riku_iki

40 ms·

781.
by riku_iki 2y ago
didn't catch hype and big VC money..
782.
by riku_iki 2y ago
> just have to plan and do it in the background this can mean building some untrivial infra, so the task is much more complicated than just using cloud vm. My point is that bottle neck and why you need cluster and not single VM is likely
783.
by riku_iki 2y ago
Then 3tb of total data volume is irrelevant. What relevant is size of column.
784.
by riku_iki 2y ago
Base on numbers you put, you would need to have stable 50-100GB/s speed to your disc system to just read 3TB data in 40-60sec. Would you mind sharing your setup?
785.
by riku_iki 2y ago
> A trillion rows of indexed data will fit on a single cloud VM and be reasonably performant. create and subsequently update index on trillion rows is very untrivial task performance wise.. Which DB you would suggest to use for this?
786.
by riku_iki 2y ago
> The general feedback we got talking to folks in the industry was that 100 GB and then user discovers that DuckDB is plagued with OOMs and dramatic performance degradations when his data is slightly larger than memory.
787.
by riku_iki 2y ago
I imagine it could be cool for some DB vendors too, so they can optimize base on direct access and skip all OS/FS overhead.
788.
by riku_iki 2y ago
its great starting insight, but again its small dataset (100GB) which almost fits memory, and I think many details are missing (for example clickbench publishes all configs and queries, and more detailed report, so vendors can reproduce&#x2
789.
by riku_iki 2y ago
clickbench is a toy benchmark: small dataset, very specific queries. Benchmarking full tcp-h (not just one query like in your post) on sizable dataset (few TBs) would be very good close to real world scenario, but vendors usually avoid this
790.
by riku_iki 2y ago
> which benchmarks faster than DuckDB for OLAP [1] that link doesn't do any performance comparison. They claim that CedarDB executes less "code branches" than duckdb, which may or may not translate to faster performance.
791.
by riku_iki 2y ago
> There were never any promises, and Putin barely even cared that it resulted in Finland and Sweden joining NATO. you don't know for sure. The only evidence is some notes from brit diplomats I think which refer on such promise.. Als
792.
by riku_iki 2y ago
In this case hardware is significant bottleneck: LLMs need lots of compute to strive.
793.
by riku_iki 2y ago
I don't understand why wouldn't google package their TPUs into some servers and sell it to consumers to grab big piece of this cake..
794.
by riku_iki 2y ago
> 2) The s-1 filings are prepared and audited by third party companies. does it completely release liability?
795.
by riku_iki 2y ago
> But when the CEO lies on a resume, nothing happens. this can be considered as essential job function, so checking the quality of lying is a part of interviewing process.
796.
by riku_iki 2y ago
how other providers (gmail) work in this case?..
797.
by riku_iki 2y ago
> as you'll see in the Google doc linked from the announcement, they are referring to the use of computer systems to assist mathematical research, which falls into a two step process I reviewed the doc, and it is just curated list
798.
by riku_iki 2y ago
> I wonder how long NVIDIA can justify its current market cap once people realize just how cheap it is to run inference on these models given that LLM performance is plateauing next wave driving demand can be actual new products develope
799.
by riku_iki 2y ago
More like Mark gave them 100k GPUs, and they are not sure what exactly to do with them..
800.
by riku_iki 2y ago
> I would expect nobody to escalate over a Taiwan Invasion. there is semiconductor industry on the table. I think there is high chance NATO will be suppressing invasion forces through launching anti ship missiles from aircrafts and cruis
801.
by riku_iki 2y ago
> AI capability has been following an exponential growth curve Math benchmark: Minevra, Jun 2022: 50.3% Opus, Mar 2024: 60.1% And there is a high chance it leaked to Opus training data, since it is old and now popular benchmark.
802.
by riku_iki 2y ago
> but I highly doubt fans would show up to watch robo-cars race you must be talking about humans. AIs may show up to watch robo-cars races.
803.
by riku_iki 2y ago
> The loser would go nuclear. it depends if loss will be significant enough to justify mutual annihilation. Assume Russia attacked Finland, and NATO started military operation and lost. It will be very unlikely France, Brits and American
804.
by riku_iki 2y ago
I am wondering if such moderated model can induce significant security risk, for example it can generate exploitable code, or try to trigger some action base on some specific input.
805.
by riku_iki 2y ago
I think the point of the doc is that "human imperfection" is very prominent concept in trained LLM..
806.
by riku_iki 2y ago
The worrying part is that first concept in the doc they show/found is "human imperfection". Hope this is just coincidence..
807.
by riku_iki 2y ago
unless there are some automigration tools are in place..
808.
by riku_iki 2y ago
> Rust compilation, WASM compilation, Docker builds shell scripts that produced a VM image, Python was in there somehow too was it through bazel rules? The worry is if some of those rules will get bug or missing necessary feature, it wil
809.
by riku_iki 2y ago
> Most of Google's revenue, to pick one example. For every flashy start-up or cool open source project, there is an army of people quietly delivering 10% relative improvements that drive $100M revenue lift in their particular niche
810.
by riku_iki 2y ago
> The market is "pricing in" this future potential. or short gain opportunity with goal to quite before crash.

More ›