Interested in how they compare to each other, difficulty of getting into, culture and WLB, and I guess TC never hurt.
You'd know what research problems excite you and which projects you'd want to work on as these places all publish. I'd base your choice on that
My experience at MSR left a lot to be desired. MSR is not what it used to be, especially in AI it is barely recognized in the field. You probably haven’t heard of anything big came out of MSR for AI this year and when something big did came out in past such as Resnet or Malmo, those people almost immediately left for greener pastures. Ironically, there are tons of small and insignificant things comes out of MSR. In fact, in current NeurIPS, they were #5 in terms of paper count however it’s highly questionable how many of these will exert any influence in the field. As far as AI is concerned, MSR has mastered the art of optimizing paper count while magically avoiding to publish anything of long term importance. All these isn’t helped by the fact that Microsoft leadership has lost interest from long term research creating a huge internal struggle for justifying value and chasing after short term impact on products to make leadership happy. After LeCunn left as head, FAIR is cooling down dramatically. They had lot of hiring problem, internal struggles of productization/value justification and severely limited funding. They are are #8 on NeurIPS paper count ranking but it’s hard to recall if anything significant coming out from there just like MSR. OpenAI perhaps has highest impact on the field per researcher capita as other commentator has said. Their Gym framework is de facto standard and their spinning up is now go to for most new comers. They have done enormous amount of work in virtually every sub field with long lasting contributions. Interestingly they don’t seem to care much about publishing papers so there are almost ZERO low significance papers churning out from there! FAIR and MSR has lot of learn from them. Google Brain takes the cake when it comes to total raw productivity as well as long lasting impact. They are tour de force in the field. For NeurIPS publication count, they are #1 and their next biggest competitor is not even half way there to match them! They do produce lots of noise just like MSR or FAIR, however, difference being that every year they have made slew of serious breakthroughs which will be part of long term history of this field. However, I must also say that they are the largest of all labs, much better funded, encouraged to have long term aspirations, have safety and stability under highly recognized leader like Jeff Dean so lot of muscle power + freedom + stability to do big things.
Health & Wellness
7h
3408
Why are women naked in gym?
Tech Industry
4h
330
Google has a workforce crisis thanks to their leftist/woke ideology
2024 Tax
6h
1034
Biden’s new tax proposal is wild
Cars
Yesterday
1558
Why are Americans obsessed with SUV?
Tech Industry
13h
2103
Go woke, go broke: Google fires 28 employees involved in pro-Hamas protest
OpenAI is a joke compared to Brain, FAIR and MSR. They routinely hype up incremental results which are more or less derivative of existing breakthrough results from the other three research groups. Their most recent release - emergent multi-agent tool use - is a good example. Brain and MSR both edge out FAIR but all three are extremely strong research groups in their own right. MSR is much more active in a wider variety of fundamental research than Brain is, and FAIR focuses more on applied research.
Dude, you are smoking something and obviously not in the field. OpenAI has done some very legit work and lot of it now standard in RL research. In fact they have the highest productivity of any of these labs. Sure, they need to hype to keep funding coming in but everyone else is doing it too. They are literally the smallest lab but they have legit breakthrough to claim in everything ranging from NLP, robotics, games, OSS.
Okay, sure. Why don't you name some of those breakthroughs, specifically, and we can debate them further?