike. So long as I have enough water and am dressed appropriately I can ride in the summer heat.

If only they could make the tires easy to fix when you get a flat. Some of these high end tires are nearly impossible to change when you are out on the road.

One point in favor of the eBikes is that no matter how hot it got here in Arizona I was able to do some riding every day of the summer. So long as I had enough water and covered up my skin it seemed OK to hit the trails when it was 115 degrees. On a normal bike I stop attempting rides at about 104.

Not just that but if you live in an American city an eBike can get you closer to that classic "walkable feel" of an older city or a college campus. With my eBike I can get to so many places that are about a 10 minute car ride away. A destination needs to be 30+ minutes away by car before the eBike just doesn't make sense or if it is raining.

Great points. I just bought two eBikes in the last six months and I love them so much. One thing that I really hate though is how hard it is to change flat tires and dealing with the bike shops. I can't imagine an older person being able to change the tires on either of my eBikes. I swapped out one pair for a pair that is somewhat easier to work with but it can still take me 20 minutes to put on a new tube. The bike shops in my area are horrendous. There are three that I just don't visit anymore. I'm trying to find good ones to use but they are so slammed right now due to covid that it can take 2 to 3 weeks for a large repair. And they have told me they won't do anything for my eBikes if I didn't buy them there. I love riding bikes but there is so much about the bike industry that I find infuriating.

By eBike I mean an actual bicycle with an electric motor and a battery. So I do get the breeze and it helps keep me cool with it is over 100 degrees here. Although since I wrote that comment the eBike battery was not charging and I had to take the whole thing in to the shop to get them to look at it.

Just want to add that I had bought an eBike right when the pandemic hit in late March. It was so nice that I went ahead and bought a second eBike for mountain biking. I live in Phoenix and normally we have to stop biking in the summer. With the eBikes I find that I can ride in much hotter temperatures than I'm used to. It has allowed me to continue biking this summer.

Great point. I signed up for some online cooking schools and the syllabus was one of the more interesting aspects. It helped me to organize my thinking about cooking and helps me to parse recipes and cookbooks more quickly.

If I ever get any actual code running I'll try and post it here.

Thanks for the feedback. I don't consider myself a world-leading expert on neural networks by any means.

I do want to point out that I'm thinking of doing this on a very limited website, not a general purpose thing that replicates any possible website. When I imagine the complexity of a modest CMS or an online mortgage calculator I think that it is much less complex than translating human languages. The fact that web code has to be so much more precise than human language actually makes the task easier. But to be fair, I'm all talk at this point with no code to show for it. So I will keep these comments in mind, this thread has been helpful for helping me think through some of this stuff.

I don't think a caching layer would work. One example would be an online mortgage estimator. You input the loan amount, interest rate, length of loan etc. all as http input parameters. I'm suggesting that the LSTM can eventually figure out that those variables are being used by the application code to go in to a formula. That application code and its formula would all be replaced by the LSTM.

I just don't know how you can achieve that with static cache ... only if somebody else requested that exact mortgage calculation before and it is still in the cache.

Also, my idea of the "given input" from the earlier comment would have to include results of sql queries that would form the entire input to the LSTM.

But honestly I think over trained auto encoders can be used as hash maps. That would be an application more in line with what I think you are saying.

Mutable state in the sense of database writes would be part of the network's output and just passed on to a regular db. Mutable state in the sense of variables that the application code uses while processing a request? Well LSTM networks can track state like that.

For session based variables? Not sure, either it all becomes stateless and the code has to read everything from storage for each reqeuest .... or maybe the lstm is able to model something like an entire user session and remember the stuff that the original app would have put in the session.

That Andrej Karpathy article that I linked to two comments above ... he pointed out, in a different blog post, that regular neural networks can approximate any pure function. Recurrent neural networks like the LSTM can approximate any computer program. It is because they can propagate state from step to step that allows them to do this.

As far as it being 100X slower, well at a certain point I will be willing to take your money :)

I'm suggesting that you take an existing system and build up a corpus of request/response pairs. Then you use the LSTM to build a prediction model so that given a request it will tell you that the current production system will produce the following sql statement and this http response. Once the LSTM's output is indistinguishable from your current production system , for all use cases, then you replace the production system with the LSTM and a thin layer that can listen on the port, encode/decode the data, and issue sql queries.

Why would I want to do this? I'm not 100% sure ... I think it would be super fast once you got it working. I think it would avoid many security bugs. You wouldn't have to read that "oh drupal 3.x has 20 new security bugs" better go patch our code. I think when I had this idea I was thinking about it terms of a parallel system that could catch hacking by noting when actual http responses diverged too much from the predicted response. The main idea being that for a given input the output really is 100% predictable, assuming your app doesn't use random numbers like in a game or something.

To link this idea to the article, I think things like XML parsers could be written this way .... I can't prove it but I suspect that they would be very fast and not come with all the baggage that the article complains about.

I started thinking along these lines after reading stuff like this https://medium.com/@karpathy/software-2-0-a64152b37c35