TikTok vs Amazon

[pale wave]

def tiktok if ban doesnt affect

[blazing vortex]

same position you're in lol

[unborn star]

Amazon

Better name

More stable

[marble nebula]

I am in the same position as you - mle intern for tik tok, and also sde intern for amazon

[dusk matrix]

the work is quite different, the position matters more than the company in your case i think

[hushed kraken]

I thought AWS was better than TikTok

Why are people voting TikTok?

What makes it better than AWS?

[marble nebula]

That might be true. Although I'm not quite sure how team matching works at Amazon. Given my background is almost entirely ml based I would hope to get matched to an ml team? But in any case the ml done for an mle position at tik tok (at least per their job description and company reputation) seems quite different

[manic sonnet]

I am also in the same position, tiktok vs amazon

[dusk matrix]

I was mle intern at another chinese big tech in 2023, some of my coworkers have interned at bytedance/tt. we did mostly experiments with the recommender systems models and not so much “software engineer” with like actually serving models and stuff. If that route interests you then definitely tiktok. If you were more thinking like ml platform engineering and stuff then maybe zon

[winter void]

I'd go tiktok, higher bar for entry and good team it seems

but it's pretty close

[marble nebula]

wait this is very interesting. So you were actually doing model fine tuning etc with company compute, and testing different architectures and things like that? Whereas for amazon I'm still not sure what kind of ml you would actually do, that is if you actually get matched to an ml team. I hope its not just creating a langchain

[dusk matrix]

The architecture and hyperparams were pretty fixed for most of my experiments, it was on an already mature and quite large model. I did more experiments on the type of data that gets added as new features.
If your tiktok job positions mentions stuff around algorithm then its probably pretty close to that

[dusk matrix]

Im currently interning as an ml infra swe at a small-midsize company in canada, we work on things like ensuring high uptime, maintaining fast serving speed and sutff like that, so very different from my previous mle internship

Thats why i say position might matter more here cause they can be veeeery different jobs

[marble nebula]

Thanks so much for these invaluable insights! So how research was done for these positions? As in did you guys track recent literature and make decisions/updates to the code based on a literature search? I have looked on the internet but it doesn't seem like theres a lot of information on what mles do at these companies. As you probably know for university research its kind of like making a codebase from scratch based on some research direction or borrowing code from related projects and changing stuff inside the code and seeing if things are better based on the theory/empirical results in a controlled setting

[dusk matrix]

Yes and no. For my case, there could be research involved but id say its someone like a senior or a team leads that decides to see if they want to, lets say test out an architectural tweak or new feature. Then once the project is defined ish the job goes down to general mle.
Not much stuff are implemented from scratch for our team, we were ads rec so its a mature and old team, especially for interns the project scope didnt really go beyond just mildly tweaking existing model.
It doesnt sound like a lot of work but our projects were end to end from almost unstructured data to model evaluation and hundreds of terabytes of data go through that pipeline for a project. Training a single model for my projects was like 6 hours.

Prod ml is quite different than university research

Especially as entry level, expect much smaller scope and longer time spent on things like data engineering, feature crafting and other stuff besides actual tuning

Again this is team based, its probably way different if its a more research oriented team or like some llm team

[marble nebula]

I see. Thanks very much once again for this wealth of info. It was for this position: Machine Learning Engineer Intern (E-commerce Governance Algorithms) if it helps. From the job description it seems they are working on this: Interest or experience in anti-fraud, anti-spam, platform-integrity, or similar fields. It's interesting you guys even did some pretraining and also quite a bit of end to end work. It doesn't seem to be a recommender system based team more like an anomoly detection/safety team, but having worked previously with some very simple movie lens tasks for research with a distillbert based collaborative filtering what you did seems interesting and definately something I don't think a pure sde intern role at amazon would be working on? May be wrong about this. I have the manager's name so maybe I could reach out and set up a meeting

[dusk matrix]

apologies for the shallow choice of word but the vibe is quite different.

despite being mostly work on models in production, the team still maintained kind of a more researech vibe compared to swe jobs. we had like biweekly meetings with people talking about new paper in the industry stuff, which is quite cool.

Touching any large-ish ml model on prod in general is not something a pure sde does, and definitely not what an intern swe does. It's hard to answer this question cause they pretty much have no overlap in work except maybe data engineering being swe work in some companies and mle work in others.

one isnt better than the other tho, im back to swe cause i find myself quite untalented in ml 😄 Just gotta find ur thing.

[wanton swan]

amazon if final internship, tiktok if not final internship

res value for tiktok is better, but you don't wanna be there full time

also tiktok higher ng tc doesn't matter if you can't liquidate

[wanton swan]

not the kinda risk i’d take imo

ur rsus depend on politicians

[marble nebula]

What happens if you say no to amazon is there some kind of internal blacklist? Or does it not really matter

[marble nebula]

I guess in terms of research perspective in machine learning, like for PhD admissions or just like general outlook from the industry, what do people think of tiktok or chinese companies in terms of their research output or research contributions? Or in general for the kind of ml that they do? My concern is basically how recruiters or admissions people would perceive taking a sde intern role at amazon vs a more specialized ml role at tiktok. I guess since deepseek v3 wowed a lot of giants of ml, and it came from a chinese startup, it probably shouldn't be too negative?

[dusk matrix]

honestly idk cause id lean towards positive as well

chinese big tech publish a ton especially in rec sys

and in llm and adjacent ones all these days too

deepseek is good, qwen is good, i cant imagine it being bad in any way

My concern is basically how recruiters or admissions people would perceive taking a sde intern role at amazon vs a more specialized ml role at tiktok.
if your applying to a swe job then the recruiter would probably perceive an mle role like 0.5x value compared to a previous swe job

vice versa i think it applies too

proobably because how different their tasks r