359 karma · joined August 24, 2020
But not searching to see if something had been posted before on a forum is a bit different[0]
My hypothesis (that in NA centric but may apply globally) has two parts:
1. Evolution of startups from entrepreneurs with an idea to an extension of corporate r&d / innovation, where startups are given money to ideate instead of to build an existing product.
2. junk bonds -> dot com -> cdos -> vc funds. Money chasing high yields
There would be a lot of cool ways to improve the model by giving feedback, either showing training images where the model is uncertain, or some more advanced explanations for classifications flagged as incorrect, in order to guide the user to gather the training data that can improve it.
And possibly providing a summary of where it knows it works well.
There are a lot of benefits there, both for improving models people are building but also to help users understand why their model is performing as it does.
I expect, as they allude to, it will begin with simple classification tasks in order to stick with the clean user experience they've built. But I'm super eager to see what they propose in this area.
What I came here to say is that when I have a csv file I'm looking at with awk, thos first thing I do is a `wc -l`, then get NF (number of fields) for the first row of the csv (here e.g. 80), followed by `awk -F, 'NF==80'` | wc -l`. If the numbers don't match, then I know it's not parsing properly.
The most common issue of course is commas in quotes strings. I have a small script that removes these, so I can still use awk easily. Anything more complex like newlines in quotes strings and maybe it's time to question if awk is still worth it.
I would love to hear the advantages of combining aggregation and plotting the way Vega does.
Coding is cool because learning a small amount can actually unlock a lot, and you dont need to be a computer scientist to write an app or even build a business around software, but "learning to code" thinking it gives you equal standing with software professionals would be like thinking you could provide input structural engineering discussions because you took intro to statics.
I've tried to reason about this, and one hypothesis is that all matter has compulsions, like to get to lower gravitational potential energy, to react this way or that, etc. And that desire can be thought of as conciousness. So for more complex assemblies of matter, conciousness becomes the sum of those compulsions, but still reflects the sum of what the base components want to do - oxidize stuff, etc. But the point is, conciousness is the manifestation of the inherent desires we have, based on the rules of the universe.
Just a hypothesis of course, not testable, but my demonstration is that if you pour water into a maze with an outlet, it will solve it, based on its desire to reach a lower potential energy, and to make hydrogen bonds.
Also another good example is cooks, who get hired based on working a shift unpaid to see how they fit with the team.
Asking for a hands on demo is pretty common.
Yes, sure. Religion is super personal, it doesn't automatically mean you believe every word of a text. There may be some that demand that, but it's a broad spectrum that definitely admits looser interpretations.
Training a language model, I would argue, is equivalent (in terms of being derivative) to generating a list of the top ten words in a corpus of text. Not really a derived work.
Two observations: 1. A lot of "management" is readily automated, look at uber, it automates the dispatcher and keeps the manual workers. This idea could be extended further up the management ladder, its just that middle management is more of a stakeholder to the consultants pushing automation so they focus on eliminating workers.
2. The real scary thought to me is not manual jobs being automated, but rather humans only being used for their spatial reasoning and grasping abilities and having all other agency automated away. Anecdotally this may already be the case e.g. in some warehouses.
I think this was a great example of how people missed the point of how blockchain added value to bitcoin, got excited about the wrong things, and caused the boom/bust we saw: bitcoin replaced anti-counterfeit measures with the blockchain itself, and centralized transaction processing with distributed miners. It didnt replace a database, it used a decentralized database for a new purpose.
Other new ways of using blockchain have clearly not emerged, but I feel like the rollercoaster of popularity came from people completely misunderstanding how it added value to cryptocurrency and looking for the wrong kind of applications. Distributed tasks (bitcoin is uber for transaction verification) are where it's at, not a database.
If you can reliably find this kind of work, then go for it. But what I've seen as a consultant is that everyone likes to do that part, but what clients usually want to contract for, and what makes money is the steady, often boring, implementation part.
Targeting the MVP work you are often competing with others who will heavily discount the initial work hoping to get the lucrative implementation. What makes it harder is the uncertainty involved, there is a high bar to being trusted enough contract for something that had not been defined yet.
In the management consulting world which I am more familiar with, the distinction is between "strategy" work - the focus of well reputed (in some circles) firms with very strong networks like McKinsey, and implementation or "transformation" type work from IBM or a Big 4. The dream is to be McKinsey and come in and demonstrate a brilliant plan that others are stuck implementing, but the reality is that that kind of work is tough to sell.