03-28-2023 12:31 AM
@pawhan11 wrote:
Tend do disagree on generation. You will need less people to do the same, one skilled developer that will know what he wants + the right querries to those models and will check that code later.
My experience after few days of using gpt4 for .net and unit teste generation it gets it right and good 90% of time. I have only 2 years exp in text based after switching from 10 years in labview, when i compare time to write this boring code by myself vs generation + output verification it is so much faster.
I expect it might go into direction whare we will more focus on interesting part like architecture, system design and code generation will be assisted by those models. Now it is way better to query model vs stackoverflow and google search for most of my cases.
And that's the whole problem right there. 10% of the time there will be something wrong, but since you didn't generate the code yourself, you'll have no idea where to start looking.
03-28-2023 01:17 AM
The requirements for a LabVIEW AI implementation are about the same as for an Electronic Schematic or PCB AI implementation. A lot of AI development in that market sector is already well underway. For NI, the issue is what AI strategy is the best way forward considering its development resource base. If NI is feeling rich then it can develop internally. If budgets are tight then NI could join an open source project and then bring the results back into LabVIEW.
I haven't needed AI yet for LabVIEW. But I have needed and used metaprogramming to accelerate design.
03-28-2023 03:18 AM
Debugging? It is less consuming than building up the whole thing.
03-28-2023 03:31 AM
Really depends on the bug, i've had one that took months to resolve.
03-28-2023 05:40 AM
Just one? I've had many that were well out of sight for a long time. Some I never managed to solve (like a crash on exiting a program).
Still ChatGPT is generating a huge hype because it is one the first language models to produce goodish code. I would treat it like I handle junior code: with eyes open, a lot of review and coaching (I don't know how coaching would come into play). Still it appears to be a fraction of the cost of a junior programmer and more readily available (no hiring overhead, etc).
I can see this technology being a huge step forward in a few years. As programmers we would focus on the broad strokes, gathering requirements and translating them for an AI to code and validating the code once written.
As of yet I would trust the language model with unit tests and syntax completion (in text languages). But we'll have to see some serious work from NI for these technologies to interface with LabVIEW. VI scripting can be very slow at times (and github copilot produces several different suggestions at a time). Also given how long certain features took to be implemented in the past, I'm not holding my breath on that one.
03-28-2023 08:50 AM
@Artst wrote:
If the COVID-19 vaccine manufacturers can weasel their way out of any liability for their vaccines, and still be able to sell it, what is to prevent the same for AI?
Have you looked at the terms on any software you've bought lately? They don't take responsibility for anything. That is the way in corporate 'Merica.
03-28-2023 11:37 AM
@pawhan11 wrote:
Tend do disagree on generation. You will need less people to do the same, one skilled developer that will know what he wants + the right querries to those models and will check that code later.
My experience after few days of using gpt4 for .net and unit teste generation it gets it right and good 90% of time. I have only 2 years exp in text based after switching from 10 years in labview, when i compare time to write this boring code by myself vs generation + output verification it is so much faster.
I expect it might go into direction whare we will more focus on interesting part like architecture, system design and code generation will be assisted by those models. Now it is way better to query model vs stackoverflow and google search for most of my cases.
Great to hear some first hand experience! Thanks. I suppose that my initial view is that text-based programming languages now have a flexible "VI scripter".
03-28-2023 04:47 PM - edited 03-28-2023 04:48 PM
@benjamin-hinrichs wrote:
Just one? I've had many that were well out of sight for a long time. Some I never managed to solve (like a crash on exiting a program).
Still ChatGPT is generating a huge hype because it is one the first language models to produce goodish code. I would treat it like I handle junior code: with eyes open, a lot of review and coaching (I don't know how coaching would come into play). Still it appears to be a fraction of the cost of a junior programmer and more readily available (no hiring overhead, etc).
I can see this technology being a huge step forward in a few years. As programmers we would focus on the broad strokes, gathering requirements and translating them for an AI to code and validating the code once written.
As of yet I would trust the language model with unit tests and syntax completion (in text languages). But we'll have to see some serious work from NI for these technologies to interface with LabVIEW. VI scripting can be very slow at times (and github copilot produces several different suggestions at a time). Also given how long certain features took to be implemented in the past, I'm not holding my breath on that one.
I'm looking forward to a decade or two of second-rate software with subtle bugs that no one will ever be able to figure out... Wait, aren't we there already?
04-18-2023 03:02 AM
My company uses more than 40 licenses of LV2016. We have already spent a lot of money, but it is simply impossible to upgrade all these licenses to the latest version. What company spends so much money on a software that can be easily replaced by a cheap/free one? If they still stick to this model, we will keep 2016 or switch to a different language (C#?). We also no longer buy new licenses.
04-18-2023 09:04 AM
My expectation is that 2016 will be quite functional for at least a few more years. The programs work, the software is usable, and you OWN the programming environment. How likely is it that NI will put forward features, or improvements in the next few years that will make the transition attractive. In your case, upgrading from 2016 would be expensive in any case. NI has an uncertain future. It looks like they are trying to maximize profits by pushing their support and sales off to third parties, and are in active talks to find a buyer. The resulting short term increase in profit may make them attractive enough to be picked up by a private equity group that is only interested in the return they can squeeze out. The future of NI is uncertain. It is prudent to coast on an existing license while exploring alternate options, or hoping the company makes a concerted effort to better serve their customer. Then again, there is always a chance that another party may rise up to create a credible replacement for LabVIEW, maybe even an open source environment community supported.. My license is permanent and frozen. When talking to young people about programming for industrial environments, I always mention LabVIEW, and suggest they look elsewhere. It's a sad thing.