I don't feel like it's any faster for things I'm not really already familiar with. For instance I asked it to write me a Makefile. It wrote one. It looked plausible but I don't know enough about Make to know. So I had to do loads of reading about Make just to verify that the AI answer was correct. Basically the same amount as just learning anyway.
yup just literally talked a client out of a nice little mobile project by telling him my rate. lets say I had AI by my side - I'd be quoting the same rate, but number of hours would be lower. project might be a go then.
Then youre wrong. There is no market incentive for layoffs due to GPT-like technology, as I have demonstrated above. Similar breakthroughs "replacing jobs" have happened before in the field of software engineering, this is nothing new or unprecedented. Its merely another tool that will become in widespread use to increase production.
What i think will lead to mass layoffs is the current recession rather.
yes - and in another way, too. a lot of the demand for programmers is driven by successive waves of new technology adoption and investment - and AI is looking to be a motherload that should keep us going for awhile.
> If the language model is large enough to contain the entirety of the documentation and the LSP itself, then why bother integrating with the LSP?
If your goal is to get a response to an LSP query, why on earth would you use an LLM trained on data where >99.9999% of that data has nothing to do with answering an LSP query?
Why would I switch out an LSP server of 100% accuracy for an LLM that’s slower and has lower accuracy?
Imagine an LLM tuned to eliminate misunderstanding and ask why at least 5 levels deep… Without fearing to irritate the boss or to create an impression of being not smart, both possibly harmful for human career but irrelevant to unthinking software tool.
I too like science fiction. People keep acting like it will be easy to bolt on things like eliminate misunderstandings onto LLMs and quite frankly I would be incredibly surprised if that happens any time soon.
Eliminating misunderstanding comes down to willingness to ask more questions if you have low confidence. The main reason this doesn’t happen is subordinates afraid to look stupid or lose jobs. None are concerns to an unthinking machine.
Maybe some day. But I tried it just now on a database design I‘ve been working on for two months and it spits out something superficially close immediately from a two sentence prompt. On one hand that’s impressive, it‘s interesting and somewhat correct but all the interesting parts are missing or wrong and it never get‘s beyond that, not even with my help. No sane person would answer so confidently yet superficially useless.
A sane approach would be to start understanding the requirements and work from there, trying to figure out where the challenges are.
> capable of at-least C++ "constexpr"-style compile-time computation, which shouldn't even be possible if one presumes GPT is "just" a giant database storing only multidimensional word similarity scores and sequence distribution from text inference
I don't see how being a giant word-DB necessarily disqualifies compile-time computation. You can view computation as applying a series of term rewrite rules to an input until some sort of termination condition (or indefinite loop). In the case of these AI, the input is the prompt and predicting the next token is a limited form of term rewriting (where the rules are probabilistic and based off the network), and because code and explanations were probably included in the training data, it seems reasonable to me that the "rewrite rules" of Python bled a little bit into the AI.
It makes me insanely curious about the internal structures though. I gave that site 2 similar examples: one produces a correct explanation while another produces an incorrect explanation. The difference: a deleted line of insignificant whitespace
From those 2 examples, I think its pretty clear that the AI's "rewrite rules" don't always coincide with Python's, but I would expect this to be mitigated by targeted training (like Copilot).
I haven't messed with it much since I find the process for the 'good one' (Midjourney) annoying (go onto a Discord channel, type your request in public for a bunch of people, wait a while, hunt for it amongs the long channel of requests, etc).
I'm assuming the process has gotten better since, but I don't know. I'm mostly just using free vector art and simple colors/shapes or whatever simple things I can make in Blender for my art still, in part because there's such a backlash against using any A.I. art right now.
Most of it is judging by what people have been saying in groups online. Some people have found it very useful and use it extensively, like for their board game art.
It doesn't even have to get 100% of the way there (for coding games based on rulesets). Even 75% would probably save me a lot of time and allow me to find the energy to close the gap as opposed to not even starting since I have so many other projects going on.
You mean such as when high level languages became mainstream and we no longer needed to code in assembly language? Or when IDEs became widely available? The underlying design skills are still difficult to acquire and not displaced by new tools - that is at least until SoftwareArchitectGPT comes along...
It is different this time. In the past automation effected some domains more and some domains not at all. People moved to those other domains. AI can run all the domains humans do and more that they can't.
Never went away; been hearing about it since msft sold the idea to managers in regards to SharePoint... Fucking SharePoint.
They can design their own forms and hook the data up to bits of automation. Like magic for muggles.
Yes, this is also why senior engs get payed so well. Actually if you are junior dev you basically cost money, imo. However a lot of companies hire those in hopes they will make their career and stay longer - especially startups do that, as they can slide on the company culture hook way more. Also they need them as senior engs require equivalents of "secretary" to handle less import things.
I don't mean to sound mean, senior devs are also secretaries of their cto and so on.
Even in the gaming industry. Typically you have people developing an engine, common frameworks... tools that downstream work to lower branches - now I can just type to chatgpt rather than go through requesting/reviewing, see quicker where I mis-designed my "framework" etc... I am afraid it's gonna be not great for junior engs all together.
Every org has varying level of engineers. This technology will make cheap grind, well cheap. We see the progression. With the speed of changes it seems we should all be worried. (as why does it matter despite being last we fell a year after)
42 comments
[ 0.25 ms ] story [ 168 ms ] thread> There wont be any unprecedented mass layoffs
Baghdad Bob strikes again.
What i think will lead to mass layoffs is the current recession rather.
If your goal is to get a response to an LSP query, why on earth would you use an LLM trained on data where >99.9999% of that data has nothing to do with answering an LSP query?
Why would I switch out an LSP server of 100% accuracy for an LLM that’s slower and has lower accuracy?
A sane approach would be to start understanding the requirements and work from there, trying to figure out where the challenges are.
GPT can‘t do this currently.
A real test of it's abilities would be synthesizing a NEW type of nuclear reactor.
> capable of at-least C++ "constexpr"-style compile-time computation, which shouldn't even be possible if one presumes GPT is "just" a giant database storing only multidimensional word similarity scores and sequence distribution from text inference
I don't see how being a giant word-DB necessarily disqualifies compile-time computation. You can view computation as applying a series of term rewrite rules to an input until some sort of termination condition (or indefinite loop). In the case of these AI, the input is the prompt and predicting the next token is a limited form of term rewriting (where the rules are probabilistic and based off the network), and because code and explanations were probably included in the training data, it seems reasonable to me that the "rewrite rules" of Python bled a little bit into the AI.
It makes me insanely curious about the internal structures though. I gave that site 2 similar examples: one produces a correct explanation while another produces an incorrect explanation. The difference: a deleted line of insignificant whitespace
* https://whatdoesthiscodedo.com/g/dd2af89
* https://whatdoesthiscodedo.com/g/45ea060
From those 2 examples, I think its pretty clear that the AI's "rewrite rules" don't always coincide with Python's, but I would expect this to be mitigated by targeted training (like Copilot).
I'm assuming the process has gotten better since, but I don't know. I'm mostly just using free vector art and simple colors/shapes or whatever simple things I can make in Blender for my art still, in part because there's such a backlash against using any A.I. art right now.
Most of it is judging by what people have been saying in groups online. Some people have found it very useful and use it extensively, like for their board game art.
It doesn't even have to get 100% of the way there (for coding games based on rulesets). Even 75% would probably save me a lot of time and allow me to find the energy to close the gap as opposed to not even starting since I have so many other projects going on.
https://www.youtube.com/watch?v=7Pq-S557XQU
I don't mean to sound mean, senior devs are also secretaries of their cto and so on.
Every org has varying level of engineers. This technology will make cheap grind, well cheap. We see the progression. With the speed of changes it seems we should all be worried. (as why does it matter despite being last we fell a year after)