@AdamChainz These things are a joy if writing tests to start with but worth keeping in mind the indirection they create can be hard to follow for the next person. You want tests to be as low friction to add to and work with as possible, as mentally satisfying as it can be to do otherwise.
@beatport Site loads each time but (1) I get random £ or $ or € for prices each refresh. (2) Can add things to cart but not pay at check out as gives me instant JS error alert. (3) If I filter for one subgenre in track list it filters on a different one. (2/2)
New @beatport site is so broken. It made me think about how modern javascript sites broken is so much more unpleasant than old fashioned server side broken as I can sort of use bits of the site but it’s a horrible experience. (1/2)
This thread is an incredible insight.
Because we trained GPT on code, it means we’ve trained it on billions of individual micro-logical predicate sequences, so it “knows” logic by thinking in code.
@AdamChainz Hah, quite. "soon".
At the mo I think it's right about 90% of the time, but the bugs in that other 10% are harder to spot as its someone else's code.
Would think that if you have some code that functions correctly, even if unmaintainable (old libs / just a mess) at some point soon you'll be able to have ChatGPT/github's upcoming tools tidy it up for you and make it beautiful in the style of your own code (or your fave dev).
Would also be able to document for you. Exciting for both our own old abandoned projects but also discarded open source projects. Obvs don't be lazy and get the AI to write tests for you first,.
Can imagine something evolving out of ChatGPT for code refactoring being a game changer for that situation where you have an inherited project that is "kind of working in prod" but near impossible to maintain because it's a mess under the hood.
@ThomasSimonini@huggingface Ha oh wow so you do! I've not found NLP and diffusion models as compelling as RL to get under the hood of, despite the fruits of ChatGPT/Stable Diffusion. My next plan is to try wielding RL at a couple of things I'd had in mind. Hopefully will have something interesting to share.
Just finished the @huggingface deep reinforcement learning course. Good fun. Having a track laid out one chapter at a time reminded me of a time when buying a new tech book and doing that was how we learnt stuff. Cheers @ThomasSimonini
@ESYudkowsky “When it starts threatening people & writing stories about it taking over the world & exterminating mankind we’ll definitely hit the button & shut it down til we can figure out what’s happening.”
Reality: “LOL that’s hilarious, let’s post more articles & tell the AI to read them”
Github Copilot (and I suppose ChatGPT) is a bit like that person you worked with who you know bullshits half the time but speaks with such confident gusto you can't help falling for it again.
To be fair, also delivers the goods lots too.
Never makes the tea though.
Twitter is great because it gives you unfiltered visibility into the thoughts and decision-making process of some of the most successful people in the world. This has completely cured my impostor syndrome.
Pondering if @elonmusk's Twitter poll is so he can test out his new autonomous tesla killer bee bots.
Anyone who voted yes is going to find themselves attacked by a swarm like in that Black Mirror episode.
Next year anyway, when it's ready. 😜