Eridrus

28 ms·

421.
by Eridrus 3y ago
The article does seem to be punching a straw man though. A 60/40 portfolio is basically only recommended for those approaching retirement who cannot stomach high volatility. Target date funds that are commonly used would put 30 year ol
422.
by Eridrus 3y ago
All the big clouds are bad for this, GCP has actually been the best big cloud in my experience.
423.
by Eridrus 3y ago
Many advocacy groups (particularly those making claims about human trafficking) fudge their numbers to make their cause seem more important, and you can see why, we would probably not be discussing this article without such an eye catching
424.
by Eridrus 3y ago
I don't know about how many people keep holding, but we just experienced a 30%+ fall in tech stocks (and 25% fall in sp500) last year. I held all my stocks because, well, what else was I going to invest in? And if I don't have a
425.
by Eridrus 3y ago
The amount of money being spent on ML compute is pretty high. If it stays high, cost will be a relevant axis that others can compete on. GPUs are also not super available atm. We'll see if this is a temporary issue or one that persists
426.
by Eridrus 3y ago
Did you not read the article? He mentions this explicitly. What happens first is a lot of income restructuring to avoid paying those taxes: https://www.latimes.com/business/la-fi-nocera-tax-avoidance-... I don't t
427.
by Eridrus 3y ago
The answer is always "it depends", but <10% is the right ballpark. I would say 5-10% is probably what you would expect to go to employees depending on the specific trajectory things went. Generally, companies start by creating
428.
by Eridrus 3y ago
Be young and fit :p I used to be able to work from anywhere on the shittiest laptop hunched over on the shittiest desk... getting older and then less fit during the pandemic, and well, now I can't work without a good chair because my b
429.
by Eridrus 3y ago
Broadway musicals were disrupted a long time ago, what is left is crumbs for people who have a niche interest; these jobs will not be replaced by AI because it is already an anachronism that has been replaced by other entertainment for the
430.
by Eridrus 3y ago
Thanks!
431.
by Eridrus 3y ago
Is ALiBi still the sota for this setting, or have there been advances beyond this in the last 8 months? I know there has been a lot of interest in longer context lengths recently.
432.
by Eridrus 3y ago
This seems heavily impacted by the specific proxy being used for "engagement". In fact the article notes "In 2012 YouTube switched from maximizing clicks to maximizing watch-time. They found it led to a short-term decrease in
433.
by Eridrus 4y ago
I am aware of the limitations, but covering some predefined area with no human intervention is the definition of L4. Clearly there is non-zero development risk here, but I don't think it's going so badly that it's definitely
434.
by Eridrus 4y ago
Not even geohot is asserting that "SAE Level 2 is the most practical self-driving we are ever going to get", his assertion is that he thinks nobody is going to get there in the next 10 years. He may be right that we won't get
435.
by Eridrus 4y ago
I don't think they did, but celebrities are nothing if not PR machines, so I'm sure it was simpler to get this signed.
436.
by Eridrus 4y ago
In general, SWE/SRE do not have access to user data. SWEs just have no prod access at all, and SREs generally only have the ability to run signed binaries with code that has been code reviewed and submitted, though there are obviously
437.
by Eridrus 4y ago
The author (from their previous post) is primarily interested in US vs Rest of World comparisons. I don't think it's fair, but the reality is that companies pay what it costs, and in Europe there are fewer companies willing/a
438.
by Eridrus 4y ago
Is there any discussion on how this work will differ from Common Crawl?
439.
by Eridrus 4y ago
Given the fact that x86 instructions are not generally completed in a single cycle, how much of this increased clock do we actually expect to translate into faster performance? Presumably we're not expecting a 50% performance increase
440.
by Eridrus 4y ago
I barely come in to the office any more, and this has been fine for me, I know all the senior folks in my org, I have a big meaty project that I am driving that has appropriate staffing, and it's largely not from the city I live in, so
441.
by Eridrus 4y ago
Is everyone with relevant inference costs not doing this already? I am so confused how there seems to be a startup around having a work queue that does batching...
442.
by Eridrus 4y ago
ArXiv is fine for what it does, but it does not provide any sort of dissemination support. IMO the best thing about CS conferences is the poster track where you can walk by hundreds of posters, and the information is (when done well) much m
443.
by Eridrus 4y ago
I'm not really an expert on this, so I don't know why someone is doing this here, but it sounds like a DKIM replay attack: https://proton.me/news/dkim-replay-attack-breakdown
444.
by Eridrus 4y ago
The performance of different queries is different across a typical RDBMS and something like Spark/BigQuery. A traditional database is efficient (i.e. cost effective) at doing a lot of the same query repeatedly, e.g. looking up a custom
445.
by Eridrus 5y ago
I think this startup is the result of a tight labor market making employers look for less credentialed candidates, a general skepticism of credentials in the tech sector, and society's general fascination with "whiz kids". Si
446.
by Eridrus 5y ago
Beyond what makes a good model, IME at a FAANG building an ML product, the bulk of the work in practice tends to be general software engineering. There need to be a sufficient number of people who understand the actual ML pieces under the h
447.
by Eridrus 5y ago
I think an undervalued bit of value here is that everyone is guessing the same word every day, enabling the social sharing of "wordle plots". So even if you have a recreation, you need to own the canonical word list to gain the so
448.
by Eridrus 5y ago
The most compelling applications I have heard are for simulating quantum chemistry.
449.
by Eridrus 5y ago
There is a lot of defeatism floating around these days that isn't really warranted, there's good evidence that what you study matters as much as where, e.g. https://www.air.org/news/press-release/when-it-
450.
by Eridrus 5y ago
I feel like NER is a poorly designed task in general. You're eventually trying to link the entities to some kind of KB, so you should be injecting that entity information into your system for detecting mentions.

More ›