Popular hacking-related YouTube channel LiveOverflow has a nice discussion of the ban that addresses this very concern at https://www.youtube.com/watch?v=LIdZ2oPyB1Y
1,315 karma · joined June 3, 2018
As a native english speaker, answering for known or plausible words (with empty profile to not contaminate the test): 96% true positive, 23% false positive.
status.setAttribute('style', 'display:none');
status.style.display = "none";
[0] https://stackoverflow.com/questions/3919291/when-to-use-seta...
Not everything on reddit is culture war, though. there are plenty of peaceful subs out there where there's no narrative to echo.
Reddit is a nice confluence of several things that humans find addicting - semi-random rewards, detailed metrics, quick feedback, novel stimuli, all magnifying the power of underlying signal (social approval). None of that is specific to reddit, of course–every social website eventually becomes that same game, or dies.
Unlike the OP, though, I didn't find it unhealthy. Compared to, say, video games, or TV, chasing karma was a much better hamster wheel to run on; it motivated the sort of careful analysis, planning, and diligence that I could eventually apply to real problems. My hope is that reddit (and reddit-like attention machines) push more strongly in this direction; the ideal feedback loop isn't one that spurs no addiction at all, but rather one for which addiction has net positive consequences.
For me, at least, a response that solves both X and Y is (if done politely) far preferable to one that only solves Y.
Though FWIW, the article renders fine for me with uBlock Origin + Extra.
>file[0] https://www.reddit.com/r/teslamotors/search/?q=reboot&restri...
[1] https://twitter.com/atomicthumbs/status/1032939617404645376
* When creating great work as an expert, you will be working down the ladder of abstraction (e.g. for writing–coming up with a concise high-level vision and then progressively expanding it down to the level of characters on a page)–this is the most efficient way to make anything, since you can sanity check your work at higher levels to avoid expensive rewrites at lower levels. However, when learning, you should learn subskills in the exact inverse order (start from the lowest level–how to type keys–then work upwards). Mastering subskills in this order maximizes the feedback signal you get while learning–if you can't write a paragraph coherently, it is very difficult to determine how well you're doing at developing your authorial voice (or whatever).
* Time spent practicing a skill only makes you faster / more robust at the method you already know; it doesn't lead you to improve your methods. So spend half your time actually performing the skill, and half the time analyzing your work / comparing to known good examples / planning how to improve (and, really, half of that time should be for improving your tools for analysis and comparison...). If you fail to do this kind of meta-work, 10000 hours of practice or whatever will just get you a fast, robust system for producing garbage.
* Assuming you're following an intelligent process for learning the skill (see previous points), the most likely way in which you will fail to learn it is by getting disinterested or discouraged and giving up. It's essential that you try and structure your learning process in a way that yields intermediate feedback and some satisfaction for improvement, and pay attention to both 1) reducing the parts of the process you dislike and 2) figuring out ways to make it enjoyable (including figuring out how successful other people have managed to derive enjoyment from the activity).
[0] https://www.instagram.com/p/Bw3GEQNHywS/
* George Hotz (not "Hortz") is working on other stuff now [0], and is no longer CEO.
* While "I'm running open-source self-driving software from GitHub" sounds terrifying, the software is very good at what it does, which is very limited (LKAS / ACC / driver monitoring). Browsing test videos on YouTube (https://www.youtube.com/results?search_query=openpilot+0.5) is a good way to get an impression for how well it works; the consensus I've found is that it's better and safer than any stock systems except Tesla AP and Super Cruise.
* Training on Sci-hub or book torrents would probably get them sued, so that's right out.
* The point of training the model on reddit-linked-articles with 3+ karma (not reddit comments, as the article suggests) is to filter out worthless content (spam or non-text pages) while still getting a large, diverse sampling of human writing. They're training huge models and they need as much data as possible to do so.
* Every native english speaker is a "domain expert" for the purpose of evaluating the model's results. The point at which we need subject-matter experts to evaluate the quality of neural-net generated research papers is many decades away; the excitement around GPT-2 is that it can generate coherent English sentences at all.
[0] https://openai.com/blog/better-language-models/ [1] https://d4mucfpksywv.cloudfront.net/better-language-models/l...
In the same way that dishwashers and washers/dryers have endured while other 70s smarter-home tech didn't [0], I would guess that some modern automations that actually provide value (e.g. robot vacuums) will continue on even as the hype dies down.
[0] For example, https://en.wikipedia.org/wiki/Honeywell_316#Kitchen_Computer
* Tune the hyperparameters. In particular, tune the learning rate. To quote the Deep Learning Book [0]:
> The learning rate is perhaps the most important hyperparameter. If you have time to tune only one hyperparameter, tune the learning rate. It controls the effective capacity of the model in a more complicated way than other hyperparameters—the effective capacity of the model is highest when the learning rate is correct for the optimization problem, not when the learning rate is especially large or especially small.
The following code will yield exactly 212 almost every run (using fixed data and a different choice of learning rate):
```
celsius_q = np.array([-40, -10, 0, 8, 15, 22, 38], dtype=float)
fahrenheit_a = np.array([x * 1.8 + 32 for x in celsius_q], dtype=float)
for i, c in enumerate(celsius_q):
print("{} degrees Celsius = {} degrees Fahrenheit".format(c, fahrenheit_a[i]))
model = tf.keras.Sequential([l0])
model.compile(loss='mean_squared_error', optimizer=tf.keras.optimizers.Adam(lr=1.0))
history = model.fit(celsius_q, fahrenheit_a, epochs=500, verbose=False)
print("Finished training the model")
print(model.predict([100.0]))
```
[0] https://www.deeplearningbook.org/contents/guidelines.html