4 ms·
Utter delusion in such threads to think this has anything to do with AI. MSFT saw a way to cut costs and the lax regulatory environment and just went with it.
by jimsimmons 4y ago
Utter delusion in such threads to think this has anything to do with AI.
MSFT saw a way to cut costs and the lax regulatory environment and just went with it. GitHub also closed down their SF office, had some layoffs here too and went fully remote
- 29athrowaway 4y agoThat office was cool. It had a replica of the oval office.
- zumu 4y agoNonetheless, very few people went in, even before the pandemic. They should have closed it years ago, tbh.
- hulitu 4y ago> MSFT saw a way to cut costs I thought indians were cheap (and that's why companies hire them in the first place). Is MS not able to pay them ?
- Apes 4y agoAll employees across the globe are cheap compared to even a single C-level, but how many C-levels have you seen let go during this "recession"?
- hulitu 4y agoI'm sorry. I think you don't get it. They fired employees, not managers. /s Those firms are in the long run, condemned. For me the move seems strange because a lot of companies in EU are expanding in India and are hiring a lot of indians. And, in contrast with China, they never said that the reason to open branches in India was to be able to get a foot on the market there. It's like India has cheap developers but no promising market.
- huijzer 4y agoHow about both? Both lax regulatory environment and expected AI productivity gains in the (near) future.
- karmasimida 4y agoWould say maybe for GitHub, as they are the core company that is driving AI for coding, they might need more devs, not fewer of them in regular years. AI's impact will come, in next several years, but now it is too soon.
- phillipcarter 4y agoMicrosoft is practically logjammed on developer time for any project in active development. Why on earth would someone get rid of the very people who are best to drive these AI tools to create software?
- jimsimmons 4y agoLet me fire these actual workers doing real work with the hope that $magic_upcoming_tech will hold the ship. Great plan
- PartiallyTyped 4y agoPerhaps 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 agoI 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.