5 ms·

Let me fire these actual workers doing real work with the hope that $magic_upcoming_tech will hold the ship. Great plan

PartiallyTyped 4y ago
Perhaps internal models are just that good. It wouldn't surprise me given that people knew what GPT4 was capable of 6-9 months before public release.
- throwaway4233 4y ago
  I doubt that the internal models would be able to function as a developer though based on my limited knowledge. It would need to 1. Independently pick work. 2. QA it, sort out code reviews and then finally add them to the release cycle. 3. Write documentation or write change logs for it. While all of this can be automated, there would need to be manual verification at some point (to fix broken tests or update configuration values) and that would require an engineer who is aware of the involved domain. And to adjust for vacations, sickness and other events, multiple engineers would be needed. All of whom need to have experience working with that area of the codebase, for which hands-on experience would be hard to gain, since the models are doing most of the grunt work. And at this point the company would be paying for both the engineers and the operational cost of running those models.
  - eggsmediumrare 4y ago
    No one is saying there won't be engineers in the loop. Just fewer of them.
  - PartiallyTyped 4y ago
    I never claimed that there will not be engineers in the loop. The operational cost is many times smaller than that of an engineer, and with proper loops it can work tirelessly.
KeplerBoy 4y ago
magic upcoming tech developed at MSFT/GitHub/OpenAI. If anyone is convinced of the tech it's them.
- Quarrelsome 4y ago
  If it makes developers more productive then you can simply do more as an organisation with more developers. How many issues are still open in your Github? Typically there are thousands open in any going concern. Cutting headcount is a consequence of cuts which is a consequence of priorities, legal, under performance, politics, etc. The very idea that some idiot has the power within the org to just sack developers because they have misunderstood how generative AI works is..... oh wait, that is extremely realistic because humans are very stupid. But also I would suggest it is not indicative of any sort of incoming trend and its unlikely to be best practice.
  - aunty_helen 4y ago
    You're neglecting capital costs and the dominant market position GitHub has. Capital being spent on GitHub to develop it could be used elsewhere or banked as profit if you had a smaller team. GitHub isn't in a position that it's market dominance will be threatened by a slower pace of development. You're also talking in absolutes, GitHub says there's a 40% productivity boost from using ai. Assuming (simplistically) that this results in a 40% required headcount reduction for the same output, you can then see: 0% headcount reduction, 40% improvement in velocity 40% headcount reduction, 0% improvement in velocity Scale these two factors mentally to see that it's not all or nothing, maybe they've gone 20/20 and so some people still get to keep their jobs.
    - Quarrelsome 4y ago
      > GitHub says there's a 40% productivity boost from using ai. That is their own marketing bumpfh and idk if any org should dog-food sales copy as strategy. I am saying that cutting development teams by 40% on the wonky maths that generative AI makes developers 40% more "developery" won't be best-practice. I don't think its a leap before you look job and I don't think the toolsets and practices are mature enough to bank that 40%, at least, not yet. Therefore I think the slashing of the engineering team in India is not related to AI but rather budget or political or smth else. I guess at the executive level its about whether you think you can increase sales with more R&D or whether you have entirely saturated your market or organisational capacity. Still, I would argue that mindlessly cost cutting is more the domain of the post-IPO, than that of the entrepreneur.

Perhaps internal models are just that good. It wouldn't surprise me given that people knew what GPT4 was capable of 6-9 months before public release.

throwaway4233 4y ago
I doubt that the internal models would be able to function as a developer though based on my limited knowledge. It would need to 1. Independently pick work. 2. QA it, sort out code reviews and then finally add them to the release cycle. 3. Write documentation or write change logs for it. While all of this can be automated, there would need to be manual verification at some point (to fix broken tests or update configuration values) and that would require an engineer who is aware of the involved domain. And to adjust for vacations, sickness and other events, multiple engineers would be needed. All of whom need to have experience working with that area of the codebase, for which hands-on experience would be hard to gain, since the models are doing most of the grunt work. And at this point the company would be paying for both the engineers and the operational cost of running those models.
- eggsmediumrare 4y ago
  No one is saying there won't be engineers in the loop. Just fewer of them.
- PartiallyTyped 4y ago
  I never claimed that there will not be engineers in the loop. The operational cost is many times smaller than that of an engineer, and with proper loops it can work tirelessly.

KeplerBoy 4y ago

magic upcoming tech developed at MSFT/GitHub/OpenAI. If anyone is convinced of the tech it's them.

Quarrelsome 4y ago
If it makes developers more productive then you can simply do more as an organisation with more developers. How many issues are still open in your Github? Typically there are thousands open in any going concern. Cutting headcount is a consequence of cuts which is a consequence of priorities, legal, under performance, politics, etc. The very idea that some idiot has the power within the org to just sack developers because they have misunderstood how generative AI works is..... oh wait, that is extremely realistic because humans are very stupid. But also I would suggest it is not indicative of any sort of incoming trend and its unlikely to be best practice.
- aunty_helen 4y ago
  You're neglecting capital costs and the dominant market position GitHub has. Capital being spent on GitHub to develop it could be used elsewhere or banked as profit if you had a smaller team. GitHub isn't in a position that it's market dominance will be threatened by a slower pace of development. You're also talking in absolutes, GitHub says there's a 40% productivity boost from using ai. Assuming (simplistically) that this results in a 40% required headcount reduction for the same output, you can then see: 0% headcount reduction, 40% improvement in velocity 40% headcount reduction, 0% improvement in velocity Scale these two factors mentally to see that it's not all or nothing, maybe they've gone 20/20 and so some people still get to keep their jobs.
  - Quarrelsome 4y ago
    > GitHub says there's a 40% productivity boost from using ai. That is their own marketing bumpfh and idk if any org should dog-food sales copy as strategy. I am saying that cutting development teams by 40% on the wonky maths that generative AI makes developers 40% more "developery" won't be best-practice. I don't think its a leap before you look job and I don't think the toolsets and practices are mature enough to bank that 40%, at least, not yet. Therefore I think the slashing of the engineering team in India is not related to AI but rather budget or political or smth else. I guess at the executive level its about whether you think you can increase sales with more R&D or whether you have entirely saturated your market or organisational capacity. Still, I would argue that mindlessly cost cutting is more the domain of the post-IPO, than that of the entrepreneur.