Coding, Data Science, A.I., Robots |

  • Thread starter Thread starter nycfan
  • Start date Start date
  • Replies: 625
  • Views: 21K
  • Off-Topic 
Willy said:
I feel like we are all dead men walking at this point, with respect to our careers. In 200 years, if we are still around, we will have probably figured it out, and I doubt money, ownership of real estate, etc. will still exist. But the question is - how do we get there? The next few decades will be very strange. If you are 35, I’m not even sure how you begin to plan for retirement.
Click to expand...
A guy at the PAAS/SAAS software co i work for demo'd some really impressive stuff (it is already making our jobs way easier) which could probably replace 90% of our QA, Engineering, Support and Docs folks at some point in the future. The company could be fine with a few product managers, an architecture type person who understands the full pic and can drive the prompting, and a few junior architect and SME types who can routinely provide feedback and fine-tuning for the system which creates the code, the testing, maintains the code, the monitoring, writes the docs and provides the support. Once image generation gets better it could even create training materials in case the architect retires.
 
gregh1 said:
A guy at the PAAS/SAAS software co i work for demo'd some really impressive stuff (it is already making our jobs way easier) which could probably replace 90% of our QA, Engineering, Support and Docs folks at some point in the future. The company could be fine with a few product managers, an architecture type person who understands the full pic and can drive the prompting, and a few junior architect and SME types who can routinely provide feedback and fine-tuning for the system which creates the code, the testing, maintains the code, the monitoring, writes the docs and provides the support. Once image generation gets better it could even create training materials in case the architect retires.
Click to expand...
I don't think it's quite there yet but it is getting a whole heck of a lot closer. The problem we run into is the context memory. You'll be working on a problem and give it requirements. It'll start doing requirements one and two and three but forget requirements 4 through 10. So you prompted with the requirement for four and it forgets one two and three. It doesn't do a great job putting everything together but when it comes to individual pieces, it is amazing.
 
gregh1 said:
It is official: i'm slacking on internal documentation at work for sake of job security. I don't love this job, but I don't want to hasten my demise. The better our internal docs, the better our AI projects will be at minimizing the need for expensive humans on payroll.
Click to expand...
Try this. Put your code into something like Claude and ask it to create documentation and then see how good it is coming out. I think it's going to be really good but would be interested in what you see.

If I were you, I would try to start sliding into a product owner and product manager type roles. Good ones are still going to offer a lot of value.
 
gtyellowjacket said:
I don't think it's quite there yet but it is getting a whole heck of a lot closer. The problem we run into is the context memory. You'll be working on a problem and give it requirements. It'll start doing requirements one and two and three but forget requirements 4 through 10. So you prompted with the requirement for four and it forgets one two and three. It doesn't do a great job putting everything together but when it comes to individual pieces, it is amazing.
Click to expand...
Let's hope for the general white collar labor market the context issues remain a problem. And power gets expensive, chips don't get more efficient ;)
 
gtyellowjacket said:
I don't think it's quite there yet but it is getting a whole heck of a lot closer. The problem we run into is the context memory. You'll be working on a problem and give it requirements. It'll start doing requirements one and two and three but forget requirements 4 through 10. So you prompted with the requirement for four and it forgets one two and three. It doesn't do a great job putting everything together but when it comes to individual pieces, it is amazing.
Click to expand...
Honestly, my mind had some trouble just keeping up with the black magic behind the black magic. Using "DeepEval" to improve the black magic created graphs and stuff.
 
Last edited:
gregh1 said:
Honestly, my mind had some trouble just keeping up with the black magic behind the black magic. Using "DeepEval" to improve the black magic created graphs and stuff.
Click to expand...
Some of the black magic, even the geniuses don't know. LLMs are mostly black boxes that get a little gray around the edges.

The tools are a little bit better for determining why agents make decisions one way versus the other. Not great but better than evaluating why an llm gave one answer versus another.

I haven't heard of deepeval though and that is a project our team is literally working on right now. Thank you.
 
You must log in or register to reply here.
Back
Top