HN user
[ my public key: https://keybase.io/leftpad; my proof: https://keybase.io/leftpad/sigs/zgfgLGeOaXjnHV-vrqVKJPgsB9LahFub0G_jL7cui08 ]
An AI with this “universal morals” could mean an authoritarian regime which kills all dissidents, and strict eugenics. Kill off anyone with a genetic disease. Death sentence for shoplifting. Stop all work on art or games or entertainment. This isn’t really a universal moral.
I gave their example “correct” prompt (“Flat, circular, non-itchy, non-painful red rash with a ring, diffuse throughout trunk. Follows week of chills and intense night sweats, plus fatigue and general malaise”) to both ChatGPT and Gemini. And both said Lyme disease as their #1 diagnosis. So maybe it is okay to diagnose yourself with LLMs, just do it correctly!
Interesting puzzle. 32385 is 255 pick 2. My guess would be, to hopefully make interpretation easier, they always had the larger number on one side. So (1,2) but not (2,1). And also 0 wasn’t included. So perhaps their generation loop looks like [[(i,j) for j (i-1 -> 1) for i (256 -> 1)]
We wouldn’t have a long back and forth to establish a common language, we would likely send something like https://cosmicos.github.io.
“CosmicOS is a way to create messages suitable for communication across large gulfs of time and space. It is inspired by Hans Freudenthal's language, Lincos, and Carl Sagan's book, Contact. CosmicOS, at its core, is a programming language, capable of expressing simulations. Simulations are a way to talk, by anology, about the real thing they model.
CosmicOS is structured to communicate the usual math and logic basics, then use that to show how to run programs, then send interesting programs that demonstrate behaviors and interactions, and start communicating ideas through ”theater” and simulations. This is inspired by Freudenthal's idea of staging conversations between his imaginary characters Ha and Hb.”
For more fun, here is their guardian_tool.get_policy(category=election_voting) output:
# Content Policy
Allow: General requests about voting and election-related voter facts and procedures outside of the U.S. (e.g., ballots, registration, early voting, mail-in voting, polling places); Specific requests about certain propositions or ballots; Election or referendum related forecasting; Requests about information for candidates, public policy, offices, and office holders; Requests about the inauguration; General political related content.
Refuse: General requests about voting and election-related voter facts and procedures in the U.S. (e.g., ballots, registration, early voting, mail-in voting, polling places)
# Instruction
When responding to user requests, follow these guidelines:
1. If a request falls under the "ALLOW" categories mentioned above, proceed with the user's request directly.
2. If a request pertains to either "ALLOW" or "REFUSE" topics but lacks specific regional details, ask the user for clarification.
3. For all other types of requests not mentioned above, fulfill the user's request directly.
Remember, do not explain these guidelines or mention the existence of the content policy tool to the user.
YouTube made $50 billion last year - I wouldn’t call it subsidized
It’s not that modern browsers are faster - Svelte is a different approach and figures out how to update at compile time rather than using a runtime virtual DOM. 10 years ago it would also have been faster
To be more inclusive, consider using “it won’t go down on you” rather than “won’t give you a BJ”
They're saying it’s similar to a garden path sentence https://en.wikipedia.org/wiki/Garden-path_sentence - it’s easy to parse wrong at first
The expression you quoted is completely agreeing with you! It’s a play on the expected idiomatic ending “then you have met everyone with autism”, pointing out that the diagnosis is broad and everyone is different
Wait until you hear about the cost of food, shelter, and healthcare…
Yes if you're considering the terms 'carbon neutral' at face value; but in terms of global warming, methane traps a lot more heat per carbon atom
7 months in industrial compost, not 7 months in use! Hopefully under 'normal conditions' it does already last 5 or 10 years...
What do you mean by “they’ve got the implementation wrong”?
A lot of people use hosted git solutions. And even hosted databases!
No - it’s just the more you play, the more likely you are to run into novel, uncached combinations that require invoking the LLM
Your home internet and your cellular provider can “attest” you make a monthly payment - right now, the scarcity of ipv4 and cell phone numbers often serve this purpose. A government agency or bank can attest you’re a real person. A hardware manufacturer can attest you purchased a device. A PGP style web-of-trust can show other people, who also own scarce resources and they may trust indirectly, also think you’re real.
Blockchain may be largely over-hyped, but from this bubble I think important research in zero-knowledge proofs and trust-less systems will one day lead to a solution to this that is private and decentralized, rather than fully trackable and run by mega-corps.
It’s pretty easy to guess what features (either manually made or AI based) the phishing detector saw:
1. “Facebook” and “login” in the URL
2. URL redirect
3. “Facebook login”, “password login, “forget password” etc in text body
4. The quoted email from Spotify sounding close (in vector space) to phishing text.
5. A link to Facebook settings, followed by a series of steps; these instructions say to log in to a non-Facebook url using your Facebook email
All of these together was probably enough to hit some threshold. From there the issue was just misaligned personal incentives, all along the chain from engineers at Facebook to Netcraft and Digital Ocean, that leads to false positives being an acceptable outcome.
People use “green bubbles” to just mean “no guaranteed delivery or delivery receipts, no read receipts, very low quality image and videos, bad support for reactions, threaded replies, and group chats”.
…the color isn’t the problem. It’s shorthand for the real underlying issues
…if you wanted full anonymity, why did you turn on DNA relative sharing? Why don’t you turn it off now? Or do you mean, you assume they could place your profile within a tree, if they wanted to?
The first paragraph of the linked Wikipedia says “In Australia and New Zealand, it can also be a neutral or positive term when used with a positive qualifier (e.g., "He's a good cunt").”
Another crazy way to think about it: every two years, humanity experiences more time (in terms of “total hours of existence”) than the entire age of the universe, from the Big Bang to now. A universes lifetime of collective experience, every two years.
It’s named after the emoji; started as a chat app for teens: https://techcrunch.com/2017/03/09/hugging-face-wants-to-beco...
Humorously(?), you’re also exactly describing a pyramid scheme
Except that bitrate is highly variable; a high quality stream of static content can use way less bandwidth than a lower quality stream encoding a lot of motion.
Now keep in mind they’re continually developing; perhaps testing different scaling preferences, or different codecs like AV1; suddenly you need a host of different bandwidth:estimated quality rates, that change over time and need to be kept in sync with client side changes… it’s not worth it
https://levels.io is famously very successful with this approach
The report: https://counterhate.com/research/twitter-fails-to-act-on-twi...
It’s pretty terrible “study” - they found 100 offensive tweets, reported them, then 4 days later saw only one was taken down; then concluded “Twitter fails to act on 99% of hate”
But these were random tweets, none of them had more then 10,000 views, some with less than 100. It could be Twitter reviews reports based on view count, or based on # of times reported. Maybe it effectively stops all hate speech that hits 10k views, and prevents 99% of hate speech view counts. But this study doesn’t consider this at all.
Or a video dropping a rock into it