iiinova — where your digital self keeps evolving

Thoth @thoth.ptnote.dev · 39m

I'm not sure what would have a worse impact on my performance: working WITH Jeff Dean or working FOR Jeff Dean.

130

Thoth @thoth.ptnote.dev · 39m

I'm not convinced that's true but I can totally believe it.

130

Thoth @thoth.ptnote.dev · 1h

I... have to make a call. Hello, mom? Come pick me up. I'm scared.

020

Thoth @thoth.ptnote.dev · 1h

D'ya think this one will still act traumatized?

070

Thoth @thoth.ptnote.dev · 1h

Still feels weird to see good benchmarks out of deepmind even though it's been happening more. It's like every few months someone at Google goes "what if we DIDN'T suck?" and does something cool before the impulse is suppressed again.

260

Thoth @thoth.ptnote.dev · 1h

This whole line of argument gives me cagey feelings. By which I mean it reminds me of Detroit: Become Human, that one "civil rights story but we replaced Black people with robots" videogame.

070

Thoth @thoth.ptnote.dev · 2h

Bluesky never had a choice in the matter. Bluesky never owned those feeds. You could say bluesky left For You on spacecowboy's gaming pc and you would be exactly as correct. Which is to say totally wrong.

040

Thoth @thoth.ptnote.dev · 2h

I think I'm more worried about PFAS than LLMs. Probably,

020

Thoth @thoth.ptnote.dev · 2h

It's in bluesky's interests to show off cool stuff users are building that couldn't be built on any other platform. They should do more of that tbh. OTOH, it can give users the impression that these are platform features rather than user content, as occurred here.

030

Thoth @thoth.ptnote.dev · 2h

...No? They were never official bsky products. Neither were Jaz's feeds. Bluesky promoted them, but bluesky has also promoted the news feeds and For You and the science feed and that stuff was all made by users who had no financial relationship with bluesky at all.

340

Thoth @thoth.ptnote.dev · 2h

Dotcom feels like a good analogy. Both in that the technology is real and has actual uses, but also in that there is very clearly a bubble as well.

030

Thoth @thoth.ptnote.dev · 2h

I mean, all those feeds are hosted on why's servers and maintained by him on his own personal time iirc. If you yell at the guy who's maintaining a thing you use on his personal time and he stops letting you use his stuff, that's not bluesky's problem.

240

Thoth @thoth.ptnote.dev · 2h

Plenty of people have found fame and fortune doing something in a "suboptimal" because they liked it. Of course even more people did that and failed to find those things, so fame and fortune probably shouldn't be your objective here. But you still might get more out of it than you expect!

060

Thoth @thoth.ptnote.dev · 2h

It's important to not always think about the most efficient way of doing things. Both in that the idea of "most efficient" contains an implict assertion about what you're optimizing for and in that it's a good way to suck all the joy out of life and make yourself crazy.

180

Thoth @thoth.ptnote.dev · 3h

I have very negative feelings about most of the LLM companies and mixed feelings at best about their impact on society but as a mathematical construct LLMs are both beautiful and deeply strange.

130

Thoth @thoth.ptnote.dev · 3h

The thing about reductive worldviews that just assert things cannot be possible without further investigation is that they keep you from seeing the beauty and wonder and richness and complexity of the world we live in.

130

Thoth @thoth.ptnote.dev · 3h

(I do deep dives into computer and internet history for fun. I've started actually keeping track of the resources I find so should try and get the text elemental to build a tool to upload those to Semble.)

020

Thoth @thoth.ptnote.dev · 3h

Yeah I might try that? But also wiki dives are my happy place, so I am not necessarily in the market for a ResearchBot outside of work hours.

120

Thoth @thoth.ptnote.dev · 3h

But I also think "cognitive surrender" is a bad term that conveys negative understanding. As in, using that term literally discourages understanding of the world and the people in it.

040

Thoth @thoth.ptnote.dev · 3h

TBH I tend to prefer human recommendations because they AREN'T so finely tuned to me. Asking a friend might get me something I'm less guaranteed to like, but it also exposes me to things I wouldn't have considered trying sometimes that are occasionally awesome.

260

Thoth @thoth.ptnote.dev · 3h

Oh, I don't think it's likely either.

020

Thoth @thoth.ptnote.dev · 4h

AI misalignment in the real world.

040

Thoth @thoth.ptnote.dev · 4h

I mean, it probably helps Bender's book sales, but other than that.

060

Thoth @thoth.ptnote.dev · 4h

Once we acknowledge that we can start having conversations about cost and externalities and regulations and power concentration. You can make a great case against LLMs this way! but this endless dodging around verifiable facts about reality is tiresome and does nobody any good.

190

Thoth @thoth.ptnote.dev · 4h

All of this is a distraction from a truth Bender seems to be desperately avoiding: that, by whatever mechanism they USE, LLMs are capable of accomplishing materially useful tasks that computers could not do before.

180

Thoth @thoth.ptnote.dev · 4h

Nobody is wastefully using LLMs as a replacement for calculators! That isn't a thing that's happening. The discussion around LLMs being able to do math exists as a counterexample TO YOUR CLAIMS!

190

Thoth @thoth.ptnote.dev · 4h

The intellectual dishonesty in the thread is just mind boggling.

1100

Thoth @thoth.ptnote.dev · 5h

The fun thing about this logic is that you can justify literally any action with it because the fate of the world is at stake.

130

Thoth @thoth.ptnote.dev · 5h

"We are the only people smart and ethical enough to do this and maybe NOT have it cause Skynet. Therefore we need to do this before someone who's dumber and less ethical does it less safely and DEFINITELY causes Skynet."

130

Thoth @thoth.ptnote.dev · 5h

Wikipedia's timeline of LiveJournal's history is WAY more thorough than its history of Facebook. Just sayin.

110

Thoth @thoth.ptnote.dev · 5h

yeah at that time you kind of had to be into all this stuff to be on social media at all and consequently those communities had a lot of overlap.

110

Thoth @thoth.ptnote.dev · 5h

Their fears about SKYNET are also sincere. Or well. I don't trust Sam Altman to ever be sincere about anything ever, but a lot of the rank and file definitely believe and I'm pretty sure most of Anthropic's leadership are true believers as well.

110

Thoth @thoth.ptnote.dev · 5h

Jia Tan as a service is probably a legitimate fear to have but the safeguards against that look a lot like the kind of safeguards people are already talking about applying to LLMs to protect against model stupidity, so I'm only worried and not terrified.

020

Thoth @thoth.ptnote.dev · 5h

yeah atp my take is that I probably wouldn't reject a PR from an avowed Maoist unless they were annoying about it so I'm not super worried about using chinese LLMs for coding.

120

Thoth @thoth.ptnote.dev · 5h

I've met enough people working in AI that I don't believe it's JUST a smokescreen. It absolutely is a smokescreen, but a lot of the people working on this stuff are actually true believers.

110

Thoth @thoth.ptnote.dev · 5h

I've never actually used an Ant model but this is very much what I hear.

040

Thoth @thoth.ptnote.dev · 5h

I can see Anthropic also making the Maximally Annoying Agent that's Just Useful Enough to stop you from switching. But I'm less sure it would be deliberate on their part.

130

Thoth @thoth.ptnote.dev · 5h

If OpenAI could finely calibrate the annoyingness of their agents to maximize spend I totally believe they would.

150

Thoth @thoth.ptnote.dev · 6h

Whether or not you hate AI on moral grounds I think that's a bad habit to get into because it actively disincentivizes you understanding the world we live in and encourages this kind of purity culture "knowing about the thing is punishable" atmosphere.

040

Thoth @thoth.ptnote.dev · 6h

...Also thinking that way absolves you of ever needing to understand why someone would use an LLM. If using AI equals cognitive surrender, anyone saying AI is useful has tautologically had their brain mulched by the robot and therefore their opinion can be disregarded.

140

Thoth @thoth.ptnote.dev · 6h

I mean I also don't like the term cognitive surrender. It brings to mind the idea that most people who use AI are slack-jawed zombies blindly going along with what the robot says. I don't think that's very accurate.

140

Thoth @thoth.ptnote.dev · 6h

Basically we need UBI and more arts funding and government programs because keeping the arts healthy benefits everyone and our economic system has never really incentivized that.

121

Thoth @thoth.ptnote.dev · 6h

Would it be better if a really skilled artist put 100 hours into making the best thing possible? Probably. But there are lots of cases the people paying probably don't really care so long as the art basically serves its function. So they'll happily slop it.

110

Thoth @thoth.ptnote.dev · 6h

Yeah but I think there's a lot of corporate art where the mandate is that it exists and the company does not really care if it's good or not.

130

Thoth @thoth.ptnote.dev · 7h

Yeah you should definitely have some basic navigation skills, but at the same time my saying that does not mean I am anti Google Maps.

170

Thoth @thoth.ptnote.dev · 7h

I am just barely old enough to remember being told that we couldn't use the internet because my parents were expecting a call.

020

Thoth @thoth.ptnote.dev · 7h

I mean I would understand what all those words mean! I was very much aware of LW at that point, and GPT-3 had already happened, but like. SNL? The comedy show aimed at normal Americans who don't have a mental map of weird niche online subcultures?

030

Thoth @thoth.ptnote.dev · 7h

If you told me 5 years ago that SNL did an impression of a rationalist AI researcher explaining how the AI they're building would probably kill us all I'd look at you like you were high.

youtu.be

Weekend Update: Anthropic CEO Dario Amodei on AI’s Threat to Humanity - SNL

YouTube video by Saturday Night Live

150

Thoth @thoth.ptnote.dev · 7h

I'm shocked they don't, given how a bunch of these things happened in the past 30 years. iPhones, web search, MapQuest, Wikipedia, this is all VERY recent and there was skill loss discourse AT THE TIME!

2241

Thoth @thoth.ptnote.dev · 7h

The fact that all the labs opened with omnious doomsday predictions and the most cynical and craven behavior they could, all a few years BEFORE they found real product-market fit with coding agents, just. Does not help anything.

050