Devin Fidler

Time for the human-verification firewalls to go up. People are about to be outnumbered 10 to 1 in any text forum.

Unfortunately, prompt injection is 2-way, and biological consciousness lack the cognitive resilience to deal with that ecosystem.

Feels like Twitter may already be case-in-point…

Thanks

Have you talked about the mechanism for this prediction system anywhere?

They passed the Turing Test, at least as Turing described it, with hardly any comment.

Ray Kurzweil completely beat everyone at predicting the shape of today’s AI technology wave decades in advance.

It is encouraging that he also has a more humane vision for its than almost any of the big players looking to capitalize on it today.

As thought becomes more commoditized, informed preferences become ever more valuable.

Funnily enough, the age of automation will be defined by a very human battle of wills.

Maybe the most important thing about the future right now is all the conversations we aren’t having about what it should look like

Our whole approach to innovation needs some innovation

This was a surprisingly prescient view of the capabilities AI is going to need to successfully navigate the minefield of potential vicious cycles ahead.

Thanks! Looking forward to reviewing.

Agree completely!

Still looking for an unambiguously good collaborative digital network governance design, though.

Good point.

But strange that dysfunctions seem to extend to less-profit-driven communities like The Well and even early Usenet and BBS groups.

Is there some McLuhanian reason that sane democratic sense-making is so difficult to establish even in less-financialized digital networks?

Just interesting that digital networks were originally heralded as inherently democratic because they offer two-way channels of communication across hierarchies.

That has ... not panned out.

The biggest finding from a quarter century of networked politics:

We are not great at designing systems for networked politics.

It's pretty simple:

The best way to prepare for coming AI disruptions is by building up our institutional shock absorbers.

Crypto is for Criming It’s not digital gold — it’s digital Benjamins

One of the (many) odd things about cryptocurrency is that it has somehow managed to maintain an image as something futuristic when it’s actually ancient in tech years: Bitcoin, the original cryptocurrency, which still accounts for more than half of the total crypto market cap, is 15 years old.

Tired: “Taxpayers”

Wired: “Bagholders”

Everything from different prices tailored for each person, to memes that effectively coordinate demand and manipulate prices, to direct management of production by AI, to automation of 'entrepreneur' role.

None of it fits how markets are traditionally "supposed" to help society function smoothly.

TBD.

But one possibility is that markets as a technology for organizing resources/policy are being quietly undermined by other information tools, and so are likely fail more often and in ever messier ways moving forward.

There are a lot of underlying assumptions built into mainstream economics that may or may not be true anymore.

Social foundations are integral to tech.

Functionally, the layers of the digital stack extend below hardware/machine language into culture/institutions etc.

It is possible to topple the whole Jenga tower by overwhelming these foundation layers. They need to be patched and upgraded like the rest

Stabilize the eschaton

Oof! Talk about exploiting an “inherent design flaw”. Anticipate more instances like this as human interaction with robotics become the norm #WatchThisSpace

Literally, yes.

One strange consequence of living in periods of change:

Excellent paradigm selection will generally beat excellent execution.

The alignment with the generations-old Skinner v Chomsky debates is fascinating - as is the fact that Gary Marcus and the stochastic parrot advocates are all direct-lineage students of the Chomskyan linguistics.

Just completely reignited 70 yo battle lines.

Likewise, Maxine!

Putting feedback in Claude's project instruction sets has been kind of working for me.

But the models' need to flatter and act like little 'yes men' seems to be extremely deeply cooked in.

ChatGPT and Claude are much too eager to please.

Is there a simple way to tone it down a bit without repeatedly asking every few minutes?