I feel that this is a data collection activity (and thus, more advanced future models and usecases) disguised as a social media. People will provide feedback in the form of clicks/views on AI generated content (better version of RLHF) on unverified/subjective domains.
Biggest problem OpenAI has is not having an immense data backbone like Meta/Google/MSFT has. I think this is step in that direction -- create a data moat which in turn will help them make better models.
Hey, cool idea. Would you be able to tell me the tech stack for the whole app? I want to build a similar application for some other use case. I have built a static map with all my labels using leaflet in Python. To turn it into something like you have, what technologies will I need?
I want to learn more about how to rebalance my portfolio. I started with ETFs and MFs and then bought some good stocks when they were low. But I have never rebalanced it. Would you be able to share some resources about it? Also, if possible, some pointers about your script.
Rebalancing is just selling the high performers and buying the low performers. In his example, you'd keep your "safe asset" allocation at say 15% - if your other stocks did well one year, you'd sell some and buy more "safe assets" so they again constitute 15% of your total value. If stocks tanked, you'd instead sell some "safe assets" and buy more stocks, again until your "safe assets" are back at 15% of total value.
Mediocre performance is better than your top performers dropping 30% or 60%.
I can point couple companies that suddenly dropped from $90 a share to below $10 and then they never got up “Just eat takeaway.com” between 2018 and 2022 it was looking like they would go to the moon. In 2022 you can see hell of a drop and it is not going back.
If you would sell parts of it before 2022 you would lock at least some of the gains.
But I think you know better when to switch companies ;)
I noticed that the two bars were breaking differently under the hydraulic press. One was crumbling and the other (manufactured) was exploding. There was no mention of this effect in the video. It couldn't be the due to force because in the 2nd half the manufactured bar broke at a lower force. Could this factor has consequences on how manufactured sand concrete behaves with natural phenomenon (hurricanes, earthquakes, fires, etc.)
Exploding means it was keeping its integrity for longer (i.e. not compressing), and then releasing it when it couldn't anywhere.
Crumbling means it was falling apart.
A paper book will explode in a press because it does not have any way to compress and release any of the force on it, until it releases all of it in one shot.
Putting stress on the concrete requires force, and causes the concrete to deflect. Force over displacement is work, and energy can't be lost to nothingness by just breaking the concrete. Thus, the concrete releases the stored up energy as kinetic energy of its fragments.
Stronger concrete requires more stress to cause it to fail, and as such it takes more force to break it. There is logically more energy because of the higher force, so more energy gets released.
I could never put it words like in the article, but I always thought this was true (not just to books, but to all the inputs). Many times in my life, I have remembered obscure stuff from a random movie or specific parts of a conversation that I had many years back. Similarly with books, I can't recall it completely, but the sense of it is something that easily pops up.
Now I have started taking advantage of this using Anki. Whenever I read something having a particularly interesting or thought-provoking idea, I find a way to create flashcard(s) out of it. Now I am able to recall these things more often and many times they have guided me in my life or helping out friends.
Domain: Recommendation Systems, NLP (Translation, Language Modeling),
Classification Algorithms, Regression Algorithms, Deep Learning (Seq2Seq, Transformer, RNN, LSTM, GRU, CNN), Data Scraping, Visualization (matplotlib, ggplot2).
I am Senior Data Scientist with ~8 years of experience in Data Science. For the last three years, I have been working on Recommendation Engines for a B2C company. I have deployed 30+ models in production, receiving more than 400k QPS traffic. I have seen e2e journey of multiple data science projects: business problem -> formulation -> offline training/eval -> deployment -> A/B testing -> Scaling up -> Monitoring. In my new role, I am looking to apply my experience to design e2e ML systems.
Great work. Nice blend of aesthetics and simplicity of building the timeline.
Along with embed link, I was wondering if there is a way to download the HTML of the rendered timeline directly. I use static pages for my blog and would love a way to add the rendered output directly.
I write about random experiments I do in my life (data science, personal analytics, reviews, travel, etc). Been travelling a lot this year, but haven't been able to write about it much. Hoping to change that in the next few weeks.
Biggest problem OpenAI has is not having an immense data backbone like Meta/Google/MSFT has. I think this is step in that direction -- create a data moat which in turn will help them make better models.