Post AZRwIE2ApZZ77lJb2u by frabcus@mastodon.social
(DIR) More posts by frabcus@mastodon.social
(DIR) Post #AZRONJejxVJ3ZDbtEu by simon@fedi.simonwillison.net
2023-09-05T00:54:53Z
0 likes, 0 repeats
One of the things that I'm really enjoying about working with LLMs is that I feel liberated from /syntax/I'm used to spending a substantial portion of my intellectual energy when I'm writing code thinking about the very specific syntactic rules of the programming language I'm working inWhen I'm writing with Copilot, GPT-4 or Claude I hardly spend any time thinking about that at all, which frees me up to spend more time thinking about more interesting, higher level problems instead
(DIR) Post #AZROXmvA0X47E5HfRA by simon@fedi.simonwillison.net
2023-09-05T00:56:55Z
0 likes, 0 repeats
In the past I've habitually stuck to the languages I know best - Python, JavaScript, SQL - because I'm so much more productive in a language where the syntax has been baked deep into my brainI've been spreading my wings a whole lot more over the past six months - Bash, jq, AppleScript, even Go - because syntactical minutiae is no longer a barrier for me to be productive with them
(DIR) Post #AZROz3cs0UCIr2jRCK by ogi@hachyderm.io
2023-09-05T01:01:29Z
0 likes, 0 repeats
@simon learning swift (which I haven’t had to code in a few years now) has made me a better Python dev. I always encourage folks that after you become comfortable with one language, getting familiarity with other languages is likely one of the best things you can do.
(DIR) Post #AZRPAUo3I7PwL2PsYq by luis_in_brief@social.coop
2023-09-05T01:01:41Z
0 likes, 0 repeats
@simon this makes me want to program again. But boy I fear the bootstrapping curve will still be steep.
(DIR) Post #AZRPLW9ccVUvhilFVQ by pdxmisfit@lvb.io
2023-09-05T01:03:27Z
0 likes, 0 repeats
@simon jq makes me craaaaaaaaaaaazy EVERY.SINGLE.TIME.I end up doing "operation | jq '.' | grepitty grep grep foobecause I don't grok jq… still… after all this time. 😭
(DIR) Post #AZRPXeC6YMJE7BhqmO by sankao@mas.to
2023-09-05T01:04:15Z
0 likes, 0 repeats
@simon Seconded, I have used GPT4 to write my first gdb script to dig into a core file this week, which would have never happened before.
(DIR) Post #AZRPiUV8vopREpmXke by davew@mastodon.social
2023-09-05T01:04:37Z
0 likes, 0 repeats
@simon i've had the same experience.hard to believe all the people spreading bs about these new systems.
(DIR) Post #AZRPxKtkuxAFnaNBqK by maxhawkins@social.coop
2023-09-05T01:08:44Z
0 likes, 0 repeats
@simon I’ve been learning Rust this way. Sometimes I’ll write a program in Go or C++ and ask ChatGPT to convert it to Rust to see how the idioms differ.
(DIR) Post #AZRRWZtBH41yNhTMTw by natik@hachyderm.io
2023-09-05T01:29:51Z
0 likes, 0 repeats
@simon yep, big same.
(DIR) Post #AZRTW7NTTk6frZYCXI by simon@fedi.simonwillison.net
2023-09-05T01:52:24Z
0 likes, 0 repeats
@luis_in_brief that's another thing I find exciting: LLMs tend to be really good at answering those frustrating questions about environments"Give me step by step instructions for compiling this Go code" is one I've used a few times, for example
(DIR) Post #AZRbqC8EJOyC25VKZU by Brad_Rosenheim@climatejustice.rocks
2023-09-05T02:01:23Z
0 likes, 0 repeats
@davew @simon From my understanding, using large language models to code in different languages where one may not have the syntactical knowledge is certainly like rocket fuel for getting over the learning curve. But people who really understand the syntax of those languages say that the code produced is generally inefficient and can be simplified many times over.To me, this is different than students using to pass a test in grad school with prose based answers. Here, the LLM generally wrote very well (spotless syntax), but the ideas are off, non-committal, or specifically wrong. I have heard of too many instances of the code written by an LLM just flat out not working.
(DIR) Post #AZRbqCxHFbnqaPO68e by simon@fedi.simonwillison.net
2023-09-05T03:25:32Z
0 likes, 0 repeats
@Brad_Rosenheim @davew that's one of the thing that is so great about code: if an LLM writes code that doesn't actually work you find out about it pretty fast!
(DIR) Post #AZRq1hCCYVKAv5jxmi by frabcus@mastodon.social
2023-09-05T06:04:33Z
0 likes, 0 repeats
@simon Nice! I think I always did this with search engines and copying code anyway. I think with the right IDE it would be substantially quicker with an LLM. I tend to find my time more sucked into libraries / APIs for which even GPT-4 hallucinates a touch too much for me.
(DIR) Post #AZRwIE2ApZZ77lJb2u by frabcus@mastodon.social
2023-09-05T06:08:15Z
0 likes, 0 repeats
@simon I’m saying this as I’m super curious about *why* coders give such radically varied feedback upon first trialing eg Copilot. And how day to day programmers wildly differ in the internal mental actions of how we work.
(DIR) Post #AZRwIEo1xdqXWBhodk by simon@fedi.simonwillison.net
2023-09-05T07:14:40Z
0 likes, 0 repeats
@frabcus I have a hunch that a lot of it comes down to attitudeIf you're determined not to be impressed by Copilot it's very easy to convince yourself that it's uselessIt takes quite a bit of curiosity and persistence to figure out how to get useful results out of it
(DIR) Post #AZRwWEmGraBnU5YfPE by boffbowsh@mastodon.me.uk
2023-09-05T06:52:53Z
0 likes, 0 repeats
@frabcus @simon I wonder if there’s a correlation between how well someone adopts something like Copilot and how often and how effectively they work with more junior developers. Working with Copilot is an exercise in framing, patience and feedback, similar to working with juniors.
(DIR) Post #AZRwWFS6M3eLZp84bg by simon@fedi.simonwillison.net
2023-09-05T07:15:27Z
0 likes, 0 repeats
@boffbowsh @frabcus that's really interesting - I've certainly found myself treating ChatGPT Code Interpreter as something of a digital intern, coaching it in the right direction
(DIR) Post #AZSI1yUN3e0hGnA024 by frabcus@mastodon.social
2023-09-05T11:18:28Z
0 likes, 0 repeats
@simon Would love to see a video or blog post about how you use it! Particularly Copilot's UI felt inflexible to me - no chat when I was last using it. You work round that with comments?I'm anticipating much better IDE interfaces - my sense is none are quite there yet. (e.g. letting the LLM edit multiple files easily)
(DIR) Post #AZSK6kyuuAOgjYr86K by drj@typo.social
2023-09-05T11:41:08Z
0 likes, 0 repeats
@simon @frabcus that is true of literally everything. Art, History, Biology, Exercise, Activism, Python.
(DIR) Post #AZSMoDsXXdh2iWj9BQ by nfrgrt@hejfolks.social
2023-09-05T12:11:53Z
0 likes, 0 repeats
@simon I’d say it’s an absolutely legit programming model to be embedded into products that provide value, maybe copilot and the like in your case. But all the rest reminds me so much about the crypto-alike hypetrain with only managers and politicians making the wrong calls and public media spreading fud and hype. What’s horrifying to me is not “using AI” but the erratic decisions fueled by some clever marketing tricks.