3 ms·

It’s a trap to be scared and try to gamble on near future. By the time most of today’s wannabe AI gurus are good enough to get a decent job offer - AI would no

It’s a trap to be scared and try to gamble on near future. By the time most of today’s wannabe AI gurus are good enough to get a decent job offer - AI would not be such a big deal (best case scenario imo). I didn’t notice any “obvious” changes in the game that weren’t there before. For instance I don’t see anything different in requirements and comments of HRs reaching me. JVM is still same jvm, bugs are still bugs, unit tests are still unit tests. The fact that now instead of googling a problem you google and gpt it is a minor change in any real world complex engineering context that im aware of. What exactly is obvious to you that im missing?

mech422 3y ago
I agree on the AI stuff ... personally, I think it's going to need 5 years or so just to iron out all the legal and regulatory stuff before the 'next big push' In general, I think you make your money in Tech at the ends.. Either bleeding edge or trailing edge. ATM, I'm working on Rust, serverless, playing with Pijul, keeping an eye on OpenTF, and pulumi. For trailing edge stuff - I figure I'll be the worlds oldest COBOL coder by the time I retire and can use that to fund my retirement :-P
- Tommah 3y ago
  There are certain aspects of computing that will probably always be around. We'll always need fast compilers that generate fast code -- people don't want slow programs, and programmers don't want to wait for compiles to finish. Relational databases aren't disappearing anytime soon. And as we keep getting smaller and smaller devices, we'll keep shrinking things down, so we'll need new ways of representing and compressing data. The list goes on and on. If you can get comfortable with things like these, you'll always be useful to somebody. (And yes, COBOL will be with us for many years to come. In the next century, we'll probably be emulating IBM mainframes on quantum computers.)
p1esk 3y ago
The obvious thing is gpt-6 (or even gpt-5) will be able to do your job better and faster than you. This will happen in 1-2 years. That doesn’t scare you?
- diarrhea 3y ago
  I am bearish on that prediction. I’m not convinced LLMs can scale that much more. Of course, I’d be happy to be wrong. In that case, I’ll leverage the tools to become more productive.
- zilti 3y ago
  LOL maybe it will if you are a "coding by copying from StackOverflow" "developer"
- seabearDEV 3y ago
  Is there any evidence that supports this statement?
  - p1esk 3y ago
    Progression GPT1-2-3-4 in last 5 years, with no indication of any slowdown in the near future.
    - aristofun 3y ago
      there are plenty of signs that it is slowing down including current versions quality degradation, unprofitable business models (it's still too damn expensive to be covered by your average 19$/m subscription) etc.
    - badpun 3y ago
      There's no indication that the pace of innovation for LLMs in the near future will be anywhere near what it has been in the past few years.
      - p1esk 3y ago
        Is there an indication it won't be?
        badpun 3y ago
Tommah 3y ago
I think AI will still be important, but the AI of the future won't look like the AI that we have today. People have a tendency to redefine "AI" to mean whatever the new hotness is -- LLMs today, deep neural nets a few years ago -- but new things are created all the time, and fads always change. When I started grad school for comp sci nearly 20 years ago, I worked with a group that was big on "AI", but we were dealing with multi-agent systems. No neural nets or super-intelligences or anything like that. Neural nets were actually considered passé back then; I vividly recall a professor telling me that SVMs (support vector machines) were stronger than neural nets, because SVMs had a stronger theoretical foundation and were more amenable to mathematical analysis. Neural nets, on the other hand, just happen to work -- but they happen to work very very well! Deep learning didn't gain traction until after I had finished grad school. The LLMs that we have today are amazing, but there is still plenty of room for improvement. Having to train it on a huge dataset is problematic for some uses; perhaps there is a related structure that can be trained more easily and more quickly. That would also reduce the effect of OpenAI's monopoly. LLMs also have specific weaknesses, like poor performance at arithmetic. At this stage, I wouldn't really feel comfortable feeding problems into an LLM and presenting users with the LLM's answers. It's still the Wild West in many aspects. There is always an improvement on the horizon, but it's hard to tell where it will come from and when it will come. Maybe we'll have LLMs that really start to resemble intelligence, or maybe we'll have a totally different structure that does everything LLMs can do plus more.

mech422 3y ago

I agree on the AI stuff ... personally, I think it's going to need 5 years or so just to iron out all the legal and regulatory stuff before the 'next big push' In general, I think you make your money in Tech at the ends.. Either bleeding edge or trailing edge. ATM, I'm working on Rust, serverless, playing with Pijul, keeping an eye on OpenTF, and pulumi. For trailing edge stuff - I figure I'll be the worlds oldest COBOL coder by the time I retire and can use that to fund my retirement :-P

Tommah 3y ago
There are certain aspects of computing that will probably always be around. We'll always need fast compilers that generate fast code -- people don't want slow programs, and programmers don't want to wait for compiles to finish. Relational databases aren't disappearing anytime soon. And as we keep getting smaller and smaller devices, we'll keep shrinking things down, so we'll need new ways of representing and compressing data. The list goes on and on. If you can get comfortable with things like these, you'll always be useful to somebody. (And yes, COBOL will be with us for many years to come. In the next century, we'll probably be emulating IBM mainframes on quantum computers.)

p1esk 3y ago

The obvious thing is gpt-6 (or even gpt-5) will be able to do your job better and faster than you. This will happen in 1-2 years. That doesn’t scare you?

diarrhea 3y ago
I am bearish on that prediction. I’m not convinced LLMs can scale that much more. Of course, I’d be happy to be wrong. In that case, I’ll leverage the tools to become more productive.
zilti 3y ago
LOL maybe it will if you are a "coding by copying from StackOverflow" "developer"
seabearDEV 3y ago
Is there any evidence that supports this statement?
- p1esk 3y ago
  Progression GPT1-2-3-4 in last 5 years, with no indication of any slowdown in the near future.
  - aristofun 3y ago
    there are plenty of signs that it is slowing down including current versions quality degradation, unprofitable business models (it's still too damn expensive to be covered by your average 19$/m subscription) etc.
  - badpun 3y ago
    There's no indication that the pace of innovation for LLMs in the near future will be anywhere near what it has been in the past few years.
    - p1esk 3y ago
      Is there an indication it won't be?
      - badpun 3y ago

Tommah 3y ago

I think AI will still be important, but the AI of the future won't look like the AI that we have today. People have a tendency to redefine "AI" to mean whatever the new hotness is -- LLMs today, deep neural nets a few years ago -- but new things are created all the time, and fads always change. When I started grad school for comp sci nearly 20 years ago, I worked with a group that was big on "AI", but we were dealing with multi-agent systems. No neural nets or super-intelligences or anything like that. Neural nets were actually considered passé back then; I vividly recall a professor telling me that SVMs (support vector machines) were stronger than neural nets, because SVMs had a stronger theoretical foundation and were more amenable to mathematical analysis. Neural nets, on the other hand, just happen to work -- but they happen to work very very well! Deep learning didn't gain traction until after I had finished grad school. The LLMs that we have today are amazing, but there is still plenty of room for improvement. Having to train it on a huge dataset is problematic for some uses; perhaps there is a related structure that can be trained more easily and more quickly. That would also reduce the effect of OpenAI's monopoly. LLMs also have specific weaknesses, like poor performance at arithmetic. At this stage, I wouldn't really feel comfortable feeding problems into an LLM and presenting users with the LLM's answers. It's still the Wild West in many aspects. There is always an improvement on the horizon, but it's hard to tell where it will come from and when it will come. Maybe we'll have LLMs that really start to resemble intelligence, or maybe we'll have a totally different structure that does everything LLMs can do plus more.