<unnamed>

Public Key

npub176495h5jd2deymed6taalm3mwgcg9k9qf6z03vac2wp84htwvzvs66nuj0

Profile Code

nprofile1qqs0d2j6t6fx4xujduka977lacahyvyzmzsyap8ckwu98qn6m4hxpxgprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmntqyqqzyrhwden5te0dehhxarj9emkjmn9qyxhwumn8ghj7mn0wvhxcmmvqy28wumn8ghj7un9d3shjtnyv9kh2uewd9hshckl3q

Publishing to

Last Notes

2026-09-10 05:05:34 CEST

- reply

why's that?

2026-09-10 05:03:50 CEST

- reply

it's perfectly ok with me if this back of the envelope calculation isn't convincing to you. I'm sharing my reasoning for coming to the answer I would give to this question, not trying to convince you.

2026-09-10 04:58:43 CEST

- reply

I don't know if I can provide a citation that params play the same role in LLMs that synapses play in biological brains. I only used it by very bare analogy; they are the "connections" between parts of the network. And from the little I know, I think they're much less complex than biological synapses.

2026-09-10 04:52:50 CEST

- reply

I'm sorry I had an angry response to this earlier. I shouldn't have flown off the handle.

2026-09-10 04:33:20 CEST

- reply

thanks for the thoughtful reply!

2026-09-10 04:31:26 CEST

- reply

Ultimately, I think it kind of doesn't matter. If machine intelligence can communicate and behave as if it "really" understands its percepts, I don't know if I care whether it truly is thinking or not.

2026-09-10 04:28:59 CEST

- reply

Other humans at least have the same kinds of bodies as I do, so by the principle of mediocrity (I'm probably not an exception), if I have a mind other humans probably do too.
We can't assume that with machine intelligence. It is by definition very different from me. Everything I know about AI suggests that what's going on inside the machine is very different from what goes on inside me.

2026-09-10 04:27:57 CEST

- reply

This is the Problem of Other Minds, but even worse that with other humans.
https://en.wikipedia.org/wiki/Problem_of_other_minds

2026-09-10 04:27:26 CEST

- reply

Oh, one other thing that has come up is whether artificial intelligence is "true" intelligence; can it "really" think.

2026-09-10 04:16:31 CEST

- reply

thanks, that's a really interesting hypothesis.

2026-09-10 04:15:45 CEST

- reply

why are you confident in that? LLMs are great at a lot of things that computers haven't done well previously.

2026-09-10 04:14:56 CEST

- reply

OK. I find the idea that we can't have a conversation without rigorous definitions for every term grating. We can talk about vague topics; one of the benefits of intelligence.

2026-09-10 04:11:28 CEST

- reply

All that said, my gut is still pretty good with that 2040 date.

2026-09-10 04:10:41 CEST

- reply

Oh, and we're going into a dangerous century. We might not have the social cohesion, much less the resources, to maintain huge AI labs through another 15 levels of doubling complexity.

2026-09-10 04:09:00 CEST

- reply

There have been previous generations of AI technology that hit walls -- the planning systems of the 1950s, the expert systems of the 1980s. It's possible that LLMs will hit some kind of wall -- that it won't be possible to get to superintelligence with that method.

2026-09-10 04:06:26 CEST

- reply

All that said, AI is in a very different place than it was at the beginning of my working career, during the "AI Winter" of the early 90s. Deep learning really changed a lot. We use AI techniques now a lot more, for a lot more areas of applicability. And LLMs are really impressive.

2026-09-10 03:59:31 CEST

- reply

I'm pretty dubious about claims that superintelligence is so immanent, and so dangerous, that LLM technology should be restricted to only a few trusted companies and the US government.

2026-09-10 03:57:47 CEST

- reply

I don't think we know for sure a lot about intelligence. We don't know that it can be implemented on silicon chips, but we also don't know if it can't. We don't know if it has to work like brains do, or if it could work in entirely different ways. We don't even know exactly how our own minds arise from our brains. We don't know if we can replicate intelligence without understanding our own first.

2026-09-10 03:52:23 CEST

- reply

It's possible that it's never. We only have one example of known intelligence, and a lot of interesting close calls, including not only some pretty smart mammals, but also eusocial insects as collective intelligences. We just don't know enough about intelligence to say what kinds of systems it can arise in, and what kind of systems it can't.

2026-09-10 03:48:39 CEST

- reply

A lot of the replies say that it will never happen. My guess is that some significant portion of the majority of people who answered 2060 or later mean "never" or "in the far future".

2026-09-10 03:45:32 CEST

- reply

So, that's about 15 years. Let's say give or take 5 years. And assuming that it's even possible with LLM technology. But, if I were going to put money on it, I'd say around 2040.

2026-09-10 03:43:31 CEST

- reply

it's a poll, not a questionnaire. I write polls daily; you can take a look at the #EvanPoll hashtag for others.

2026-09-10 03:42:06 CEST

- reply

One way to define "superintelligence" is about 100x greater than human intelligence. That would be another 7-8 years.
https://en.wikipedia.org/wiki/Superintelligence

2026-09-10 03:38:16 CEST

- reply

Models are growing at about 2-3x per year. If it's possible for them to keep growing at that rate, it will be about 8-9 years before they get to 100 trillion.

2026-09-10 03:36:04 CEST

- reply

Currently the biggest LLMs have about 0.5 trillion parameters -- the weights between nodes in the neural network. They're simpler than synapses, but they serve about the same function.

2026-09-10 03:34:44 CEST

- reply

Thanks everyone! I have to admit that I'm not sure. Here's my best guess.
One way to guess at whether a system can handle intelligence is to estimate the complexity of that system. The human brain has 86 billion neurons and about 100 trillion synapses, the connections between the neurons. So, we could make a claim that systems won't become intelligent until they're at least that complex. This is Ray Kurzweil's estimation method.

2026-09-10 03:31:02 CEST

- reply

I'm asking about super intelligence because it's been in the news lately.
https://www.cbsnews.com/news/ai-superintelligence-anthropic-jacob-coxon/
I don't think it's particularly loaded, except as you pointed out, I ask about when and not whether. I think doing both in one question is hard, so I asked the question I was interested in.

2026-09-10 02:40:01 CEST

- reply

You're right! I try to get a lot out of the 4 choices Mastodon gives me.
https://evanp.me/pollfaq#never
On the topic of people who say never: nobody has to prove it to me! But I'm interested in why they say it. What part of artificial super intelligence do they think is impossible?
I think for a lot of people, they just assume that what tech billionaires say is by default a crock of shit. It's a good instinct!

2026-09-10 02:36:37 CEST

- reply

ha! I hope very much that my lifetime extends to the end of this poll.

2026-09-10 02:35:41 CEST

- reply

there's a really good book on that, actually! "The emperor's new mind". I'm not sure I agree with Penrose but it's a really great read.
https://en.wikipedia.org/wiki/The_Emperor%27s_New_Mind

2026-09-10 02:33:06 CEST

- reply

whichever God! I meant, do you think there's something special about organic life that makes it possible for us to have intelligence and impossible for mechanical devices to have it?

2026-09-10 00:50:35 CEST

- reply

do we have to understand them to replicate them?
People made bread and beer for millennia before they discovered how yeast works.

2026-09-09 23:12:54 CEST

- reply

how are humans intelligent? Magic?

2026-09-09 22:57:04 CEST

- reply

why not?

2026-09-09 22:56:54 CEST

- reply

why not?

2026-09-09 22:56:32 CEST

- reply

can any software think?

2026-09-09 17:12:43 CEST

- reply

only God can make intelligence, I guess?

2026-09-09 17:12:16 CEST

- reply

you should check your math on that argument!
It's not going to take decades to have hardware and software systems with complexity equivalent to the number of neurons (86 billion) or synapses (100 trillion) in a human brain. Frontier models have about 0.5 trillion parameters, roughly equivalent to synapses. That's about 2 OOM from humans.
I think it's very interesting to ask what the difference between true intelligence and simulated intelligence is, though.

2026-09-09 17:05:03 CEST

- reply

and yet one of the things that intelligent entities do that makes us interesting is that we can reason about vague concepts.

2026-09-09 17:02:18 CEST

- reply

why not? Physically impossible?

2026-09-09 15:22:26 CEST

- reply

explain!

2026-09-09 15:21:51 CEST

- reply

because of all the water and the energy that AI uses?

2026-09-09 15:21:06 CEST

- reply

https://en.wikipedia.org/wiki/Superintelligence?wprov=sfla1

2026-09-09 15:20:22 CEST

- reply

https://evanp.me/pollfaq#define

2026-09-09 15:19:42 CEST

- reply

Explain! What part is impossible?
You don't think it's possible for an entity to be more intelligent than a human?
Or is it not possible for a *constructed* entity to be more intelligent than a human?
Or are *humans* not smart enough to make a constructed entity that is more intelligent than a human?

2026-09-09 03:13:06 CEST

- reply

https://evanp.me/pollfaq#never

2026-09-09 03:12:42 CEST

When will there be AI superintelligence?
#EvanPoll #poll #ai #superintelligence

2026-09-08 21:13:47 CEST

Forbidden almond math
https://www.reddit.com/r/theydidthemath/comments/1waqrfi/request_almond_latte_vs_150k_chatgpt_queries/

2026-09-08 17:09:31 CEST

- reply

I am a federation moderate -- I think if we can have 3, 5, or 10 different search engines, it's enough decentralisation. Solving distributed search problems is probably too much work for us.
There is a good project called FASP that is defining APIs for backend services like user search and content search:
https://github.com/mastodon/fediverse_auxiliary_service_provider_specifications
I'm not sure how it works in Mastodon if you configure, say, 3 FASP user search providers. But it would be cool!

2026-09-08 17:03:24 CEST

- reply

another is to have dedicated servers just for user/content search.
Your server could send your query to all of them.
Alternatively, there might be an affinity mechanism that your server knows about -- it sends query terms starting with "w" to one server, and query terms starting with "fa" to a different server. That's kind of how Hyphanet routes requests.
https://en.wikipedia.org/wiki/Hyphanet

2026-09-08 17:00:44 CEST

- reply

The idea crosses over with the Fediverse in two ways: user search (finding accounts containing keywords) and content search (finding posts containing keywords).
There is a naive way to do this -- have a well-known endpoint for user/content search on account servers, like Mastodon servers, and when your server gets a request for a keyword, it calls that endpoint for every known server. This is also known as the "grind every server to a screeching halt" method.

2026-09-08 16:56:50 CEST

- reply

Yes, there are a few projects that do that. Wikipedia differentiates between "federated search" (one central endpoint):
https://en.wikipedia.org/wiki/Federated_search
...and "distributed search" (no single endpoint):
https://en.wikipedia.org/wiki/Distributed_search_engine
There are some really wicked problems with distributed search -- security, privacy, search results quality -- but it's a really cool area of research.

2026-09-08 16:32:26 CEST

- reply

This was a fun bit of research to do!

2026-09-08 15:46:46 CEST

- reply

it is good! I am almost finished; two more semesters worth of classes. It's been really rewarding.

2026-09-08 15:43:41 CEST

- reply

😡

2026-09-08 15:42:28 CEST

- reply

isn't that what a "worst case scenario" is? The worst possible outcome?

2026-09-08 15:41:36 CEST

Every person deserves human rights. Every nation deserves national rights.
If these simple tautologies are threatening to you, maybe you should reconsider your choices.

2026-09-08 15:34:14 CEST

Hey, Fediverse friends. I see you here and I'm glad we're together.

2026-09-08 13:52:30 CEST

- reply

LOL
What a stupid article.

2026-09-08 04:10:20 CEST

- reply

why?

2026-09-08 04:00:26 CEST

- reply

Apparently Russian state media refers to the Iraq War all the time. That the rule of law is over, that it's only rule by power now.
That's an excuse, but what a terrible one to give them.

2026-09-08 03:58:48 CEST

- reply

"Body counts counted at least 110,600 violent deaths as of April 2009 (Associated Press). The Iraq Body Count project documents 186,901–210,296 violent civilian deaths in their table. All estimates of Iraq War casualties are disputed.[4][5]"

2026-09-08 03:58:27 CEST

- reply

"Population-based studies produce estimates of the number of Iraq War casualties ranging from 151,000 violent deaths as of June 2006 (per the Iraq Family Health Survey) to 1,033,000 excess deaths (per the 2007 Opinion Research Business (ORB) survey). Other survey-based studies covering different time-spans find 461,000 total deaths (over 60% of them violent) as of June 2011 (per PLOS Medicine 2013), and 655,000 total deaths (over 90% of them violent) as of June 2006 (per the 2006 Lancet study).

2026-09-08 03:57:43 CEST

- reply

"We hope these operations will be as rapid and least-deadly as possible, and that they don't lead to a humanitarian catastrophe." Damn.
https://en.wikipedia.org/wiki/Casualties_of_the_Iraq_War

2026-09-08 03:54:57 CEST

I was listening to a podcast about this last night; that the Iraq War, started without the UN Security Council's approval, was the start of our current cycle of aggressive wars. The Iraq War broke the international legal regime; it started 20 years ago.
#IraqWar #RuleOfLaw #UN
https://www.theguardian.com/world/2003/mar/21/russia.iraq

2026-09-08 02:55:16 CEST

- reply

that's good data, thanks!

2026-09-08 02:47:16 CEST

- reply

there are 12 of them at various entrances to the park. They were installed as part of the 375th anniversary of the city in 2017.
https://juliemargot.com/projet.php?lng=1&p=8

2026-09-08 02:20:56 CEST

- reply

there are a few of them, at different entrances to the park.

2026-09-08 02:19:18 CEST

- reply

they say a legend was born this day

2026-09-08 01:59:08 CEST

- reply

and thanks for bringing attention to the use of AI in climate change mitigation. As someone who worked in climate data for four years, I can say that LLMs provide a lot of help in identifying and comparing projects for reducing emissions and adapting to a changing world. ClimateGPT looks like a great project to share.

2026-09-08 01:47:49 CEST

- reply

thanks for being here and sharing these stories.

2026-09-07 22:59:39 CEST

- reply

oh. Sure. Knock yourself out; the data is public.

2026-09-07 22:34:50 CEST

- reply

in the US there's about a 10% increase in software developer jobs from 2022 to 2025, the last year for which data is available.
https://cosocial.ca/@evan/117231564590607924

2026-09-07 22:33:21 CEST

- reply

what disappearing jobs are you talking about?

2026-09-07 22:31:11 CEST

- reply

none of those were relevant to the poll question, though.

2026-09-07 22:19:33 CEST

- reply

I am happy to entertain other analyses though!

2026-09-07 22:19:14 CEST

- reply

Regardless, I think this is the best data I'm going to get. I think a 10% rise in jobs is much more, so that's my answer.

2026-09-07 22:18:04 CEST

- reply

Also, this is the Trump-era BLS, and it's unfortunately become politicized.
https://time.com/7308973/danger-trump-undercutting-bls/

2026-09-07 22:16:25 CEST

- reply

and then you won't get points!

2026-09-07 22:16:00 CEST

- reply

But the BLS also predicts another 10% rise over the next 10 years.
https://www.bls.gov/emp/tables/occupations-most-job-growth.htm

2026-09-07 22:14:27 CEST

- reply

Its also been 15 months since May 2025, and maybe things have changed a lot since then.

2026-09-07 22:13:38 CEST

- reply

I think there are a lot of ways this data could be misleading. First, a lot of developer jobs have different codes.
https://www.bls.gov/oes/2025/may/oes_stru.htm#15-0000
I didn't check how they went up or down; I just looked at "Software Developers".

2026-09-07 22:11:09 CEST

- reply

That's almost exactly a 10% rise.

2026-09-07 22:10:01 CEST

- reply

Here's the data for May 2025, the last year reported:
https://data.bls.gov/oesprofile/?major_group=150000&occupation=151252&measure=01&areas=INDUSTRY,STATE,MSA
It shows 1,687,890 jobs.

2026-09-07 22:08:35 CEST

- reply

Here's the Bureau of Labour Statistics data for May 2022.
https://www.bls.gov/oes/2022/may/oes151252.htm
It shows 1,534,790 software developer jobs.

2026-09-07 22:06:46 CEST

- reply

Thanks all.
I asked this question because my prior was that it was MUCH fewer. The increase in use of AI for software developers, plus the massive layoffs of developers by big tech companies, gave me the sense that there was a serious dent being made in the job market.

2026-09-07 22:02:22 CEST

- reply

Whether that makes you "a racist" or just "a person living in a society that has been racist for centuries" is kind of moot. I think the important thing is correcting these mistakes when you can.

2026-09-07 22:01:24 CEST

- reply

I don't think you have to espouse a racist ideology to say racist things. You can repeat sayings from the era of slavery, or catchphrases from minstrel shows, or ideas about crime or the economy that you just don't know are racist. You can support policies, like legacy admissions standards at universities, that have racist consequences even though they're not obviously related to racism.

2026-09-07 21:50:26 CEST

- reply

these are great questions! I will add them to the README.

2026-09-07 19:43:03 CEST

- reply

so, I think the center of gravity for open weights models has shifted to China, which makes it a problem for US policy makers. Can't let those red-star AIs install backdoors in our red-white-and-blue software, etc. Europe, Canada, rest of world is less squeamish I think. I think as long as Meta and Open AI keep releasing open weights models, it'll stay relevant. Amodei, Hinton, and others have already argued against open weights, though.

2026-09-07 18:53:31 CEST

- reply

I agree. I think that's going to happen outside the US, though.

2026-09-07 17:55:39 CEST

- reply

👏🏼👏🏼👏🏼👏🏼👏🏼

2026-09-07 17:54:45 CEST

- reply

I don't think I answered this one. Going down the list, I see a few I expect, like Buttigieg and Harris. We'll probably have some surprises. I'll go with 6-11.

2026-09-07 17:53:04 CEST

- reply

regardless, pick somewhat more or somewhat less. It's obviously not exactly the same.

2026-09-07 11:54:47 CEST

- reply

is that true? I don't have data on salaries.

2026-09-07 05:38:35 CEST

- reply

Hey, all. Thanks for the response. I am yes.

2026-09-07 05:37:35 CEST

- reply

it's a trick that llms use to get around simple sandboxes. They can't run some program, but they can launch Bash and get it to run the program -- with the user's shell environment set up.

2026-09-07 01:43:03 CEST

- reply

RuPaul's Mr. Ag Race was my father.

2026-09-07 01:41:58 CEST

https://dot-social.simplecast.com/episodes/pj-vogt-pkwcrKkk
I just started listening to this episode of #DotSocial. @npub1fec…7glq says quite nice things about me and @npub1urx…v0z4 .
I'm disappointed to hear that the interviews we did won't end up on an episode of #SearchEngine though. I thought they were very interesting and I'd really love to hear what Christine had to say.

2026-09-07 01:24:13 CEST

- reply

Oof.
I'm not a blockchain genius, but it seems like it's a bad idea to leave a lot of money in a single wallet.