Whats your recommendation for backend engineers that want to transition to ml engineers?
If you contribute to open source (tensorflow, pytorch) does that help? What is your day to day like? What minimum skill set do you need to function as MLE? Does taking ML courses in coursera help?
That's a lot of questions so I'll try to answer one by one.
> Whats your recommendation for backend engineers that want to transition to ml engineers?
I get this a lot, and my first question would be, why do you want to work in ML? Is it because it's "cool"? because it's in high demand? because it pays a lot? for the second two points, see my comment. For the first point, I'm pretty cynical about it. At the end of the day it's a dev job not very different from other domains. Depending on where you end up, your day to day might look the same, but with focus in ML systems.
Nevertheless, if you really want to transition, I still think the easiest way is to do an internal transfer to a team that does ML. That's how I got this job. If you are a good engineer, most ML teams will be happy to have you and teach you the skills on the job.
Of course maybe there is no ML team in your current company. If you are in a small company you might try to influence leadership to start ML initiatives and get some experience. I've seen many candidates do this: they start an ML project in their current company, and after 6-12 months they look for a job elsewhere. This is a variant of CV driven development, and sadly often these initiatives are hamfisted and don't really solve business problems.
Otherwise you might try to get a job in ML in another company. Lucky for you I found some companies that hire engineers in ML or ML adjacent jobs using a standard software dev loop, i.e. no special ML round, although these difficult to find. The other option is to try to learn enough ML to pass an interview.
I won't lie, this is doable, but the deck is stacked against you. The job market is pretty insane right now, and you'll be competing with people with masters and PhD's in ML, plus engineers with years of experience in the field.
I believe it's fair to say that pure theoretical knowledge won't get a job, much less unless you have prestigious credentials. Instead I would focus on a blend of applied theory and personal projects. Probably it would be best to choose one ML framework and double down on it to increase your chance of finding a job that requires that particular framework.
For theory there are tons of courses and resources online, although I tend to like more walkthroughs and tutorials that guide you on building a system rather than pure theory. For practical experience, the sky is the limit. As I said there are tutorials online that guide you on a project, but that will almost certainly not be enough. You can try your hand at Kaggle competitions, and at some point you can just choose some personal project that interests you and try to build on it.
I hope any of this helps and best of luck to you. Feel free to DM me if you want to know more.
I'm from Pinecone. We use proprietary indexes. We could've used HNSW but decided the high memory consumption (ie, costly at scale) and slow index updates (ie, data gets stale) won't cut it for production use cases.
Oh I read a lot of HNSW stuff on your/Pinecone blog series. (Great learning resource btw, well done!) So I assumed you were using HNSW already. It's a news to me that you don't use it.
Because even in china they can't build nuclear as quickly (and likely as cheaply) as wind and solar. It's hugely different [0][1]. Over the last 10 years, they used 3.5x more energy (so not nameplate capacity, but actually generated energy) from solar and wind than they did from nuclear.
Further, if you download the data [1] an look at the derivative you get an even more grim picture. Not only is wind and solar increasing rapidly, but the derivative is positive (and possibly increasing itself). But when you look at nuclear, you find that not only is it not increasing as rapidly, but it's derivative is going down and heading for zero fast.
They can however, build nuclear plants wherever they want, which is why China is building them at all right now. They have plenty of renewable potential in the west, but it’s limited in the more populated east, and their transmission lines meant to carry power west to east over long distance are still limited in capacity.
I believe they mean grim if you are a nuclear cheerleader.
But it is pretty damning that a country that is motivated and willing to put its money where its mouth is, is also having troubles making it happen quickly.
> California is side stepping the whole mess with nuclear by building out solar and batteries.
I like all non-CO2 producing energy sources: nuclear, solar, wind, etc.
For any given energy source, only a subset of territories can use it: solar needs sun (hello Alaska and Wyoming), nuclear needs strong governmental, engineering and public oversight, wind, like sun, is also not universally available.
Specifically California can do all of them, which puts it in a unique position to be choosy. I believe that by going with sun+batteries, California will pave the way for less rich but sunny territories for energy independence / less CO2 pollution.
It's okay in my mind to leave betting on nuclear to the states which don't have that much sun. I am looking forward to the TerraPower plant in Wyoming being built and hopefully more to come after it.
Solar works really well in the Alaskan summer. Winters you have to figure something else out. Alaska has lots of hydro, which is really common in that region.
So 5GWh of storage capacity would be about 0.6% of average daily usage. We're going to need a lot more than that to make renewables viable for base load. How much will that cost, how long will it take to build, and where will we mine the raw materials?
Once there was a remote mountaintop monastery, nestled high in the clouds above a deep gorge. To reach the monastery, visitors were required to ride in a basket pulled up from the top. One day, halfway up the cliff, a tourist frowned at the condition of the rope. He asked his guide, "How often do they replace this rope, anyway?" The monk answered "Why, as soon as the old one breaks, of course."
Once there was a remote mountaintop monastery, nestled high in the clouds above a deep gorge. To reach the monastery, visitors were required to ride in a basket pulled up from the top, after paying the ticket fee for the ride. One day, halfway up the cliff, a tourist noticed the sparkling condition of the titanium cable upon which they were suspended. He asked his guide, why did it cost me ten years salary to take this ride? The monk answered "If we didn't charge that much, we couldn't replace the cable after every trip". The tourist asked "is that really necessary"? The monk replied "Well, we used to use a hand woven hemp rope, and it broke once during a trip. We wouldn't want that to happen again, now would we?"
Point being, it's stupid to refrain from building new nuclear plants because they're "dangerous", because that forces the old plants to be operated far past their intended service life. Better to replace the rope before it breaks, so to speak.
In a rational world, those who believe atomic energy is inherently scary would be the first ones demanding that the old plants be replaced.
> Once there was a remote mountaintop monastery, nestled high in the clouds above a deep gorge. To reach the monastery, visitors were required to ride in a basket pulled up from the top. One day, halfway up the cliff, a tourist frowned at the condition of the rope. He asked his guide, "How often do they replace this rope, anyway?" The monk answered "We have a large team of monks who continuously monitor the rope using advanced technology. We actually invented the field of reliability analysis to help us keep this rope safe. If we notice anything wrong with the rope, we immediately repair it. The rope is guarded by heavily armed guards to make sure no one sabotages it."
Best of luck to your job search, god speed