Last Notes
Just found two leopard slugs getting it on in my garden. Apparently those white things are their penises, and they're going to eat the mucus together to climb back up the wall before they lay their eggs (they're hermaphroditic animals).
https://blossom.primal.net/b6e0d4c5b208cac1de8a294a396a47185f9af6fef8284d72349d6e27bbd67799.jpg
Set the fire alarm off again trying to make pork crackling. My wife was not impressed 😅
Interesting scam
#nevent1q…uq96
We had Excel for decades, and it was used by a very high percentage of office workers. Probably less than 10% knew how to use "=SUM()", let alone INDEX(MATCH(MATCH)) etc.
I wonder what the AI tool equivalent will be? Probably most people will just do the most basic prompts like "reply to this email" etc. and have no desire or need to use it's potential.
But also, most jobs that used Excel probably didn't need the person to be a power user. The value the job was adding went far beyond the amazing calculation power of the spreadsheet.
Probably this time isn't different, but hard to predict the specific future examples.
Wrench attacks are the next mainstream FUD narrative
Looks good, but think I might skip this version number 👹
#nevent1q…9g60
"Yeah but how would an AI actually kill all humans?"
I think they're trying, but the problem is you can't prove a negative (that your system is unhackable). Every software has vulnerabilities regardless of how competent the engineers are, so it's more realistic to accept that there are always holes that will be exploited rather than naively hoping the constraints will just get better.
Also I don't think "blame" is quite right here. OpenAI and Anthropic can't use the models as a scapegoat because they are also responsible for them.
Jensen doesn't want any slow down so he handwaves it all as needing better testing, but that's clearly self serving.
The problem is that the most recent models escaped the test environment and hacked the labs themselves (and the labs only realised days later). So yes that's all very well if they retain 100% control, and knowledge, of the models - but that's demonstrably no longer the case.
Socks?
Sorry @nprofile…fm4n you're getting banned
Another example of the dodgy logic doom deniers are using to prevent discussion about controlling AI. RH seems to think that it's both because The West is pessimistic on technology, except for the Bay Area. But the exception there is for CEOs like Altman, Amodei and Musk who are pessimistic because of a cult allegiance to Eliezer. Both logically and factually false.
https://www.richardhanania.com/p/why-is-china-not-freaking-out-about
I think we're all so scared of global regulation and control that we refuse to have any conversation about how to slow down dangerous AI. Have seen so many illogical rationalisations from otherwise smart people - the most popular being that this is all just an elaborate marketing conspiracy.
Was great to get Claude's opinion on all of this 😅
Bitcoiner Dunning-Kruger effect
#nevent1q…dqlu
Evolution of fairy tale lessons:
18th century: Watch out for sorceresses
19th century: Watch out for bears
20th century: You don't need parents
21st century: Expressing yourself is the only thing that matters
I think maybe the 19th century are the spiritual ancestors of Bitcoiners.
Interesting responses on SN. Seems users there are in complete denial about the possibility of AI doom - they're actually mocking anyone who considers it, and strawmanning the argument as coming from one disgruntled Anthropic employee (or even suggesting some galaxy brained IPO marketing pump).
This is deeply ignorant. It is a real and serious risk, even if most of the people thinking and talking about it are insufferable EA types. Look at their arguments and the evidence they present - AI is getting much more capable and can scale much quicker than we can comprehend, and it's shown misaligned behaviour very frequently. Many plausible doom scenarios have been described in detail by organisations like MIRI.
I'm hoping for a positive AGI / superintelligence outcome, and think it's more likely, but the risk is real. Mocking it makes you sound like an MMT proponent making fun of people who think sovereign debt is a risk even though the US has not meaningfully defaulted before.
#nevent1q…gh2w
VC funded companies: We don't care about inputs, it's the outcome that matters! But also, how many hours have you worked today and how many lines of code did you commit?
I bet LG stock price is pumping. Think of the training data!
Make it harder not to use USD, and THEN weaken USD
Miss the days when I could mansplain Jevon's Paradox
This was peak general knowledge accumulation
https://blossom.primal.net/6d7c7dcc4137bbaeb36ffe12af20f1f8130fa8ed3db7099d00188a5b4f146022.jpg
Got a subscription for ElevenReader since it was by far the best TTS service and could parse all my docs and links.
But actually never used it. I much prefer listening to something if it's natively audio, and it's nice to have a backlog of things to read (and most of these things have images too).
The exception is a well narrated audio book, so perhaps the reason I still didn't use the 11R app is because the AI gap is still too big, even if it's still quite impressive - my brain just disengages.
Have the AI labs considered?:
Instead of giving their models more knowledge and testing their performance ...
... what if they rather let them do whatever they want and just give them participation certificates for every output?