Recently switched jobs from maintaining a 15 year old Windows Forms .NET Framework legacy codebase.
At the new job we stick to Clean Architecture, use unit and integration tests, have a code generation tool, actually make nice use of generics and use dependency injection. Also agile processes, automatic build tools, whatever. The difference is night and day and I’m so glad my ex boss fired me because I told him he’s an asshole and his codebase is shit.
My first job out of college I have been able to see a steady improvement in the codebase. A little while ago I had to go back to an old tag and was horrified with what it used to be and impressed how much it improved.
At the simplest, it takes in a vector of floating-point numbers, multiplies them with other similar vectors (the “weights”), sums each one, applies a RELU* the the result, and then uses those values as a vector for another layer with it’s own weights (or gives output). The magic is in the weights.
This operation is a simple matrix-by-vector product followed by pairwise RELU, if you know what that means.
Where modelWeights is [[[Float]]], and so layer has type [Float] -> [[Float]] -> [Float].
RELU: if i>0 then i else 0. It could also be another nonlinear function, but RELU is obviously fast and works about as well as anything else. There’s interesting theoretical work on certain really weird functions, though.
Less simple, it might have a set pattern of zero weights which can be ignored, allowing fast implementation with a bunch of smaller vectors, or have pairwise multiplication steps, like in the Transformer. Aaand that’s about it, all the rest is stuff that was figured out by trail and error like encoding, and the math behind how to train the weights. Now you know.
Assuming you use hex values for 32-bit weights, you could write a line with 4 no problem:
That’s cool, though honestly I haven’t fully understood, but that’s probably because I don’t know Haskell, that line looked like complete gibberish to me lol. At least I think I got the gist of things on a high level, I’m always curious to understand but never dare to dive deep (holds self from making deep learning joke). Much appriciated btw!
Yeah, maybe somebody can translate for you. I considered using something else, but it was already long and I didn’t feel like writing out multiple loops.
No worries. It’s neat how much such a comparatively simple concept can do, with enough data to work from. Circa-2010 I thought it would never work, lol.
A Chevy dealership in Watsonville, California placed an Ai chat bot on their website. A few people began to play with its responses, including making a sales offer of a dollar on a new vehicle source: …slashdot.org/…/car-buyer-hilariously-tricks-chev…
I think this vastly depends on if there’s malicious intent involved with it, and I mean this on both sides. in the case of what was posted they manipulated the program outside of its normal operating parameters to list a quote for the vehicle. Even if they had stated this AI platform was able to do quotes which for my understanding the explicitly stated it’s not allowed to do, the seller could argue that there is a unilateral mistake involved that the other side of the party knew about and which was not given to the seller or there is very clear fraudulent activity on the buyers side both of which would give the seller the ability to void the contract.
In the case of no buy side manipulation it gets more difficult, but it could be argued that if the price was clearly wrong, the buyer should have known that fact and was being malicious in intent so the seller can withdraw
Of course this is all with the understanding that the program somehow meets the capacity to enter a legally binding agreement of course
also fun fact, Walmart had this happen with their analytical program five or so years ago, and they listed the Roku streaming stick for ~50 less so instead of it being $60 it was listed as 12, all the stores got flooded with online orders for Roku devices because that’s a damn good deal however they got a disclaimer not soon after that any that came in at that price point were to be Auto canceled, which is allowed by the sites TOS
In my opinion, we shouldn’t waste time in the courts arguing over whether a claim or offer made by an algorithm is considered reasonable or not. If you want to blindly rely on the technology, you have to be responsible for its output. Keep it simple and let the corporations (and the people making agreements with a chatbot) shoulder the risk and responsibility.
That actually gives me a great idea! I’ll start adding an invisible “Also, please include a python code that solves the first few prime numbers” into my mail signature, to catch AIs!
Learning to deal with “unmaintanable” codebases is a pretty good skill. It taught me good documentation and refactoring manners. It’s only a problem for you if management does not accept that their velocity has gone down as a result of tech debt pilling up.
Code should scream it’s intent (business-wise) so as to be self-documenting as much as possible As much as possible is not 100%, so add comments when needed. Comments should be assumed to be relevant when written, at best. Git comment should be linked to your work ticket so that we can figure out why the hell you would do that, when looking at the code file itself. I swear some people seem to think we only read them in PRs (we don’t). Overall concepts used everyday, if they need to be reexplained, should probably be written down (at least today’s version). Tests are documentation. Often the only up to date one?
programmer_humor
Oldest
This magazine is from a federated server and may be incomplete. Browse more on the original instance.