I said for a few years to many a downvote on HN, everyone wants AI, nobody wants to pay the true costs, the AI race will turn into a "race to the bottom" that is, who can give you the most compute for the lowest cost, and still remain profitable?
For what? There are some things I want AI for because it does it well. There are some things I don't want AI for because it just makes a mess (hallucinations). Maybe the next AI will be different and we will have the conversation again.
Indeed, it kills our planet, our culture and our economy extremely well. Oh yes, and some code monkeys enjoy that it can make computer code on the side.
> I said for a few years to many a downvote on HN, everyone wants AI, nobody wants to pay the true costs, the AI race will turn into a "race to the bottom" that is, who can give you the most compute for the lowest cost, and still remain profitable?
I keep seeing this but this line of thinking doesn't make any sense. What does it really mean?
There are expensive models that increase the probability of you doing your task under a lower cost. That means you can't use Gemma for coding your new compiler - it would just be overall costlier.
Heavier models are cheaper at more complicated tasks because they use fewer turns and fewer mistakes.
Cheaper models are more likely to be cheap at less complicated tasks. Like if you just ask Gemma "Hi" it would probably be cheaper than asking Opus.
So what does this statement really mean? People don't want to pay the extra for a more costly model? Why wouldn't you? It reduces your overall cost!
Because real Fable usage starts at $20/month, and has oppressive usage limits even at that (ridiculous) monthly price.
Compared to my $3/month GLM-5.2 subscription, I have never felt like I was leaving capabilities on the table by refusing to cough up $20 for 15 minutes of Fable use per day.
This is the wrong way to look at it. If you have a complicated task , you can solve it for cheaper if you used Fable. It will use fewer turns to achieve the same result.
You can solve it for cheaper if you use GLM but if you are involved in it more, but that defeats the purpose.
> And gemma downloads also can be from auto CI pipelines etc. Nothing concrete
I have always found NPM download numbers truly suspect. Is no one caching? Are they estimating true number of downloads base on some estimate of cache hits?
> Lastly, after an incredible 27-year run, Jeff Dean is at a moment where he wants to try something new, and we’re excited to support him in that. Jeff and Google Senior Fellow Sanjay Ghemawat are launching an independent public benefit corporation to accelerate discoveries in ML, science, and engineering.
Oof. Good for Jeff and Sanjay (who just joined Twitter), bu that is a big loss for Google. Google stock is down 5%. It might not be much of an exaggeration to say these two are worth ~$200 billion.
Whoa. So is Jeff effectively leaving Google to work on this new venture full time? Or is the venture a side project? It sounds like the former. I’m sure he’ll still have internal access as an advisor of sorts. But this feels like a seismic change. Much larger than I initially realized?
For at least a year now the standard reflexive reply to “Google seems way behind OAI and Anthropic” has been “it’ll be ok, they’ve got Dean and Hassabis.” And now they don’t. What reason is there to be bullish about Google now?
As much as everyone wants to pay accomplished celebrities, all of these companies have young nameless geniuses that are about to make one for themselves. A guard passing torch can be opportunity.
Now, whether Google is the right environment to nurture, that’s its own quandary.
Couple of things come to mind. One is that Web sites tend to actively fight AI companies' crawlers while actively courting Google's crawler.
Another is that they hold the key Transformer architecture patent. If it is still relevant (which I'm not personally clueful about) and if they start enforcing it, then we may see a reprise of the situation where Microsoft made money for years every time an Android phone was sold. Regardless of that particular patent, it's probably safe to say they'll be better-positioned than anyone else if AI companies start lobbing patent nukes at each other.
A third factor that shouldn't be discounted is that Google has access to warehouses of training data that other companies don't. Google Books alone is an Alexandria-scale archive that the courts forced them to keep to themselves. Those restrictive copyright decisions may turn out to be a blessing in disguise for Google because no one else will have been able to scrape the data.
It just tells you about their intended audience: Investors rather than consumers.
Nowadays, every company—even huge ones—prefers to be seen as a "growth opportunity", so they are trying to play up their ability to create new products which will somehow be so incredible that the line keeps moving up forever.
That said, there is definitely a correlation between companies chasing new products and leaving old ones to become crap.
I'd bet my mortgage that if the first sentence was "We've got amazing products, amazing talent, and world-class compute", you'd be in the comments complaining that they didn't put talent first.
Hasn't that always been the case with Google? Outside of a few that are mostly good and have stuck around, I've always seen Google has having great tech but being quite bad and making products out of it.
If all the other AI companies are promising AGI by Q4 of next year, what else can you do to satisfy shareholders than also jump on that same bandwagon?
This is a promotion for Demis and this could be a path for Demis to be CEO of Alphabet in the future in the AI era.
Google already invested in Discovery Loop (Jeff Dean, Orol Vinyals, Quoc Le and Sanjay Ghemawat's company), so what is happening is still a win in investment terms.
The question is about Sundar's future at Google, he is a mobile era CEO at Alphabet and I would hazard a guess he will probably step down in less than 3 years.
I don't see Demis becoming CEO. He's a scientist and researcher. He wouldn't want to be bogged down by minutiae of corporate politics, org structure, government relations, mobile hardware, etc. Chair lets him have authority to explore any path of interest without overhead of operations.
My impression, inside and out of G, was that Sundar (and Ruth) were about scaling down R&D expenses (as a fraction of revenue), and focusing on exploiting the monopolies.
Perhaps now there could be a shift back to investing in R&D to get fresh monopolies.
That's the fun part. We don't. Could have happened 10 years ago without us noticing. I don't expect my skin cells to be able to recognize "me" anymore than I expect we will be able to recognize a superintelligent AGI. If it arrived 10 years ago, then the last 10 years could simply be its PR campaign. We wouldn't even be able to tell if it was successfully achieving its "goals" or not, assuming an AGI even has "goals"
Can we please stop saying anything about "AGI"? I remember when people would be like "AGI in 3 months" / "AGI in 2024" / "AGI is confirmed in 2025" like stop, you don't know if it's even a thing, let alone if it's coming/imminent.
Considering these are the best stats they could find, gemini usage+general situation must be really, really bleak.
High demand means nothing. A model being live is nothing to brag about. And gemma downloads also can be from auto CI pipelines etc. Nothing concrete
I said for a few years to many a downvote on HN, everyone wants AI, nobody wants to pay the true costs, the AI race will turn into a "race to the bottom" that is, who can give you the most compute for the lowest cost, and still remain profitable?
But people want what AI does for them. It's a tragedy of the commons situation.
I keep seeing this but this line of thinking doesn't make any sense. What does it really mean?
There are expensive models that increase the probability of you doing your task under a lower cost. That means you can't use Gemma for coding your new compiler - it would just be overall costlier.
Heavier models are cheaper at more complicated tasks because they use fewer turns and fewer mistakes.
Cheaper models are more likely to be cheap at less complicated tasks. Like if you just ask Gemma "Hi" it would probably be cheaper than asking Opus.
So what does this statement really mean? People don't want to pay the extra for a more costly model? Why wouldn't you? It reduces your overall cost!
Because real Fable usage starts at $20/month, and has oppressive usage limits even at that (ridiculous) monthly price.
Compared to my $3/month GLM-5.2 subscription, I have never felt like I was leaving capabilities on the table by refusing to cough up $20 for 15 minutes of Fable use per day.
You can solve it for cheaper if you use GLM but if you are involved in it more, but that defeats the purpose.
I have always found NPM download numbers truly suspect. Is no one caching? Are they estimating true number of downloads base on some estimate of cache hits?
Oof. Good for Jeff and Sanjay (who just joined Twitter), bu that is a big loss for Google. Google stock is down 5%. It might not be much of an exaggeration to say these two are worth ~$200 billion.
Now, whether Google is the right environment to nurture, that’s its own quandary.
Another is that they hold the key Transformer architecture patent. If it is still relevant (which I'm not personally clueful about) and if they start enforcing it, then we may see a reprise of the situation where Microsoft made money for years every time an Android phone was sold. Regardless of that particular patent, it's probably safe to say they'll be better-positioned than anyone else if AI companies start lobbing patent nukes at each other.
A third factor that shouldn't be discounted is that Google has access to warehouses of training data that other companies don't. Google Books alone is an Alexandria-scale archive that the courts forced them to keep to themselves. Those restrictive copyright decisions may turn out to be a blessing in disguise for Google because no one else will have been able to scrape the data.
>We’ve got amazing talent, world-class compute and products…
Products are third on the list. Google is an incubator for talent first and foremost. Products are an afterthought
Nowadays, every company—even huge ones—prefers to be seen as a "growth opportunity", so they are trying to play up their ability to create new products which will somehow be so incredible that the line keeps moving up forever.
That said, there is definitely a correlation between companies chasing new products and leaving old ones to become crap.
Founders are first. Ideas are second.
1) Repair search that had been broken by AI initiatives.
2) Include less invasive AI with ads for those that need to be spoon fed.
3) Pretend to work on AGI and data centers in space.
4) Sell shovels and TPUs to the gold diggers.
Whatever happened to Prabhakar Raghavan? Got kicked upstairs and we barely hear from him nowadays.
It's only a matter of time before Demis leaves and joins Anthropic or OpenAI.
This is a promotion for Demis and this could be a path for Demis to be CEO of Alphabet in the future in the AI era.
Google already invested in Discovery Loop (Jeff Dean, Orol Vinyals, Quoc Le and Sanjay Ghemawat's company), so what is happening is still a win in investment terms.
The question is about Sundar's future at Google, he is a mobile era CEO at Alphabet and I would hazard a guess he will probably step down in less than 3 years.
His current title is CEO of Google DeepMind. Becoming Chief Scientist of Alphabet seems to be a step away from the path to replacing Sundar, no?
Perhaps now there could be a shift back to investing in R&D to get fresh monopolies.
He was the guy branding Google an AI-first company back when they invented the transformer.
These guys...