15 dollars per square kilometer is quite affordable for this level of resolution! Is there any info about satellite coverage or plans that you can share?
Do you get good results when adding scraped youtube audio? My model performance on LibriSpeech dev drops a bit when adding youtube audio to the training dataset ( my guess is likely due to poor alignment from auto generated captions ).
I haven't trained on LibriSpeech exclusively, but yes, the perf on LibriSpeech dev is quite bad, around ~60.0 WER. If the poor alignment of yt captions is the issue, maybe concatenating multiple samples helps a bit.
Cool project, seems like your model have similar WER as mine (4th reference in readme). Do you plan to do any pre-training on the encoder part in the future? Maybe something like this[1]
Hey black cat, I have some work in preprint for a NeuroIPS workshop, demonstrating negative results of different audio distances on pitch tasks. There is one particular w2v result I'd like your feedback on.
Do you mind emailing me? Lastname at gmail dot com (see my profile for my name)
Does Intel only sold the NAND manufacturing parts to SK Hynix or it also includes the SSD controllers IP? Because most of Intel SSD performance came from the controller not the NAND flash themselves, in fact Intel usually lag behind in NAND flash manufacturing.
It will probably include the controller IP. However, none of Intel's IP will be transferred until 2025. That should give them time to disentangle any of the IP that's shared between the NAND SSD controllers and the Optane SSD controllers, and give Hynix time to decide whether it's worth continuing development of both their existing enterprise SSD controllers and Intel's designs.
Just my 2cents, alot of people mentioned alot criteria which I heard during my days as (empathy : backend engineer, abstract thinking : android dev, communication: company tech lead). But I think the core is the mindset to self improve that's best for you right NOW (self discipline).
This is really really really hard to do, since we are mostly lazy beings who rather click 50 arrow up key to find the one important command than remembering it or just dont mind to repeat a mundane process without finding a shortcut to do it faster. Programmers who I met that manage to finish a task in 10 minutes instead of 1 day usually have this trait. But I guess this applies to other jobs as well ¯\_(ツ)_/¯
Btw, I used to work as a part timer at my university department. One day one of the staff ask me to reformat about 90 Excels document to her desire format, she spend 10 minutes showing me how to type them out one by one. After this, I went to setup Python environment in that computer and google how to read excel docs. In a few hours I have finish the work that what would took me more than half a day to finish.
EE Master here, this is a wide topic but usually from my experience (friends around who work in chip industry, ie TSMC, MediaTek).
Generally, you have the hardware design (Layout designer, R&D, Process engineer ) this typically need a Master and mostly PhD in EE or related fields. From what I heard in Taiwan, TSMC only accepts Master/PhD for these position and internship.
Then you have the interface side which writes firmware codes for these chips ( SSD controler ). These are more like your typical software engineering, you will need at least BC degreee and well verse in C/C++. But usually it's hard to get in with a BC degree unless you are recommended by one of the employee.
If you don't have any of the mentioned degree, the fastest way is to get one. It is really hard to learn yourself since this is a highly automated industry with most of the work available in maintainence and cutting edge R&D.
Otherwise, your job scope in chip industry is limited to a few ( Firmware is a good start )
> Then you have the interface side which writes firmware codes for these chips ( SSD controler ). These are more like your typical software engineering, you will need at least BC degreee and well verse in C/C++. But usually it's hard to get in with a BC degree unless you are recommended by one of the employee.
Idk how one gets in at all seeming as how the few jobs I've seen in this sort of function want a decade of experience writing firmware for the same class of device as whatever the company is doing.
For neural networks unless it's calibrated otherwise the confidence score has little meaning to the accuracy. [1] see figure 4 and you will understand what I mean
Adsense has one small info button on the top right where you can click and report, however, for Youtube video ads, there's no visible way to report them as well, especially softcore porn. What's worse is that Youtube doesn't care whether the current audience is suitable or not as long as the ads were shown at the intended target ( ie gaming related video )
I did not know about this device so I did a quick research: it's essentially an optimised version of early plasma devices that use instabilities to create a very compact, dense structure that can reach really high ion temperatures. The main problem I see is that this structure is very short-lived: on the order of 10 nanoseconds.
These devices existed for a long time and have been used as X rays or neutron sources but not for fusion power. LPP have explained the exact machanism at the basis of the DPF fusion and have optimised the radiation losses and electron heating to reach higher temperatures but from this to claiming viable fusion power is quite a stretch.
Then there's this bit: "switching to pB11 fuel will give us a 100-fold increase in yield" -- compared to deuterium. That's straight-up backwards. pB11 power density is actually more like 100 times less. See https://en.wikipedia.org/wiki/Nuclear_fusion
Yes, he's published papers about his experiments, but there's nothing in them about how he's going to get to net energy output -- so that part is not peer reviewed.
