boto3

yes, human average reading speed is ~250 words per minute, but to be fair, it's not a widely known stat.

I am thinking some efficient RAG experience. E.g. you can ask "Tell me about the Donald Trump trial."

could not resist:

https://chat.openai.com/share/6e8bd7ec-32f8-4ed8-a44f-c08a3c...

gotta say, ChatGPT really knows how to make up a story, and I'd not be surprised if Sutskever and Brockman will indeed be co-CEOs :)

Oh, you meant interaction as a joint training with images, actions, feedback etc. That would be the next generation I guess.

I am simply thinking of interaction here as similar to learning a language in a classroom. First the teacher provides sample questions/answers, then the teacher asks the students to come up with answers themselves, and tell them which one is better. The end result here is I think ChatGPT is quite good at answering questions and can pass as a human, especially if it's augmented with a fact database, so obviously wrong answers can be pruned.

It did, actually. The model was trained with multiple rounds of reinforcement learning where human judges provided the feedback: first with full answers, and then with ranking of answers as most relevant.

So the model in production is probably frozen, but before that it went through multiple rounds of interaction with the world.

Zuck probably just massaged him: Yann we should be cool as ChatGPT.

No one wants war, including Russian citizens:

https://twitter.com/ragipsoylu/status/1496905869983768581

Fuck Putin. You'll rot in hell.

Seems like the metaverse is not selling well to the Wall street in the real universe :)

What prevents the author from selling another token with the same content? How is the content unique to the buyer?

Leverage. Call options at 2000 strike expiring on 9/16/22 cost roughly 1000 now. For each dollar GOOG goes up (or down), the deep in money options' price will probably move roughly one dollar in the same direction, as long as GOOG price doesn't go down too much. So instead of investing on 100 Google shares for 300K dollars, she can buy 3 call options (each call option is for 100 shares) for the same money and makes three times as much if GOOG goes up, and loses three times as much (or more) if it goes down: e.g., GOOG is under 2000 in 9/16, she will lose all 200K dollars.

Essentially, she's bullish on GOOG. She's also rich enough to take the risk to leverage 3 times. She's even more bullish on RBLX though.

The referred ProPublica article has a lot more details:

https://www.propublica.org/article/the-secret-irs-files-trov...

To me the most interesting bit in the article is they managed to scrape Google results at scale.

This is sad. If the report is correct, Robinhood might be negligent here. I don't know if there's security law required, but Schwab has 4 levels of option tradings [0], and you'd need to apply and get approved. I've investing/trading for more than 10 years now, but they've only given level 0, and rejected my level 1 application. It seems this guy's trade is at level 2.

[0] https://help.streetsmart.schwab.com/pro/4.36/Content/Option_...

Everything is factored in the real estate price: the weather, the school district, the night life, the ethnic grocery stores, crime rate etc. and last but not least, the opportunity to change to a new job with much higher comp.

Real estate, while not as liquid as stock market, is quite efficient in my opinion.

It's not that hard to DIY. You can start with: https://facebook.github.io/planout/

> startups have tremendously high variance. Most startups are st. There's a few of them that are really exceptional, and if you land at an exceptional one, you can do really well. If you randomly pick something, you'll probably have a bad time.

Great words of wisdom. How does one find a great one though? Even for YC with their expertise and experience, most of the startups they fund are not great.

That just goes to show that Pearl's approach on causal reasoning is solid. Instead of looking patterns and correlation in data, one could consider building causal graphs from the domain knowledge.

Discussed earlier: https://news.ycombinator.com/item?id=4166183

So 600M is the upper bound. Drew Houston owns roughly 25% of the company now, and he probably owned ~75% at when YC invested, for a dilution rate of 3. I'd say YC investment probably has dilution rate of 10, for the real amount of 60M. Not bad, but not Whatsapp scale either.

ok, I'll be the one to ask. How much is this check worth now?

The gap between the levels of abstraction that humans and machines operate is much bigger than the most AI researchers think. No amount of computation for various kinds of gradients can compensate for that. The next AI breakthrough will be a radical development in knowledge modeling and representation.

I'd say the choice really depends on the what you are trying to do. Python, Java (and even Golang) all have their pros and cons. If you are doing personal projects that involve with data processing, and maybe a simple webapp, nothing beats Python and its ecosystem. If you are planning to grow a team, and the projects are enterprise-ish, then Java is a good contender. In my case, while I wrote some 30K lines of Golang in my last job a few years back (and still got compliments from the current maintainers to this day), I just tolerated the language and never enjoyed it. Before that, I did a lot of Java and before that C and Perl. I don't program daily anymore, and only play with data science / ML ideas these days, and so I do Python (mostly in Jupyter notebooks) and run the scripts on real datasets in remote servers, with a simple Flask app to display the results/charts.

How do you handle aggregated metrics, e.g. request count? What does the instance (either server/container) expose at localhost:9001/metrics?

Sorry for the harsh words, but this looks like a solution looking for a problem. Any CEO of a company that's lucky enough to have customers contacting them would spend a not trivial amount of their time reading and replying to the contacts. For example, Jan Koum, Whatsapp founder said that they only started hiring dedicated customer service staff when they reach 150M users [0].

[0] https://youtu.be/8-pJa11YvCs?t=849

I think the demographics that just use the default search engine are probably most desirable for the advertising publisher as they are more likely to click on ads.

There's a reason for that. NB needs to fit more params than LR (and other linear discriminant learners), e.g. for binary classification, NB needs to fit 2*N params while LR just N.

Google is making 50 USD per user, potentially an order of magnitude more from a US user from ads, so I am quite certain that your offer of $25 USD is a low baller :)

What kind of code search are you thinking of? In my experience, code search could be useful when one works with a big and unfamiliar code base, but even then good architecture documentation and a good IDE would help more. And when one really needs string search, `git grep` is usually fast enough (for me on a 5GB code base).

I used to think it's just economic, meaning the poorer area/country would have higher crime rate. But now I've travelled quite extensively around the world, I think crime is mostly related to two things: inequality and social trust. They are also strongly related as inequality leads to distrust in society/government and in tern that distrust leads to people trying to accumulate personal wealth/power at all cost, even at others' expense.

"Uber has created more jobs faster than any company in history"? Walmart currently employs more than 1M people right now, in the US.