joshka

43 ms·

571.
by joshka 4y ago
Got off the waitlist for GPT4 the other day and have been experimenting with it a bunch for coding tasks (generation, review, planning, bug fixing etc.). I prefer 3.5 for the speed and cost and move to 4 for accuracy / detail when nece
572.
by joshka 4y ago
More succinctly in the context of the article. Should GPT be used as a secure code detection mechanism? Probably not (though it can be). But it is really useful in generating better tooling to do secure code detection.
573.
by joshka 4y ago
Take a look at how Alpaca was trained with 52k instructions that were generated from GPT-3. https://replicate.com/blog/replicate-alpaca shows how to do this yourself fairly cheaply (~$50 rented compute if my math is co
574.
by joshka 4y ago
Lying is an intentional act where misleading is the purpose. LLMs don’t have a purpose. They are methods. Saying LLMs lie is like saying a recipe will give you food poisoning. It misses a step (cooking in the analogy). That step is part of
575.
by joshka 4y ago
Have you tried regularly using the ChatGPT interface from your phone? It’s obvious that the UI is written by people that don’t take UX seriously.
576.
by joshka 4y ago
You're right it can never be as precise, but LLMs might allow us to identify and highlight the imprecision easily.
577.
by joshka 4y ago
(Feel free to reply and vocalize your respose rather than just slapping a downvote)
578.
by joshka 4y ago
Hey GPT please generate tests for the code that you write that prove that all paths exhibit correct error handling behavior...
579.
by joshka 4y ago
What does Library actually mean though? ;)
580.
by joshka 4y ago
Stats from https://replicate.com/blog/replicate-alpaca suggest 90 minutes on 4 x A100/80GB. This is 3.5 hours on A100/40GB - I'm guessing LoRA is probably parallizable.
581.
by joshka 4y ago
I was thinking about https://github.com/rahuldan/codesearch for context.
582.
by joshka 4y ago
Hard disagree. I actually think the 80/20 mentioned in this article is low if done correctly. Let's think about the software engineering process for a minute and work out which bits to automate. Here's GPT3.5 on the process (
583.
by joshka 4y ago
Lol. Yup. Though it’s possible that it was overfitting its response to my noob questions. I’ve had better luck since.
584.
by joshka 4y ago
The tokio tutorial is (mostly) pretty well written and worth a read. There's a lot of good docs like rust-by-example and the rust book, but for my liking, those both seem to spend too much time on each topics rather than distilling the
585.
by joshka 4y ago
As a fellow rust n00b, one of the big problems I'm seeing with using ChatGPT to generate rust is that it often doesn't have a good model of ownership semantics. This has caused me lots of burnt time working through plausible but n
586.
by joshka 4y ago
Doing so yourself isn't too difficult... https://langchain.readthedocs.io/en/latest/modules/memory/ex...
587.
by joshka 4y ago
If it's the product I think it is (I don't recall the exact name), it's not putting the repo into GPT. It's calculating embeddings on the code in the repo, storing those in a vector db and providing context from the stor
588.
by joshka 4y ago
I'm not sure that the commenter was negging here but instead showing that the language used in the prompt was imprecise (in a similar way that we might get system requirements in reality). The takeaway I got was more that the commenter
589.
by joshka 4y ago
I've found similar success in using this approach rather than going direct to "write me some code AI-code-monkey". You can also take the following (which are the same sorts of things you'd want to do as a programmer writ
590.
by joshka 4y ago
Reagan GOP Rupert Murdoch
591.
by joshka 4y ago
Side note it’s xe or they or she rather than he. See https://xeiaso.net/pronouns
592.
by joshka 4y ago
Talk about the output rather than the process. Builds an inference. Generates a response. Evaluates possible inferences.
593.
by joshka 4y ago
Try giving working examples as input. I’d generally expect better results
594.
by joshka 4y ago
Markdown (just the bits that makes this better).
595.
by joshka 4y ago
Talk to former employees. Connect via linked in
596.
by joshka 4y ago
I love this idea. I did a related experiment[1] about a month ago with GPT-3 where I started with a stub that takes a prompt and self modifies the existing code. import os import openai instruction = input("E
597.
by joshka 4y ago
There's a good Bjarne Stroustrup quote (mentioned in Clean Code) that is more useful here (as well as a bunch more on optimization / efficiency that are worth reading): "I like my code to be elegant and efficient. The logic s
598.
by joshka 4y ago
I'm not unaware of this. It however doesn’t fit all my use cases and hence the price is still worth it.
599.
by joshka 4y ago
A more specific example is the idea of sitting down and listing all the attributes that define a customer profile (name, addresses, date of birth, etc) and then designing the CRUD around those things. Contrast this to starting with what are
600.
by joshka 4y ago
I've seen a lot of different enterprise systems. In my experience it's not the db that's causes this inflexibility issue, but designing the architecture of systems data-first (using Data Dictionaries / ER diagrams etc.)

More ›