A lot of “AI” investment in countries like India is just annotation and labeling kind of work .
It has so become huge there, that is getting to similar size as the traditional call centre industry, hundreds of thousand of people work on curating data for model training.
Google hiring a senior lead who can scale low skill business processing massively is just inline with that.
Any compute investment is largely local . Internet and Electricity is expensive and erratic, there is red tape, loads of import duties, there are dozens of cheaper places for cheap data centers.
All this to say it is not the kind of outsourcing you think it is.
—-
There is a some amount of high tech outsourcing happening, those are driven by challenges in getting even a business visa in last 5 years.
those numbers are quite small in aggregate won’t be news worthy.
India simply does not have that kind of high quality talent pool .
The education system is both expensive and very poor. There is shortage of qualified teachers, even bigger shortage of good ones. Students and parents just want to get a “degree” /maximize scores to get a job, There is little interest to learn.
Given that this is presumably all data used for RLHF'ing, I wonder how much of this is responsible for things like LLM "sycophancy" or the issue of "hallucinations". What if the reward-hacking entity isn't the LLM/optimizer but the human annotator in the loop.
Sycophancy makes sense as a cultural disconnect, western or americam culture tends to expect superiors to be treated as verbal equals while most other cultures layer obsequience from subordinates to superiors. Thanks for the thought.
> A lot of “AI” investment in countries like India is just annotation and labeling kind of work .
Then why are they giving one year of free accounts for developers to the whole country? Remember they have about one million CS graduates every year. Do you see the scale of costs?
This is a sales tactic, in the same way that Jetbrains and Microsoft will give out free licenses for students in the hopes that some of them will go on to work for employers who will then buy the product. It also helps convince investors about potential growth and MAU increase.
Same goes for the OpenAI and Anthropic office announcements posted by the GP, these are sales offices with very little tech employment or outsourcing happening.
Re “investments”: these are also not what you think. I remember when Amazon was partnering up with schools in the US to provide “machine learning education” where said education was thinly veiled ads for managed ML services like Sagemaker and Rekognition. I wouldn’t be surprised if the $3B investment from Microsoft was primarily free courses and certifications to cultivate a market, as well some annotation work (I know that Amazon has been hiring quite a few people to do this kind of work).
For the same reason legacy publishers sell expensive textbooks at 1/10 of the price in India, identical to the USA version save for a stamp that says you can't export them from India.
The low cost editions are significantly poor quality prints, smaller fonts crappy ligatures, margins line heights- all to squeeze it into less pages, poorer paper, monochrome, laxer quality control -it is not uncommon for few pages to have minor defects .
I appreciate the effort to make something available for that cheap, but they are bad.
It is like business class and economy on a flight. You both go to same destination at the same time sure but the journey is widely different, if you don’t care about the journey it is waste of money yes, but it cannot be denied there is a reason why it is expensive.
In this analogy hard bound would be first class . Mostly unavailable and truly a luxury.
A lot of IT work that gets done in India is low skill things that could be automated even before AI codegen tools. These jobs are particularly vulnerable to AI productivity gains .
The thesis here is that current gen tools should enable 2 people do the work of 10 and LLM models get paid the budget of 2-3 devs.
The outsourcing org comes out ahead with spending only 70% as before (salary of remaining devs will naturally increase)
——
India is by far the largest market for this. 1yr free is for same reason any VC funded company discounts their product or gives it free first, or Microsoft famously rather have you use their software pirated than not at all.
This is only to get adoption, if your workers only know how to code with Claude or Codex then you are going to buy those tools as a company
Nobody is going back once you get them to change. OpenAI et al hope to capture 30% value of the IT outsourcing sector while making to cheaper by 30%. There is no free lunch.
——
The number of CS “grads” are not a useful metric for India.
The industry shifted to hiring other engineering streams to any grads with aptitude to learn a long while back.
The vast majority of them don’t have any programming skills, don’t have critical thinking skills, are poor at communication.
They are unemployable. Don’t take it from me, NASSCOM says this too.
The degrees are largely from mills that charge up to $80k-$100k (med degrees go even higher) promising jobs but don’t teach much.
It has so become huge there, that is getting to similar size as the traditional call centre industry, hundreds of thousand of people work on curating data for model training.
Google hiring a senior lead who can scale low skill business processing massively is just inline with that.
Any compute investment is largely local . Internet and Electricity is expensive and erratic, there is red tape, loads of import duties, there are dozens of cheaper places for cheap data centers.
All this to say it is not the kind of outsourcing you think it is.
—-
There is a some amount of high tech outsourcing happening, those are driven by challenges in getting even a business visa in last 5 years.
those numbers are quite small in aggregate won’t be news worthy.
India simply does not have that kind of high quality talent pool .
The education system is both expensive and very poor. There is shortage of qualified teachers, even bigger shortage of good ones. Students and parents just want to get a “degree” /maximize scores to get a job, There is little interest to learn.