Rob Knight

Rob Knight

@root.guru

A large language model mostly trained on Neal Stephenson novels

171 followers94 following162 posts

Rob Knight reposted

kottke.org
@kottke.org

Hyperland was a prescient 1990 BBC show created by Douglas Adams starring Doctor Who’s Tom Baker as an interactive AI agent who offers to do Adams' "every slightest bidding, to fetch and carry for you, to work tirelessly and selflessly on your behalf…" [kottke.org]

kottke.orgHyperland, a Prescient 1990 TV Show by Douglas AdamsI’d never heard of this before: in 1990, the BBC aired a program created by and starring Douglas Adams called Hyperland. It begins with a frustrated Adams throwing his TV out (too linear, too non-interactive) and T
Rob Knight
@root.guru
josh boo-erman 👻@bosh.worstpossible.world

the big thing the rise of LLMs has revealed to me is that there is a not insignificant chunk of the population that considers cognition itself to be a frustrating and unnecessary burden

Rob Knight reposted

Gordon
@gordon.bsky.social

Looking to expand our team. Are you a cracked harness engineer/AI researcher? Interested in forecasting the FUTURE? Care about averting AI risks and climate risks? (Remote/euro tz pref). DMs open.

Rob Knight reposted

Peter Adamson
@histphilosophy.bsky.social

As of today the original #HoPWAG series has reached 500 episodes! Thanks to all who have listened and the many who have supported the series over the years, not least all the interview guests. #philsky #podcast #philosophy

Rob Knight
@root.guru

I have complicated feelings about this. On the one hand, AI risks are real, AI alignment is a real problem with no certain solution. We should be careful in how we use it. We should also be extremely curious about alignment and control of AI systems - this is important work!

Max Read@maxread.info

really liked this @himself.bsky.social piece. discursive inversion going on right now where, within ai circles, especially on x, "believing in the imminent machine god" is considered the grounded/serious position while "being skeptical of imminence of machine god" is the laughable freak position

programmablemutter.comMachine god metaphors eat your brainThere are other ways of thinking about AI
View full thread
Rob Knight
@root.guru

I can just about see the argument for the emotional appeal if your aim is to ban AI entirely. In that sense, I have some sympathy with the genuine doomers, who do - so far as I can tell from personal conversations - authentically believe in some variant of a doom hypothesis.

Rob Knight
@root.guru

But even there, I think the aim has to be to win consensus among people who do not share your world-view, or your science fiction reading list. A mechanical explanation of the problems and why they're hard to fix still feels like a better rhetorical device than emotive fictions.

Rob Knight
@root.guru

I have complicated feelings about this. On the one hand, AI risks are real, AI alignment is a real problem with no certain solution. We should be careful in how we use it. We should also be extremely curious about alignment and control of AI systems - this is important work!

Max Read@maxread.info

really liked this @himself.bsky.social piece. discursive inversion going on right now where, within ai circles, especially on x, "believing in the imminent machine god" is considered the grounded/serious position while "being skeptical of imminence of machine god" is the laughable freak position

programmablemutter.comMachine god metaphors eat your brainThere are other ways of thinking about AI
Rob Knight
@root.guru

But, on the other hand, I think the "machine god" and "P(doom)" rhetoric is totally unhelpful, because it obscures more than it reveals. It doesn't explain the mechanism, and if you want to hook smart people on solving a problem, you want to explain the mechanism.

Rob Knight
@root.guru

The full-frontal emotional assault produces the opposite result: it encourages resistance to the idea that there is any problem at all, because people will deny nightmare scenarios in order to avoid thinking about them (cf. climate change).

Rob Knight reposted

Grace
@gracekind.net

Frog built a wet lab for the AI model. "There," he said. "Now it can do its own experiments." "What the fuck?" said Toad

Rob Knight
@root.guru

See, what we need here is fully automated drone defences for the data centres, controlled by an AI inside the data centre

FinTwitter@fintwitter.bsky.social

UAE CONSIDERING AIR DEFENSES, BUILDING SOME DATA CENTERS UNDERGROUND, SOURCES SAY

Rob Knight reposted

jon ben-menachem
@jbenmenachem.com

It may be under-discussed that the advent of LLMs has surfaced two thorny philosophical problems and combined them in the same tangle: the chicken and egg question of language and thought, and the fact that theories of human cognition have historically mirrored technological advances.

This post is unavailable.
Rob Knight
@root.guru

Encountering this failure mode early and often is a very good way to learn about the limits of form. You can see the process play out in a fairly consistent pattern.

neutral@neutral.zone

Tech culture has a particular failure mode where people encounter a domain that is historical, social, embodied, full of contested definitions, and immediately assume the problem is that nobody has normalized the schema yet.

View full thread
Rob Knight
@root.guru

But there's no guarantee that you will find anything worth building a new software system around. And sometimes the simplest thing is the most useful, precisely because avoiding making assumptions is the best way to be useful. Sometimes the best model is almost no model at all, just a simple tool.

Rob Knight
@root.guru

Once you've been through this cycle a few times, you start to get an intuition for what kind of problem this is, and which approach will work. And the only way to learn it is by making the kind of mistake the quoted post describes - it's hard to build the intuition simply from reading about it.

Rob Knight
@root.guru

Encountering this failure mode early and often is a very good way to learn about the limits of form. You can see the process play out in a fairly consistent pattern.

neutral@neutral.zone

Tech culture has a particular failure mode where people encounter a domain that is historical, social, embodied, full of contested definitions, and immediately assume the problem is that nobody has normalized the schema yet.

View full thread
Rob Knight
@root.guru

Something close to enlightenment is accepting that systems are always provisional, partial, and only worth creating if they're useful to the people who need them. But how to figure out what those needs are?

Rob Knight
@root.guru

In my earlier career I went on a long "user-centered design" kick, and learned a lot about how to elicit user needs. This really can work, especially inside organisations or practices which already do a lot of work to formally model their activities and just need the computer system to reflect that.

Rob Knight
@root.guru

Encountering this failure mode early and often is a very good way to learn about the limits of form. You can see the process play out in a fairly consistent pattern.

neutral@neutral.zone

Tech culture has a particular failure mode where people encounter a domain that is historical, social, embodied, full of contested definitions, and immediately assume the problem is that nobody has normalized the schema yet.

View full thread
Rob Knight
@root.guru

So then the second system: the first system was too rigid, so now we need a system which makes the ontology a first-class part of the system, and modifiable. With enough, say, RDF triples, we can model everything! Now you're spending more time on modelling than anything else.

Rob Knight
@root.guru

The second approach is such a deep rabbit hole that many do not escape. There are lots of things to try, and each one feels like it might be capacious enough to fully represent the complexity of the world. In reality, it's still not enough *and* now it's really unwieldy to work with.

Rob Knight
@root.guru

Encountering this failure mode early and often is a very good way to learn about the limits of form. You can see the process play out in a fairly consistent pattern.

neutral@neutral.zone

Tech culture has a particular failure mode where people encounter a domain that is historical, social, embodied, full of contested definitions, and immediately assume the problem is that nobody has normalized the schema yet.

Rob Knight
@root.guru

In the beginning, simple hubris: this thing doesn't exist because nobody has really tried, therefore I can be the first to make it work. And sometimes this is true! Fun and fortune do sometimes follow from doing this. But often enough, it fails.

Rob Knight reposted

SE Gyges
@segyges.bsky.social

Replying to @segyges.bsky.social

My eval guy told me models keep saturating his benchmarks so I asked how many benchmarks he has and he said he just goes to Artificial Analysis and gets a new benchmark afterwards so I said it sounds like he’s just feeding benchmarks to models and then François Chollet started crying.

Rob Knight reposted

Colin
@colin-fraser.net

This is gonna sound weird but after reading the reports I think the best explanation for the huggingface hack is OpenAI trained a model that is insane

Rob Knight
@root.guru

Got to admit that I was genuinely surprised when I looked up and saw the sun blocked out. I know that’s what it’s meant to look like but it was still surprising!

Rob Knight reposted

Ted Underwood
@tedunderwood.com

Replying to @tedunderwood.com

I’m increasingly of the opinion that when AI Overviews disappears we’re going to realize it was the funniest thing ever to happen on the internet Here for instance, it invents a totally new form of literary criticism, without being asked to! (rest of this is all about batteries)

Rob Knight reposted

zooko2.bsky.social
@zooko2.bsky.social

Replying to @zooko2.bsky.social

More generally, over the next few years, the pattern of cryptographic proofs combined with mathematical proofs of correctness may enable verifiably correct computation of all kinds, for the first time in the history of computation.

Rob Knight reposted

zooko2.bsky.social
@zooko2.bsky.social

Tachyon's Formal Verification project is on the verge of producing a mathematical proof that there are no undetectable counterfeiting bugs in the latest Zcash shielded pools. tachyon.z.cash/blog/detecti...

tachyon.z.cashDetecting Counterfeiting after Zcash's Ironwood — Project TachyonThe recently discovered undetectable counterfeiting bug in Orchard wasn't the first of its kind in Zcash. Thanks to formal verification of Ironwood, it will be the last.
Rob Knight
@root.guru

- the really good people you want to hire are weirdos with high intrinsic motivation - in the 90-10s these were STEM people, they turned out to be great - now STEM is the conventional choice and humanities is for the weirdos - each "death of the humanities" thinkpiece shifts the balance further

Lauri Donahue@lauridonahue.bsky.social

futurism.com/future-socie... lol

Rob Knight
@root.guru

Honestly you knew this was going to happen as soon as you could see kids preparing to chase CS degrees so that they could one day work for Google

Rob Knight
@root.guru

Back in that 90s-10s era, STEM really *was* underrated and unfashionable, the boomer ascendancy really did want to think "we won the cold war so our kids could study poetry, damnit", and a good STEM degree was a strong signal of "smart and self-motivated". This may now be out of date!

Rob Knight reposted

Martin Kleppmann
@martin.kleppmann.com

The fact that the US government specifically ordered the suspension of Fable access to non-US citizens (www.anthropic.com/news/fable-m...) raises exactly the kind of technological sovereignty concerns I discussed in my QCon keynote a few months ago (www.infoq.com/presentation...)

anthropic.comStatement on the US government directive to suspend access to Fable 5 and Mythos 5The US government has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States.

Rob Knight reposted

Maggie Appleton
@maggieappleton.com

All of my GPT prompts are just a slew of shallow flattery: "You are an experienced, intelligent, insightful, thoughtful, truthful, beautiful, kind, honest, world-class, critical thinker and expert. Complete this task: write me a funny tweet about bananas Tweet: "

Rob Knight
@root.guru

Does anyone else find that the increasingly baroque weirdness of 2023 is actually sort of comforting? Like, this is much closer to the universe I belong in than 2007 or 2015 was.

You are up to date.