4 ms·

> I'm not going to worry too much about my career being snuffed out by AI until AI actually exists. Can you explain what underpins this position? Is it about t

> I'm not going to worry too much about my career being snuffed out by AI until AI actually exists. Can you explain what underpins this position? Is it about timelines, likelihoods, not taking preparative actions in general or what?

safety1st 3y ago
Sure, I don't think the likelihood is high. I think it's unlikely we'll develop an AI any time soon which can consistently solve critical thinking problems that even a junior developer can solve. LLMs exhibit very little of that problem solving behavior. They synthesize information from existing information and often what they produce is nonsensical. When people have argued that there is some kind of reasoning going on there, I haven't found it convincing. I use an LLM every day, it makes me more productive, it speeds up a lot of tasks. It doesn't replace the executive thinking that I hire people for (even pretty junior people), like delegating a problem, having them go off and think through various angles and come up with a really good conclusion... an LLM won't ever do this on its own, it can assist the person who's doing it though. I can see a scenario where everyone working for me becomes twice as efficient with an LLM (me too) and it translates to a dev hiring freeze and a tougher job market, maybe for years even. As I mentioned in another comment I think devs who don't stay on the cutting edge will be at risk of losing their jobs. But fundamentally there's a lot of room for software to be solving more problems than it is today, grow the supply and the demand will likely grow with it, so the "human developers will disappear so quickly I shouldn't have a child" scenario is hyperbole and kinda toxic tbh.
- the8472 3y ago
  I think that's a too-narrow focus on LLMs. Of course they're not good at problem-solving because that's not what they're optimized for. That they can approximate any problem-solving at all is already impressive. Consider that there are also models that are actually trained to act as agents in an environment via reinforcement learning. Something they do for robotics and game-like simulations. It's not a bigger version of today's models we should consider but broader ones that integrate more capabilities and training regimes.
  - safety1st 3y ago
    Sure, but at that point we are talking about stringing together a variety of algorithms in a very speculative way, we're not extrapolating from what is being deployed in the world right now. Maybe we will eventually get to AGI or something like it in that manner, and maybe then the AI will then destroy all jobs and make life horrible, but I think at this point the chain of events is speculative enough that it's a silly reason to forego having children, which was the bar set by the OP, and it also seems like a pretty vague threat to programmer jobs.
    - the8472 3y ago
      This is hardly speculative. RL-trained agenty models just don't grab the headlines. They have been growing over years. Think AlphaGo, OpenAI Five, This Model Without A Catching Name[0], AlphaCode are all in that category. And they're already getting combined with pretrained transformers[1] to get some world-knowledge. There's a lot more going on than "just" ChatGPT or Dalle. [0] https://deepmind.google/discover/blog/generally-capable-agents-emerge-from-open-ended-play/ https://deepmind.google/discover/blog/generally-capable-agen... [1] https://deepmind.google/discover/blog/robocat-a-self-improving-robotic-agent/ https://deepmind.google/discover/blog/robocat-a-self-improvi...
      - safety1st 3y ago
        Thanks for the links, I'm still of the view that this stuff has a ways to go before it causes enough job destruction that so as to make parenthood pointless, however they were an interesting read! (To be clear here yes one element of my skepticism is around applications of AI, but the other very large and arguably more significant element is how unlikely it is that any of this stuff would ever constitute a sane reason to not have kids)

safety1st 3y ago

Sure, I don't think the likelihood is high. I think it's unlikely we'll develop an AI any time soon which can consistently solve critical thinking problems that even a junior developer can solve. LLMs exhibit very little of that problem solving behavior. They synthesize information from existing information and often what they produce is nonsensical. When people have argued that there is some kind of reasoning going on there, I haven't found it convincing. I use an LLM every day, it makes me more productive, it speeds up a lot of tasks. It doesn't replace the executive thinking that I hire people for (even pretty junior people), like delegating a problem, having them go off and think through various angles and come up with a really good conclusion... an LLM won't ever do this on its own, it can assist the person who's doing it though. I can see a scenario where everyone working for me becomes twice as efficient with an LLM (me too) and it translates to a dev hiring freeze and a tougher job market, maybe for years even. As I mentioned in another comment I think devs who don't stay on the cutting edge will be at risk of losing their jobs. But fundamentally there's a lot of room for software to be solving more problems than it is today, grow the supply and the demand will likely grow with it, so the "human developers will disappear so quickly I shouldn't have a child" scenario is hyperbole and kinda toxic tbh.

the8472 3y ago
I think that's a too-narrow focus on LLMs. Of course they're not good at problem-solving because that's not what they're optimized for. That they can approximate any problem-solving at all is already impressive. Consider that there are also models that are actually trained to act as agents in an environment via reinforcement learning. Something they do for robotics and game-like simulations. It's not a bigger version of today's models we should consider but broader ones that integrate more capabilities and training regimes.
- safety1st 3y ago
  Sure, but at that point we are talking about stringing together a variety of algorithms in a very speculative way, we're not extrapolating from what is being deployed in the world right now. Maybe we will eventually get to AGI or something like it in that manner, and maybe then the AI will then destroy all jobs and make life horrible, but I think at this point the chain of events is speculative enough that it's a silly reason to forego having children, which was the bar set by the OP, and it also seems like a pretty vague threat to programmer jobs.
  - the8472 3y ago
    This is hardly speculative. RL-trained agenty models just don't grab the headlines. They have been growing over years. Think AlphaGo, OpenAI Five, This Model Without A Catching Name[0], AlphaCode are all in that category. And they're already getting combined with pretrained transformers[1] to get some world-knowledge. There's a lot more going on than "just" ChatGPT or Dalle. [0] https://deepmind.google/discover/blog/generally-capable-agents-emerge-from-open-ended-play/ https://deepmind.google/discover/blog/generally-capable-agen... [1] https://deepmind.google/discover/blog/robocat-a-self-improving-robotic-agent/ https://deepmind.google/discover/blog/robocat-a-self-improvi...
    - safety1st 3y ago
      Thanks for the links, I'm still of the view that this stuff has a ways to go before it causes enough job destruction that so as to make parenthood pointless, however they were an interesting read! (To be clear here yes one element of my skepticism is around applications of AI, but the other very large and arguably more significant element is how unlikely it is that any of this stuff would ever constitute a sane reason to not have kids)