iiinova — where your digital self keeps evolving

Babak Farhang @gnahraf.bsky.social · 05/10/2026

@bsky.app historical art on wikipedia is NOT adult content. The labeling engine routinely mislabels wikipedia_dot_org content "adult". Never encountered a case where it can be genuinely argued it could be. Add a rule, whitelist the site pls: many false positves outweigh any potential false negative

100

Babak Farhang @gnahraf.bsky.social · 05/10/2026

I didn't realize PE-owned insurance companies could be hollowed out well more than the cost of purchase. So who pays (directly) for today's 2T+ big-shed build out once they goes under? A good part of it will be the insurance companies' counterparties: policyholders, bondholders, pension funds.

011

Filed by a welfare state against federal net-donor states. Taxation for thee, representation for me. As in, you pay my bills and you listen to me

Dance (Matisse) - Wikipedia

Art historians, will note: the last frame "5 cats dancing in hell" is a juicy inversion of Matisse's Dance paintings.. en.wikipedia.org/wiki/Dance_(...

en.wikipedia.org

000

(The reason why this jumps out to me is that I'm building just such a tool.)

000

Babak Farhang @gnahraf.bsky.social · 29/09/2026

Problem 3 (paraphrasing): Agents don't seem to know who they're working for. A fundamental problem, especially when guardrails are expressed as earlier prompts. A tool that determines precedence (order) and attribution (who) of prompts and agent output, as reliably as a compiler, may help.

210

But before that, establish the rule of law.. empeach these plainly corrupt SC justices (for conduct that would land any lower court judge in jail). Without that, they'll continue to override congressional acts, lower courts, and executive power when it suits them (cf student loan forgiveness).

110

Babak Farhang @gnahraf.bsky.social · 29/09/2026

I've been thinking about this.. He's even managed to break Hollywood. He's destroyed the brand. Like, Captain America was not supposed to be a badie.

020

Babak Farhang @gnahraf.bsky.social · 28/09/2026

Haven't read his work yet, but what an apt topic for our times. Empirical evidence for my own free will is at best circumstantial. So many of my decisions are a roll of dice, on re-examination. Instead, I think of decision making as a process of discovering one's true nature. Moral nature, included

000

Babak Farhang @gnahraf.bsky.social · 27/09/2026

1. It's bad at [new] concept/idea building. 2. It has poor working memory. 3. It doesn't learn/build new concepts. 4. It can't transfer working memory to long term memory (it can't adjust its weights).

000

Babak Farhang @gnahraf.bsky.social · 27/09/2026

So if we're calling Claude Science intelligent, we're actually calling an LLM, its "verification" tooling (calculator, compiler, etc.), combined for epi-cycles LLM blurbs pass verification.. it's that *ensemble that must be intelligent. OK, but if we call it intelligent, let's call out its limits

100

Babak Farhang @gnahraf.bsky.social · 27/09/2026

I agree with his opinion that LLM achievements are low hanging fruit [that are latently baked in their training data]. But to understand what's going on, one must deconstruct the black box: LLM + tooling harness. No part is "intelligent" on it's own (an LLM can't even do arithmetic).

110

Babak Farhang @gnahraf.bsky.social · 26/09/2026

In turn, made me lookup enshittification in French.. emmerdification.

000

Babak Farhang @gnahraf.bsky.social · 24/09/2026

Very cool. Does the outline of the pyramid (in the smaller PIP) represent the focal length of the camera? I assume it must be changing, for otherwise how would an oblique distant corner of a parallelogram appear right. Also, does it also work even a corner is at infinity (local v global)?

Babak Farhang @gnahraf.bsky.social · 22/09/2026

implementation-wise, we likely would not be searching for an exact prompt sequence that faithfully recreates the product. That would be infeasible. Rather, it might be more fuzzy.. how difficult (or easy) is it to regenerate something "close" (given the model).

000

Babak Farhang @gnahraf.bsky.social · 22/09/2026

Here's why this may be possible in principle.. An LLM is a compression of its training data, and assuming, for sake of argument, that its output is deterministic, then the (Shannon, not Kolmogorov) entropy of its output is bounded by the entropy of the input prompts. Implementation-wise..

100

Babak Farhang @gnahraf.bsky.social · 22/09/2026

Should LLM's be viewed as prompt-decompression machines? The point of this viewpoint would be to develop a yardstick by which to measure a product's "originality", (the human hand) in a possibly agent-involved production. Something like a fuzzy attempt at reverse-engineering the input prompts.

100

Reposted by Babak Farhang

John Timmer @jtimmer.bsky.social · 22/09/2026

Ok, this "there are two separate brains" thing appears to be making the rounds, and I want to be absolutely clear: this is a press office overhyping some perfectly reasonable science that won't come as a surprise to most developmental biologists.

32755238

Babak Farhang @gnahraf.bsky.social · 21/09/2026

Also re these ghosts.. Natural languages are distinguished from most engineered, formal languages in their use of [pro]nouns and verb conjugation to bake in this theory of mind. Most "formal" languages (programming, math) lack this: they deal in "is"es ("the operation is surjective"), not in actors.

000

Babak Farhang @gnahraf.bsky.social · 20/09/2026

I was curious about charge, btw cuz I've read it's usually modeled as zero in BH simulations.

100

Babak Farhang @gnahraf.bsky.social · 19/09/2026

Fascinating. Couple of questions re these early blackholes, if anyone cares to chime in.. 1. Would these be expected to have formed with less or more angular momentum density (like divide by total M) than in the stellar collapse scenario? 2. Would early BH's be expected to have net charge?

100

Babak Farhang @gnahraf.bsky.social · 19/09/2026

Superb high level walk-thru of what LLMs do: the surprise they do so well, but also their inability to handle surprises. I'd call the ability to handle surprises, "real-time learning". AI agents (LLMs harnessed to external programs) can't.. bsky.app/profile/gnah... Cory explains it way better 👇

001

Babak Farhang @gnahraf.bsky.social · 19/09/2026

Congrats. PS if you'd had AI correct those leaning vertical lines, it wouldn't have been a crime. 🙃

000

Babak Farhang @gnahraf.bsky.social · 17/09/2026

All true. But to be clear, that last part, their saying it's about to kill us all, that part is BS. The more fear they create, the more "better-be-on-winner-side" investors they attract, the more FOMO for an eventual IPO. It's a money sink, and this hype-based financial engineering *will collapse.

020

Babak Farhang @gnahraf.bsky.social · 17/09/2026

Yea. I'm tired of our meek reporters. Tell it to him in his face: "Why should Americans trust you on the economy, trade, and their finances when they can all see with their own eyes that you'd flunk a middle-school math quiz on percentages?"

Babak Farhang @gnahraf.bsky.social · 16/09/2026

To be fair, a human also uses external tools. But what's missing in LLMs is an ability to *learn, their "context window" is bounded. And cognition *requires an ability to learn, to light a light bulb, to adjust the weights, as it were. LLMs can't. Why the superintelligence threat then is fantasy.

Babak Farhang @gnahraf.bsky.social · 16/09/2026

Whether this combination of LLM / external-toolset can be called cognition, for me at least, is settled. No. The agent does surprise on occasion, eg w a surprising insight exhibiting deep understanding. But then moments later it'll disappoint, reminding I've anthropomorphized a chatbot's epicycles.

000

Babak Farhang @gnahraf.bsky.social · 16/09/2026

LLMs generate Math proofs much the same way that they generate software. To succeed at either task, they rely on external tools for a host of sub-tasks, even basic arithmetic, and ultimately a compiler to check their work. They typically get plenty wrong at each try, but eventually get it right.

200

Babak Farhang @gnahraf.bsky.social · 15/09/2026

All made possible thanks to that slippery extra judicial snake Roberts slithering in the shadow

000

Babak Farhang @gnahraf.bsky.social · 13/09/2026

The ones that still buy Treasuries

Babak Farhang @gnahraf.bsky.social · 13/09/2026

Yes, "1789 Capital", aptly named after the French Revolution. Amazing situational awareness

000

Babak Farhang @gnahraf.bsky.social · 13/09/2026

Doodled this self portrait using "Keep". My profile pic, I did years ago in a .bmp file using Paint Brush 🙃

000

Babak Farhang @gnahraf.bsky.social · 12/09/2026

".. added that he had jumped off the top of the South Tower as well, but the nasty media didn’t bother to take a photo." Perfect.

Babak Farhang @gnahraf.bsky.social · 12/09/2026

We should get to this, but we can't delay the root canal (impeaching corrupt SC justices, restoring the constitutional order, rule of law). Without that, none of this will matter.

Babak Farhang @gnahraf.bsky.social · 12/09/2026

Re AI and the regulatory capture angle.. A lot of people this side of the pond seem not to have gotten the memo: The US has lost international standing. We can regulate the fuck we want.. the world has moved on. To wit, Microsoft Office, EU, Canada, saga of an ICC judge losing access, on and on.

000

Babak Farhang @gnahraf.bsky.social · 11/09/2026

.. we see nominally working product produced by non-programmers that is bloated, cluttered, and unmaintainable. And mathematicians too are now facing the same dilemma: nominally working product (proofs) produced at scale with near zero discernment.

030

Babak Farhang @gnahraf.bsky.social · 11/09/2026

It's true that [almost] no one codes in assembly language anymore, but a programmer must still understand a host of first-principles concepts (data structures, algos, etc) to be good at the job in *any language. That includes the use of natural languages for programming. In the software world..

121

Babak Farhang @gnahraf.bsky.social · 10/09/2026

Do you think Sanders (and/or his staff) is being deliberately obtuse? I don't think he himself would be opposed to getting better informed on this topic. It's easy to be had by the likes of Bostrom, so I'm wondering if and how you reached out to him and his team to set them straight.

000

Babak Farhang @gnahraf.bsky.social · 09/09/2026

Give that zombie a hat. The sun is gleaming thru its combover, the unction melting uneven

000

Babak Farhang @gnahraf.bsky.social · 06/09/2026

Swiveling the inner rectangle along with the outer one as a single plane might also be visually (maybe even pedagogically) interesting

000

I believe you. Like Bluesky, Mastodon can be that way

000

He refutes such rumors in his latest post to the thread (~2hrs ago)

130

Oh, you needn't convince me it promotes brain rot. We've been cooking our brains for a while now.. Even before AI, the fact most of us stopped using pen and paper created a palpable difference in cognitive depth.

000

I like what you're doing with jenesis. Your talk reminds me of my maven build hack, namely shutting down the wifi so that it stops checking central for artifacts it's already downloaded 🙄 🙃

110