summerlight

8 ms·

61.
by summerlight 2y ago
> $250k is a staggering salary... not everyone lives in San Francisco. Or America for that matter. Still you need to spend at least $250K (which direct compensation would be close to $150K) to hire a competent browser dev. And I'm n
62.
by summerlight 2y ago
> 20 developers at $150k each = $3M It should be at least 100 devs at $250k each, which is still a severe underestimation. Note that there are many different types of mandatory expenses that roughly matches to the direct compensation, so
63.
by summerlight 2y ago
You severely underestimate the engineering cost of modern web browser. Assuming a sufficient value-addition fork, a team of 20 cannot even catch up the Chromium upstream. Good luck coming up with a new engine compatible with Chrome; MS trie
64.
by summerlight 2y ago
There are certain demands and if you don't do anything, those will be taken over by competitors and you lose controls. This is especially important for Google as they see LLM as a significant portion of future Cloud business and probab
65.
by summerlight 2y ago
And I think this is the problem. They have all the necessary pieces and are not yet very successful at stitching them together. Google has a very strong foundation and execution skills but failed to effectively govern it. Not sure if this i
66.
by summerlight 2y ago
I think reliable managers probably don't have enough time to write codes (unless they work 70~80 hours week), but at least they should be able to review other's code.
67.
by summerlight 2y ago
I think there's still significant investments into the project. Android is big enough so any sort of greenfield experiments cannot be done cheaply. Also there were so many custom variants of Linux based operating systems in Google and
68.
by summerlight 2y ago
Does ECC do anything with memory safety? This is about physical errors while the article is talking about software bugs. Those two are almost orthogonal.
69.
by summerlight 2y ago
I guess 1.5 Pro is too expensive to run this kind of product given it offers quite generous limits. More likely 2.0 Flash or Flash lite.
70.
by summerlight 2y ago
My experience is that Gemini works relatively well on larger contexts. Not perfect, but more reliable.
71.
by summerlight 2y ago
I guess 90% is for "benchmark", which is typically tailored to be challenging to parse.
72.
by summerlight 2y ago
It's a strict improvement over the technology before LLM, which usually just assigns random cryptic symbols.
73.
by summerlight 2y ago
Isn't it table-valued function? IIRC, the SQL standard still doesn't have it but it's almost universally supported extension across vendors.
74.
by summerlight 2y ago
If there's any company who can afford "real-time LLM training" at this moment, I'm 100% sure they will win this AI race since they probably have at least ~10x compute compared to competitors. Of course, no one can do tha
75.
by summerlight 2y ago
My take is that `auto` is basically a tool to reduce local redundancies rather than typing convenience. Rule of thumb: you should avoid `auto` unless it actually improves readability (e.g. significant reductions of syntactic redundancies),
76.
by summerlight 2y ago
Google seriously needs to scale up their generative models to all of crawling/indexing/ranking infrastructure. Their current ranking models are not capable of dealing with the next-gen web filled with 99% gen AI craps. I think the
77.
by summerlight 2y ago
This is because programming is not a work in a continuous solution space. Think in this way; you're almost guaranteed to introduce obvious bugs by randomly changing just a single bit/token. Assembler, compiler, stronger type syste
78.
by summerlight 2y ago
You may think in that way, but in order to make a such strong statement you probably want to do your homework first, like searching up the discussions and its rationale. Usually there is a good reason to pass a law if it's almost unive
79.
by summerlight 2y ago
When you write something like that, you probably want to add what are "far more important issues to worry over" and why pursuing this legislation would cause delay on that issues. And just in case you're not aware of, "f
80.
by summerlight 2y ago
This is a prevalent misconception that assumes advertisers don't care about how their money is spent! Advertisers and Google are actually concerned about SEO garbage. Nowadays, most advertisers tend to pay based on # of conversions, it
81.
by summerlight 2y ago
https://support.google.com/websearch/answer/9281931?hl=en They're already doing that with paid quality raters. I suppose your question is about opening this up more widely, but that's basically giving th
82.
by summerlight 2y ago
Amazon and Meta is at an early stage for their TPU equivalent and I don't think they're ready for production loads. Only Google has comparable silicons but I suspect even Google TPU are mostly for internal products rather than con
83.
by summerlight 2y ago
And... most of the important projects are still using proto2 and there is no realistic way to do interop between those two formats. IIRC, you cannot use proto2 enum in proto3 for some obscure technical reason! This creates a backsliding iss
84.
by summerlight 2y ago
> Think about it, almost everyone I know rarely clicks on ads or buys from ads anymore. This is a typical tech echo chamber. There is a significant number of people who make direct purchases through ads. > But if they can get good vid
85.
by summerlight 2y ago
This aligns with my knowledge. I don't know much about LLM, but TPU has been used for training deep prediction models in ads at least from 2018, though there were some gap filled by CPU/GPU for a while. Nowaday, TPU capacity is pr
86.
by summerlight 2y ago
I have no knowledge of Google specific cases, but in many teams smaller models are trained upon bigger frontier models through distillation. So the frontier models come first then smaller models later.
87.
by summerlight 2y ago
It is interesting to see that they keep focusing on the cheapest model instead of the frontier model. Probably because of their primary (internal?) customer's need?
88.
by summerlight 2y ago
This makes sense. If they don't outsource, they need to run millions of cars. This will cost Alphabet hundreds of billions capex, which is not cheap even for them. This is not just the money problem, but also has significant implicatio
89.
by summerlight 2y ago
While this is impressive, yet still looks like a very early prototype. The overall nuance seems that it doesn't try to be a standalone product but a part of broader R&D projects toward general agents... I doubt if they even have an
90.
by summerlight 2y ago
No, the constitution mandates the president to stand down. The cabinet approval is just a procedural one to make the boundary clear between authorities. The supreme court's interpretation is that if they don't approve it within a

More ›