Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
riku_iki
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
37 ms
·
811.
▲
by
riku_iki
2y ago
there are some multi-millionaires and billionaires clubs likely
812.
▲
by
riku_iki
2y ago
> I’m guessing you are in denial that we can make a simulated reasoning machine? some people actually try, and see that LLMs are not there yet
813.
▲
by
riku_iki
2y ago
> The TPU is basically an ASIC as far as I know; it competes against CUDA in a very small subset of it's featureset. CUDA is essentially a composition layer on top of multiple GPU features that optimizes them for general-purpose com
814.
▲
by
riku_iki
2y ago
> I hate the industry for never learning how to get along and make a real CUDA competitor. there are google TPUs. Do they provide better pefrormance/dollar, or google also charges high margin, or Nvidia is doing some unique optimiza
815.
▲
by
riku_iki
2y ago
> I also became noticeably weaker when it came to lifting weights you may need up to 6 month for full keto adaptation according to some internet accounts.
816.
▲
by
riku_iki
2y ago
Because deep ocean game is very different, those cruisers will be moving constantly with 60mph speed in much larger area than black sea, so one need more robust intelligence system and higher speed and range to catch them.
817.
▲
by
riku_iki
2y ago
Austin has area for expansion, and SF not much..
818.
▲
by
riku_iki
2y ago
regardless of propaganda, if the facts are correct it is the same speed how US can build cruisers.
819.
▲
by
riku_iki
2y ago
its unclear how efficient they are in deep ocean.
820.
▲
by
riku_iki
2y ago
Could you tell how scalable JVector is? How many vectors it can handle, like millions, billions, hundreds of billions?
821.
▲
by
riku_iki
2y ago
> here are a lot of different hash functions. Cryptographic hashes, locality sensitive hashes, hashes for checksum, etc. and there are some standard hash functions in the lib, which cover 98% of usecases. I think the same is embeddings,
822.
▲
by
riku_iki
2y ago
sure, but somehow distribution and reckless do not impact export prices?
823.
▲
by
riku_iki
2y ago
> while everybody (Apple included!) is already working on building hardware to avoid using Nvidia's H100s. the question is if they will succeed. Only Google so far has competitive product(TPUs).
824.
▲
by
riku_iki
2y ago
> Making a car is easy, manufacturing a car at scale it looks like tesla manufactured cars just fine, there is no shortage of teslas on market. Demand on EV didn't pick up.
825.
▲
by
riku_iki
2y ago
Or decrease. It looks like there is 200B industry is already sucking funds, but no end-user app with strong revenue stream is on the market yet.
826.
▲
by
riku_iki
2y ago
What is the end task(e.g. RAG, or just vector search for question answering), are you satisfied with results in terms of quality?
827.
▲
by
riku_iki
2y ago
I think it will be (much) more expensive, but I am not an expert.
828.
▲
by
riku_iki
2y ago
> They already have a profit cap at 10% of revenue. what about cap on very high executive bonuses?
829.
▲
by
riku_iki
2y ago
> California exports a ton of energy during the bright months. cool, but prices for electricity in California still somehow 3 times higher than places where electricity is exported?..
830.
▲
by
riku_iki
2y ago
> Hypothetically, what could they do against a potential script that pre-downloads all new videos from my subscriptions apply some robot detection tech and block robots/scripts
831.
▲
by
riku_iki
2y ago
they will eventually will block that client too if it become too popular.
832.
▲
by
riku_iki
2y ago
apt search would return tons of libparquet-java/c/python packages if it was popular.
833.
▲
by
riku_iki
2y ago
Likely this assumes that parquet has internal compression applied, and CSV is uncompressed.
834.
▲
by
riku_iki
2y ago
> "By default" means the most natural and common way to index an array, namely the [] syntax. Sorry, I challenge your authority to decide what is "most natural way".
835.
▲
by
riku_iki
2y ago
> I think your experience might be due to -- what it sounds like -- parsing the entire CSV into memory first (CREATE TABLE) and then processing after. That is not an apples-to-apples original discussion was about CSV vs parquet "rea
836.
▲
by
riku_iki
2y ago
> The parquet file has metadata that allows duckdb to only read the parts that are actually used, reducing total amount of data read from disk/network. this makes sense, and what I hoped to have. But in reality looks like parsing CS
837.
▲
by
riku_iki
2y ago
> Honestly some of the best engineers at DM. like who? > Unsurprising really given that he was offering ~$10M/year in com where did you get this from?
838.
▲
by
riku_iki
2y ago
It sounds like he did published some results at pre-LLM era before 2017, and there was silence after that.
839.
▲
by
riku_iki
2y ago
Tesla and SpaceX are undeniable successes of peak Mask at his core expertise. Boring and Neiralink are just valuations with unknown revenue base and future which may be not justified, and weak proof to justify another valuation.
840.
▲
by
riku_iki
2y ago
> which works for CSVs that fit in memory. what? Why CSV is required to fit in memory in this case? I tested CSVs which are far larger than memory, and it works just fine.
More ›