Sol Irvine

mostly harmless • building and

[Sol Irvine @solirvine]

東本願寺

[Sol Irvine @solirvine]

@scottastevenson @tszzl I just finished hand-writing a piece that uses exactly the shape (bold lead sentences), and then I saw this. At some point, all good writing will be assumed to be fake, right?

[Sol Irvine @solirvine]

@ElliotVaucher Good analogy. Contractors also own the pit crew, and the buck stops with the team principal.

[Sol Irvine @solirvine]

It's really interesting to see how Japanese builders do things differently than in the U.S. This crew is top notch, too.

The before/after on this room in particular is going to be nuts.

[Sol Irvine @solirvine]

@MichaelQiu14 @rcorpatauxTS Sounds like "reliable correlation", not the same as the reality Hoffman is talking about. I don't think you're wrong, though. Hoffman's blanket denial bakes in assumptions about whether local correlation is reliable enough to be "as good as real". I lean more pragmatic, too.

[Sol Irvine @solirvine]

People should take Donald Hoffman *very* seriously. This is not a casual person, and these are not idle speculations.

[American Alchemy @AmericanALCHMY]

MIT scientist says aliens are likely all around us! We just can't see them...

Donald Hoffman says the mathematics indicates embodied consciousness like ours is a rare exception in the universe.

The universe should be teeming with intelligence more advanced than humans, operating without traditional bodies.

Hoffman calls our human form one of the cheapest, most restrictive "virtual reality headsets" in existence.

The default state of consciousness, according to the math, has no body at all.

"Embodiment is very, very small."

"The universe would be teeming with life that is more advanced than humans and also doesn't have traditional bodies."

"Our particular kind of view of the world, the kind of, I'll call it a headset, the virtual reality headset that we're using for our embodiment, is one of the more trivial kinds of headsets."

"Our headset is one of the cheapest, most restrictive possible."

"There are all sorts of variations that are far more interesting and complicated than what we've got."

"Embodiment is the exception, not the rule."

Full episode with Donald Hoffman out now.

[Sol Irvine @solirvine]

@sethtjf Hard to deny that AI marked a significant shift in the willingness of otherwise-nontechnical people to venture into the wilderness of terminals and devops in order to personalize the experience. The "claw" fomo really put that into sharp relief.

[Sol Irvine @solirvine]

@MichaelQiu14 @rcorpatauxTS The linkage between your experience of saltiness and the NaCL is purely circumstantial. It's not evidence that saltiness-experience is an embedded feature of NaCL.

[Sol Irvine @solirvine]

@MattDaWhite Never heard of him. I don't even want to ask, so I won't.

[Sol Irvine @solirvine]

Really triggered a lot of akshually guys here.

[Sol Irvine @solirvine]

@CrystopherLoves I went down the rabbit hole with GPT 5.5 high. Is this idle nonsense, or earnest inquiry?

[Sol Irvine @solirvine]

> Something like 80% of all known animal species are arthropods.

Do you ever think about the vastly-more-successful siblings we have on this Earth? These beings will almost certainly outlive us, and they are living representations of hundreds of millions of years of all-out war.

[Sol Irvine @solirvine]

@baynoni I'm sympathetic to the mind-boggling dev ops challenge, but it's more like "standing in line" than "thinking".

[Sol Irvine @solirvine]

I can't prove it, but it really feels like a lot of Opus' "thinking" time is actually "waiting for resources to free up" time.

[Elliot Vaucher @ElliotVaucher]

So strange to think AI might be redistributing so many cards and at the same time not changing anything to current separation of expertises in the market.

Today, anyone can use off the shelf RL techniques, build a service around the fine tuned llm, plugin tools and MCP and call it a proprietary agent, even messing around with opensource harnesses to try and build something of their own. Then reality kicks in, and you’re faced with latency optimisation at the nanosecond, inference problems and cache issues, costs at the token that ramp up fast, sandboxing on different environments and OS, and you’re back to 101 computer science. How do you optimise the transformation of electricity to tokens at each and every layer of the stack.

Exactly the same way there are people who’s unique set of skills make them particularly good at orchestrating a huge m&a deal, there are people who live and breath information systems. Who dream about algorithms.

It seems to me that the only question anyone trying to play a role or make a move in this revolution should ask themselves is : are we operating at the correct level of the computing stack.

For some, electricity is the answer (GPU, inference), for some it’s model training (RL and so on), for some the harness, and for some the actual integration of these complex systems in their own domain of expertise.

[_gabrielShapir0 @lex_node]

the first NVCA-standard tokenized securities stock sale launched on Ethereum this week on @MetaLeX_Labs

another step toward putting corporate finance on Ethereum, the right way!

[Sol Irvine @solirvine]

The fact that every frontier LLM constantly trips over the NextJS server/client boundary despite having access to best practices skills and the entire internet tells you something about the ergonomics of `revalidatePath`, `router.refresh()`, etc.

[Sol Irvine @solirvine]

@fleetingbits Why would any firm want to create visibility for a third-party platform whose incentive is to enable competing firms and itself? It makes no sense to me.

[Sol Irvine @solirvine]

@daniellevine I've been an em-dash user for decades—one of those people who has to curb my natural written voice because AI adopted a similar style.

[Sol Irvine @solirvine]

@aCameronhuff @MEKowalski Claude Legal is open source and incredibly generic. See for yourself: https://t.co/cMZHaceEbY

It's designed to be extended to fit the firm/client/lawyer.

[Sol Irvine @solirvine]

@MEKowalski @aCameronhuff K&E is not building an LLM, and there's nothing "already out there" that distills K&E's tacit knowledge, guardrails, preferences, and secrets. If I had to guess, it'll be a mixture of LLMs (some fine-tuned, some frontier) and several agent harnesses tailored to practice areas.