7 ms·

Giant Corporations Are Hoarding the World’s AI Talent

38 points by sonabinu 10y ago

arcanus 10y ago
It's more pervasive than AI: The megacorps are hoarding all the world's top technical talent. The start-up value proposition is fundamentally weaker, at present. The salaries are substantially lower, and IPOs are too scarce (and too dilluted!) to make that lottery ticket appealing. This is to say nothing of the other megacorp benefits in 401k matching, maternity/paternity leave, etc. The top research labs at the big companies have also adopted many of the start up perks (flexible hours, food services, etc.) so you aren't losing out on lifestyle.
- zzzcpan 10y ago
  > The megacorps are hoarding all the world's top technical talent. I don't think that's true. The scale they operate on doesn't allow them to have good hiring processes, good management, etc., they can only get top talent by acquiring some other small company.
  - seanp2k2 10y ago
    Or pay ridiculous amounts, which they do. A lot of people are willing to be bored / frustrated if the dough is right.
- marssaxman 10y ago
  And yet - no matter what they're paying, you're still stuck working for a megacorp, with all the process overhead, repetitive drudgery, and narrowly siloed lack of opportunity that implies. Megacorps simply don't offer the kind of creative freedom you get for free at a startup - not unless you're willing and able put in a few years "proving yourself" all over again after you've been hired. Even then, your odds are not good. Been there, done that, three times now - never again.
  - Chris2048 10y ago
    > Megacorps simply don't offer the kind of creative freedom In what sense?
    - marssaxman 10y ago
      You get one small piece of a big puzzle to work on. Everything else belongs to someone else. Large organizations depend on structure; crossing its boundaries creates a problem for people who are trying to use the structure to simplify communication and thereby get more things done (i.e., managers). Initiative is therefore only welcome within the narrow scope of the one project you've been assigned; working on anything else creates complication which has to be resolved through meetings, design docs, and other formal process, which generally irritates people and takes more time than whatever it was you were actually trying to accomplish.
      - Chris2048 10y ago
        But focus and creativity are different.
        marssaxman 10y ago
        And...? I've told you what I experienced; I can't tell what you're questioning or observing here.
        Chris2048 10y ago
        The 'AND?" is you said corporations don't give you "creative freedom", but talked about "crossing its boundaries" of " the one project you've been assigned". The freedom of "creative freedom" usually means the freedom to be creative in you projects. Not having to focus on one particular project is a different thing. Even in a startup, which might be less silo-ed, actions on individual tasks needn't allow more creativity than in a large corporation. For example, I might have to leave infrastructure up to another team, rather than building it myself, but that doesn't mean I can't be creative in the work I am assigned, it just has to be focused on that assigned task.
Terr_ 10y ago
I wonder if this is an extension of the "only hire, never train" problem that a lot of companies have these days.
- ilaksh 10y ago
  Its an extension of the overall problem of over-centralization which has always been one of the main issues for civilization in different forms. See http://reddit.com/r/rad_decentralization http://reddit.com/r/rad_decentralization
- kem 10y ago
  I think so. There's a fine line that demarcates ML from data science from big data, for example, and it seems there's lots of chasing small differences that goes on as heat shifts from one fad to another. My guess is that a lot of people in one group could be moved to another, but rather than say "oh, you're working on X+epsilon, could you learn X and do that instead?" its "Oh you don't do X, we'll go hire someone who does X."
ilaksh 10y ago
One thing about AI research now is a lot of stuff seems to be built on pretty big local GPU clusters or pretty expensive AWS stuff. I had an idea the other day that could make that more convenient/affordable. The trick of course is coming up/applying different/new algorithms that don't rely on fast synchronization. Ideally that would be some kind of AI language that could describe/compress complex models/filters/parallel functions/etc. (https://github.com/runvnc/ailang https://github.com/runvnc/ailang) Then the nodes could exchange 'knowledge' in that special language. Combine something like turbo.js with a webrtc rpc swarm. Use rpc to pull out of task queue. So we can do GPGPU processing on web-based clusters in the browser. Can just use public gists with 'gpgpucluster' at end of description. Then use requirebin. Also use webtorrent/instant.io for imagenet example. Need to break imagenet up into a ton of little torrents. Then make an example that maybe just searches for certain classes like 'cat'. Example doesn't need to do actual machine learning, just store images in graphics memory pull them out when they search. And send them back. Maybe just use a subset of imagenet so that multiple clients can share the dataset and so work together to answer queries. EDIT: Thanks for the downvotes! I thought it was an interesting idea. But then again, the more interesting I think an idea is, the more likely it is to be downvoted all to heck. Hard for me to suppress my desire to share my ideas even though they are not appreciated.
- Cacti 10y ago
  It's hard to tell what you're asking for. There is no getting around the amount of computation required, and there are already plenty of systems to distribute workload the old fashioned way. In addition, while many projects rely on clusters of GPUs, a great deal of cutting edge research can be done on one or two decent commercial grade GPUs. A single new NVIDIA 1070 (and even better, the coming 1080 Ti) is cheap ($400) and is comparable to many good supercomputers in 2000, for example. An AWS with a K80 is only about 30 cents an hour at spot prices... after a week of training you could have yourself a state of the art result that wasn't even possible a mere few years ago. Not too shabby! Things have never been cheaper.

38 points by sonabinu 10y ago

It's more pervasive than AI: The megacorps are hoarding all the world's top technical talent. The start-up value proposition is fundamentally weaker, at present. The salaries are substantially lower, and IPOs are too scarce (and too dilluted!) to make that lottery ticket appealing. This is to say nothing of the other megacorp benefits in 401k matching, maternity/paternity leave, etc. The top research labs at the big companies have also adopted many of the start up perks (flexible hours, food services, etc.) so you aren't losing out on lifestyle.

zzzcpan 10y ago
> The megacorps are hoarding all the world's top technical talent. I don't think that's true. The scale they operate on doesn't allow them to have good hiring processes, good management, etc., they can only get top talent by acquiring some other small company.
- seanp2k2 10y ago
  Or pay ridiculous amounts, which they do. A lot of people are willing to be bored / frustrated if the dough is right.
marssaxman 10y ago
And yet - no matter what they're paying, you're still stuck working for a megacorp, with all the process overhead, repetitive drudgery, and narrowly siloed lack of opportunity that implies. Megacorps simply don't offer the kind of creative freedom you get for free at a startup - not unless you're willing and able put in a few years "proving yourself" all over again after you've been hired. Even then, your odds are not good. Been there, done that, three times now - never again.
- Chris2048 10y ago
  > Megacorps simply don't offer the kind of creative freedom In what sense?
  - marssaxman 10y ago
    You get one small piece of a big puzzle to work on. Everything else belongs to someone else. Large organizations depend on structure; crossing its boundaries creates a problem for people who are trying to use the structure to simplify communication and thereby get more things done (i.e., managers). Initiative is therefore only welcome within the narrow scope of the one project you've been assigned; working on anything else creates complication which has to be resolved through meetings, design docs, and other formal process, which generally irritates people and takes more time than whatever it was you were actually trying to accomplish.
    - Chris2048 10y ago
      But focus and creativity are different.
      - marssaxman 10y ago
        And...? I've told you what I experienced; I can't tell what you're questioning or observing here.
        Chris2048 10y ago
        The 'AND?" is you said corporations don't give you "creative freedom", but talked about "crossing its boundaries" of " the one project you've been assigned". The freedom of "creative freedom" usually means the freedom to be creative in you projects. Not having to focus on one particular project is a different thing. Even in a startup, which might be less silo-ed, actions on individual tasks needn't allow more creativity than in a large corporation. For example, I might have to leave infrastructure up to another team, rather than building it myself, but that doesn't mean I can't be creative in the work I am assigned, it just has to be focused on that assigned task.

Terr_ 10y ago

I wonder if this is an extension of the "only hire, never train" problem that a lot of companies have these days.

ilaksh 10y ago
Its an extension of the overall problem of over-centralization which has always been one of the main issues for civilization in different forms. See http://reddit.com/r/rad_decentralization http://reddit.com/r/rad_decentralization
kem 10y ago
I think so. There's a fine line that demarcates ML from data science from big data, for example, and it seems there's lots of chasing small differences that goes on as heat shifts from one fad to another. My guess is that a lot of people in one group could be moved to another, but rather than say "oh, you're working on X+epsilon, could you learn X and do that instead?" its "Oh you don't do X, we'll go hire someone who does X."

ilaksh 10y ago

One thing about AI research now is a lot of stuff seems to be built on pretty big local GPU clusters or pretty expensive AWS stuff. I had an idea the other day that could make that more convenient/affordable. The trick of course is coming up/applying different/new algorithms that don't rely on fast synchronization. Ideally that would be some kind of AI language that could describe/compress complex models/filters/parallel functions/etc. (https://github.com/runvnc/ailang https://github.com/runvnc/ailang) Then the nodes could exchange 'knowledge' in that special language. Combine something like turbo.js with a webrtc rpc swarm. Use rpc to pull out of task queue. So we can do GPGPU processing on web-based clusters in the browser. Can just use public gists with 'gpgpucluster' at end of description. Then use requirebin. Also use webtorrent/instant.io for imagenet example. Need to break imagenet up into a ton of little torrents. Then make an example that maybe just searches for certain classes like 'cat'. Example doesn't need to do actual machine learning, just store images in graphics memory pull them out when they search. And send them back. Maybe just use a subset of imagenet so that multiple clients can share the dataset and so work together to answer queries. EDIT: Thanks for the downvotes! I thought it was an interesting idea. But then again, the more interesting I think an idea is, the more likely it is to be downvoted all to heck. Hard for me to suppress my desire to share my ideas even though they are not appreciated.

Cacti 10y ago
It's hard to tell what you're asking for. There is no getting around the amount of computation required, and there are already plenty of systems to distribute workload the old fashioned way. In addition, while many projects rely on clusters of GPUs, a great deal of cutting edge research can be done on one or two decent commercial grade GPUs. A single new NVIDIA 1070 (and even better, the coming 1080 Ti) is cheap ($400) and is comparable to many good supercomputers in 2000, for example. An AWS with a K80 is only about 30 cents an hour at spot prices... after a week of training you could have yourself a state of the art result that wasn't even possible a mere few years ago. Not too shabby! Things have never been cheaper.