4 ms·

The "why" for AI right now is because everyone else is doing it and you can't risk not having a critical pillar needed to compete. I think AI is a loser for ev

The "why" for AI right now is because everyone else is doing it and you can't risk not having a critical pillar needed to compete. I think AI is a loser for everyone but Nvidia for a while because of this. Google, Meta, Apple, Microsoft, etc will all be forced to spend billions to attempt AI but it will be years before it is good enough to generate significant profits. Even then, there may only be 1-2 winners, like Google was with traditional search.

pennybanks 2y ago
amd is looking real nice right now as well.
scarmig 2y ago
There's also an interesting dynamic: costs scale exponentially (in a handwavey way) with quality. If you're willing to stay a half generation behind, you can create models a whole lot cheaper than the frontier labs. So Meta can always put significant price pressure on Google/OAI/Anthropic for relatively cheap. The only way I see AI making anyone trillions is if someone cracks the nut of true AGI and somehow manages to keep it under wraps.
- xracy 2y ago
  Here's a question... Does AGI actually change anything about how any of these companies make money? What makes someone trillions of dollars from AGI?
  - _boffin_ 2y ago
    > What makes someone trillions of dollars from AGI? solving many million dollar problems.
    - xracy 2y ago
      I don't think this question is as obvious as everyone else clearly thinks it is. But like here's the progression: 1. How does AI currently make money? (the answer is it doesn't, but like what's the world we think exists where a chat bot is monetizeable?) 2. What does AGI need to look like where it can actually solve many million dollar problems? 3. What million dollar problems does it solve? If the answers to this are as easy as they sound, then the responses to this should be easy to churn out, but instead everyone has shortcut that AGI == $$$, without showing the logic path. a response could look like: AGI solves P != NP -> $1M AGI solves customer service -> $100B AGI solves ??? -> $1B I'm making up ideas and numbers here, but this doesn't seem like that weird of an assumption to test given how everyone on here just takes it as gospel.
ozim 2y ago
I have a different bet. While AI from MSFT won’t be as ground breaking as hyped now, it still will be big enhancement for Office and Cloud offerings. Microsoft providing Word/Excel/Powerpoint with existing capabilities to big companies is going to be long run winner. NVidia is going to be always just like all hardware providers strong but still not in a place like MSFT having governments and F1000 corporations basically paying MSFT all the money they want.
mgl 2y ago
Well, I strongly believe there are already areas where LLMs' ability to flexibly transform e.g. textual content can bring lots of business value, e.g. in data reporting and analytics, see: https://www.linkedin.com/feed/update/urn:li:activity:7224046931377872897/ https://www.linkedin.com/feed/update/urn:li:activity:7224046...
throw0101b 2y ago
> The "why" for AI right now is because everyone else is doing it and you can't risk not having a critical pillar needed to compete. Basically the same "why" as the Dot Com stuff: you have to sell pet food online because everyone else is selling stuff online. At some point the hype will subside and we'll get the actual useful day-to-day applications of whatever technology is in vogue right now. Tech hype dates back to (at least) canal mania: * https://en.wikipedia.org/wiki/Technological_Revolutions_and_Financial_Capital https://en.wikipedia.org/wiki/Technological_Revolutions_and_... * https://www.pwlcapital.com/investing-technological-revolutions/ https://www.pwlcapital.com/investing-technological-revolutio...
onlyrealcuzzo 2y ago
Google doesn't really buy nVidia chips. They mostly use home grown TPUs. The main signals on Search have been AI/ML for >5 years now. At least for Google, AI isn't just a parlor trick.
- v7p1Qbt1im 2y ago
  They use both. Especially for their cloud customers. They offer Cloud GPU as well as Cloud TPU as a service on Google Cloud.
- derefr 2y ago
  > The main signals on Search have been AI/ML for >5 years now. Say more? "Passing the entire web through ML-model inference to generate 'signals' — IQL expressions? — at the same global-concurrent throughput as previous CPU-based indexing" sounds like something that should have required Google to get so many TPUs fabbed that it would have eaten the entire world's chip-fab capacity for multiple years. It's hard to imagine what they could have done to get around that. Did Google come up with some kind of model-hardcoded ASICs that could do this single task orders-of-magnitude more efficiently than more "flexible" GPU/TPU-based approaches would? Maybe, for example, they used some process to convert the weights of an arbitrary Transformer into VLSI for a search memory / memristor network / etc — essentially giving them an inferencing DSP? Or maybe they figured out a mostly-lossless method (that either doesn't use ML itself — or maybe uses ML hyper-optimized into some CPU-viable model that fits in L2 cache) to pre-process + normalize webpage documents into chunks that could be effectively content-deduplicated? (And then the ensuing inference passes — being context-free — would only ever need to be done once for a given chunk, significantly reducing the inter-page costs, and massively reducing the same-page over-time costs, of inference.) Or something else I can't even think of. (I'm guessing it's this one.) Either way, crazy stuff.

pennybanks 2y ago

amd is looking real nice right now as well.

scarmig 2y ago

There's also an interesting dynamic: costs scale exponentially (in a handwavey way) with quality. If you're willing to stay a half generation behind, you can create models a whole lot cheaper than the frontier labs. So Meta can always put significant price pressure on Google/OAI/Anthropic for relatively cheap. The only way I see AI making anyone trillions is if someone cracks the nut of true AGI and somehow manages to keep it under wraps.

xracy 2y ago
Here's a question... Does AGI actually change anything about how any of these companies make money? What makes someone trillions of dollars from AGI?
- _boffin_ 2y ago
  > What makes someone trillions of dollars from AGI? solving many million dollar problems.
  - xracy 2y ago
    I don't think this question is as obvious as everyone else clearly thinks it is. But like here's the progression: 1. How does AI currently make money? (the answer is it doesn't, but like what's the world we think exists where a chat bot is monetizeable?) 2. What does AGI need to look like where it can actually solve many million dollar problems? 3. What million dollar problems does it solve? If the answers to this are as easy as they sound, then the responses to this should be easy to churn out, but instead everyone has shortcut that AGI == $$$, without showing the logic path. a response could look like: AGI solves P != NP -> $1M AGI solves customer service -> $100B AGI solves ??? -> $1B I'm making up ideas and numbers here, but this doesn't seem like that weird of an assumption to test given how everyone on here just takes it as gospel.

ozim 2y ago

I have a different bet. While AI from MSFT won’t be as ground breaking as hyped now, it still will be big enhancement for Office and Cloud offerings. Microsoft providing Word/Excel/Powerpoint with existing capabilities to big companies is going to be long run winner. NVidia is going to be always just like all hardware providers strong but still not in a place like MSFT having governments and F1000 corporations basically paying MSFT all the money they want.

mgl 2y ago

Well, I strongly believe there are already areas where LLMs' ability to flexibly transform e.g. textual content can bring lots of business value, e.g. in data reporting and analytics, see: https://www.linkedin.com/feed/update/urn:li:activity:7224046931377872897/ https://www.linkedin.com/feed/update/urn:li:activity:7224046...

throw0101b 2y ago

> The "why" for AI right now is because everyone else is doing it and you can't risk not having a critical pillar needed to compete. Basically the same "why" as the Dot Com stuff: you have to sell pet food online because everyone else is selling stuff online. At some point the hype will subside and we'll get the actual useful day-to-day applications of whatever technology is in vogue right now. Tech hype dates back to (at least) canal mania: * https://en.wikipedia.org/wiki/Technological_Revolutions_and_Financial_Capital https://en.wikipedia.org/wiki/Technological_Revolutions_and_... * https://www.pwlcapital.com/investing-technological-revolutions/ https://www.pwlcapital.com/investing-technological-revolutio...

onlyrealcuzzo 2y ago

Google doesn't really buy nVidia chips. They mostly use home grown TPUs. The main signals on Search have been AI/ML for >5 years now. At least for Google, AI isn't just a parlor trick.

v7p1Qbt1im 2y ago
They use both. Especially for their cloud customers. They offer Cloud GPU as well as Cloud TPU as a service on Google Cloud.
derefr 2y ago
> The main signals on Search have been AI/ML for >5 years now. Say more? "Passing the entire web through ML-model inference to generate 'signals' — IQL expressions? — at the same global-concurrent throughput as previous CPU-based indexing" sounds like something that should have required Google to get so many TPUs fabbed that it would have eaten the entire world's chip-fab capacity for multiple years. It's hard to imagine what they could have done to get around that. Did Google come up with some kind of model-hardcoded ASICs that could do this single task orders-of-magnitude more efficiently than more "flexible" GPU/TPU-based approaches would? Maybe, for example, they used some process to convert the weights of an arbitrary Transformer into VLSI for a search memory / memristor network / etc — essentially giving them an inferencing DSP? Or maybe they figured out a mostly-lossless method (that either doesn't use ML itself — or maybe uses ML hyper-optimized into some CPU-viable model that fits in L2 cache) to pre-process + normalize webpage documents into chunks that could be effectively content-deduplicated? (And then the ensuing inference passes — being context-free — would only ever need to be done once for a given chunk, significantly reducing the inter-page costs, and massively reducing the same-page over-time costs, of inference.) Or something else I can't even think of. (I'm guessing it's this one.) Either way, crazy stuff.