Vinamr

[nisarga @ni5arga]

be sarvam, fork @OpenAI's apache-2.0 licensed open source code, strip the attribution, call it sovereign ai. incredible engineering.

these are our frontier labs man.

[Vinamr @vinamrsachdeva]

@TomM251385 @NateSilver538 I usually have more than one task as well, but I find it extremely hard to pay attention while switching context quickly, so I reserve switching for when I can't proceed with the current task for a predictably long chunk of time.

[Robin Hanson @robinhanson]

Academics have most of their influence on society via channels where they are not held to the research standards by which they were selected: teaching, media, and government. Are we to hope research instills habits so strong that they don't deviate from them in other channels?

[Lane || Boot.dev @wagslane]

Whenever you hear someone say some hack is a "10x multiplier", just translate it in your head to "1.2x"

It's gonna be closer to 1.2 than it will be to 1.5, and its basically never 10x

[trade econ liker @twst12612648]

techbros discovering the difference between comparative advantage vs absolute advantage in real time. just because with AI they can build all the tools internally doesn't mean the opportunity cost was worth it

[Zach Roseman @zachrose51]

@PalmerLuckey the data actually shows NYC is safer (!) than London when you look at all serious crime, not just homicides (a tiny portion of crime):

London has 1,725 serious crimes per 100K population vs NYC at 1,149.

Homicide rate gets all the focus because it's almost 4x higher in NYC, but those are a tiny minority of serious crimes - just 0.35% of serious crime!

If you walk through London, you are more likely to be the victim of serious crime like the following than in NYC:
- Assault
- Rape & related SA
- Robbery
- Homicide
- Etc.

The way NYPD & London Met report data is inconsistent, so I worked with Codex to try normalize - surely people can come up with different conclusions, but by no measure is London 3x safer than New York or even close to that.

If you want to check / analyze yourself, here are the assumptions in .md file and CSV, with sourcing: https://t.co/cgj1gqj4Kr

[nisarga @ni5arga]

you gotta be kidding me, the indian government asked to block GitHub URLs in an official order

[Paul Graham @paulg]

One way you can recognize AI slop is when the ideas and the diction don't match — when the ideas are completely ordinary, but the diction is that of someone announcing a brilliant discovery that they're really excited about.

[Bryan Caplan @bryan_caplan]

I oppose guardrails for awesome new technologies.

Instead, I favor guardrails for the people who build guardrails for awesome new technologies.

They're the ones with a terrible track record of derailing progress.

[Crémieux @cremieuxrecueil]

AI slop is definitely a problem, but human slop annoys me more.

Some of the worst human slop IMO is beautiful prose written by people who can't quantitatively evaluate anything they're talking about.

It performs *incredibly well* on Substack, and it shouldn't.

[nisarga @ni5arga]

Indian courts are trying to make KYC mandatory for domain registrations and remove WHOIS privacy. this is extremely alarming for privacy, security researchers, journalists and anyone who runs a website.

Read more here: https://t.co/PaB2Gk4p3g

[Vinamr @vinamrsachdeva]

@kingofthecoastt @Noahpinion I found a paper that addresses this, thanks to @RebelEconProf's latest piece! I haven't read it yet, but in Section II.4 of https://t.co/F8giYIRLSs (pp. 13-16), Earl Thompson lays out an argument for why wartime expropriation + peacetime subsidies might be better.

[Vinamr @vinamrsachdeva]

@kingofthecoastt @Noahpinion Bc it's better to let pvt firms (rather than the government) forecast future consumption of critical wartime goods and take on the risk? What else? Maybe this is a bit unexplored but in case it isn't, any paper/s you could recommend for understanding other factors at play here?

[Justine Moore @venturetwins]

Scrolled on my LinkedIn feed for the first time in a while.

It's almost entirely LLM-generated, but I actually have no problem with it because it's a substantial improvement over what people used to post there.

[Vinamr @vinamrsachdeva]

@asymptosis @Noahpinion @grok Oh sorry, not sure about the recent bump

[Vinamr @vinamrsachdeva]

@asymptosis @Noahpinion @grok I think he's tricking the bots into identifying themselves

[spor @sporadica]

not dunking on Dario here bc Mythos is to GPT2 what a lion is to a bumblebee,

but it’s insane that GPT-2, and successive models, have had ~0% impact on causing bad universal outcomes & so SHOULD HAVE caused AI safety folks to update their priors YET none of them seem to have??

[Lane || Boot.dev @wagslane]

The next round of AI models will only be available to developers with below 130 IQ. That will average out the cumulative intelligence to levels deemed safe by the federal government.

[Valentin Ignatev @valigo]

Andrej Karpathy lost access to Fable due to regulations, so they made him a slack bot salesguy 😭

[Aditi Agrawal @Aditi_muses]

This is the key consequence of today’s judgement, as I reported yesterday. It legitimises the govt’s power to issue feature changes.

[Karthik Tadepalli @karthiktadepall]

After a long break, I'm back to devposting! I argue that the development econ focus on "institutions" was a mistake – not (just) because the evidence was bad, but because the concept of "good institutions" is tautological and uninformative for policy.