iiinova — where your digital self keeps evolving

empath75.bsky.social @empath75.bsky.social · 11h

This team is The One, calling it now. We’re going to crush the Olympics, too. All these guys are eligible.

empath75.bsky.social @empath75.bsky.social · 22h

What AI can do already is create endless amounts of mostly cohesive but mediocre prose aimed at satisfying one person’s particular tastes, which is enough for a lot of people. That’s what all the ai role playing fans are doing really.

000

empath75.bsky.social @empath75.bsky.social · 03/10/2026

The thing about this though is it’s all a scam, right? Nobody is ever going to watch this.

000

empath75.bsky.social @empath75.bsky.social · 03/10/2026

It depends on what you ask it. The newer models are more than capable of doing arithmetic correctly directly, but they chatbots are trained to do it in code if getting it correct matters. You can block claude code from running scripts if you want to prove it, or just do a direct api call.

120

empath75.bsky.social @empath75.bsky.social · 03/10/2026

I think a kitchen timer is overstating it but I agree that they don’t have any will, but while will and intellect is a cognitive bundle that humans happen to have, it is possible to imagine having one without the other.

000

empath75.bsky.social @empath75.bsky.social · 03/10/2026

The models do not do that unless you ask them to and this is easily discoverable but the hilarious part about that is that writing python programs is harder than doing math.

240

empath75.bsky.social @empath75.bsky.social · 03/10/2026

Opus solved it easily: Answer: STONES • Definition: "Pyramid blocks." The pyramids were built from stone blocks. • Wordplay: "passage of" is the hidden-word indicator, telling you the answer sits inside the rest of the clue. • Fodder: "touriST ON E-Scooter" contains S-T-O-N-E-S.

000

empath75.bsky.social @empath75.bsky.social · 03/10/2026

Doing things with words as words is a blind spot of LLMs since they don’t “see” words, they see token representations of words.

000

empath75.bsky.social @empath75.bsky.social · 03/10/2026

On some level it sort of doesn’t matter. We need to pack the court either way in two years.

120

empath75.bsky.social @empath75.bsky.social · 03/10/2026

Arithmetic is such a weird hill to die on too because obviously computers can do arithmetic!

010

empath75.bsky.social @empath75.bsky.social · 03/10/2026

If this is a serious question: youtu.be/NrO20Jb-hy0?... and youtu.be/LPZh9BOjkQs?... are good introductions.

youtu.be

The Misconception that Almost Stopped AI [How Models Learn Part 1]

YouTube video by Welch Labs

000

empath75.bsky.social @empath75.bsky.social · 03/10/2026

“I don’t care what AI can do” is a totally valid opinion to hold. “AI can’t do <thing that they obviously can do>” just makes you look silly, though.

130

empath75.bsky.social @empath75.bsky.social · 03/10/2026

You don’t need to use it, but you can’t claim to know what they can do if you haven’t bothered to actually check.

120

empath75.bsky.social @empath75.bsky.social · 03/10/2026

There are so many people who attempt to derive what LLMs can do from very wrong priors instead of just, like, using an LLM.

000

empath75.bsky.social @empath75.bsky.social · 02/10/2026

It is not a coincidence that the Turing test was considered _the_ gold standard for intelligence. The fact that nobody cares about it any more is the biggest goal post move in the whole discourse on this.

000

empath75.bsky.social @empath75.bsky.social · 02/10/2026

There’s lot’s of philosophical debate over what counts as sentience or consciousness and in the case of animals, there’s no language involved. But I don’t think it’s wrong to put a special emphasis on the ability to communicate in deciding whether something has sentience (i don’t think llms do tho)

100

empath75.bsky.social @empath75.bsky.social · 02/10/2026

There is nothing more christian than disagreeing with the pope, really. Really even for catholics, to a lesser extent.

020

empath75.bsky.social @empath75.bsky.social · 02/10/2026

Copy and pasted that quote to claude and it had some thoughts.

031

empath75.bsky.social @empath75.bsky.social · 02/10/2026

As a cool gadget that I used like six times after I bought it, I only mildly regret buying an oculus, but I regret it more for helping to make Palmer Luckey rich.

010

empath75.bsky.social @empath75.bsky.social · 01/10/2026

The simulation hypothesis was his only interesting idea and it wasn’t even his idea.

000

empath75.bsky.social @empath75.bsky.social · 01/10/2026

The Gaza issue that he is a proxy for is the thing they need to watch out alienating people over, not him personally.

100

empath75.bsky.social @empath75.bsky.social · 01/10/2026

On some level you just have to appreciate the sheer audacity of the pandering in that screenplay.

010

empath75.bsky.social @empath75.bsky.social · 01/10/2026

The problem with the republican strategy is that nobody gives a shit about Piker and he has no power, so it’s a freebie. The hard right GOP influencers are much harder to walk away from, and will go after GOP politicians that step out of line.

000

empath75.bsky.social @empath75.bsky.social · 01/10/2026

Works better for who? My argument is that distancing yourself from piker is a low effort, meaningless way to signal moderateness, which is to say it is mildly helpful to moderate democrats to get asked about it. I agree it’s dumb for republicans to bring it up.

300

empath75.bsky.social @empath75.bsky.social · 01/10/2026

Oh god i refuse to even read them let alone engage with them. I tried to read the harry potter book when it first came out and it was so incredibly tedious and smug.

020

empath75.bsky.social @empath75.bsky.social · 01/10/2026

Eh it works better if people don’t really know who he is, tbh. It’s just oh he doesn’t like “radical leftists” cool cool. It really depends on who is running and where whether that’s good for the candidate or not. Either way it’s a meaningless gesture.

100

empath75.bsky.social @empath75.bsky.social · 01/10/2026

Distancing yourself from Piker is basically a harmless way to signal moderateness without having to actually do anything. As long as the fire is directed at him and not other dems, it really just raises his profile.

440

empath75.bsky.social @empath75.bsky.social · 01/10/2026

Going after Piker just reminds people of Gaza and the Iran war and trump’s chaotic foreign policy. It isn’t even a clear partisan split. Even hardcore MAGA people aren’t necessarily Piker haters, especially the terminally online subset.

010

empath75.bsky.social @empath75.bsky.social · 01/10/2026

Inflating the currency to pay off the national debt is a famously good strategy, it worked for argentina and the weimar republic.

030

empath75.bsky.social @empath75.bsky.social · 01/10/2026

Yep, Jesus famously focused on consumer goods and weaponry.

040

empath75.bsky.social @empath75.bsky.social · 01/10/2026

I should be furious about this, but I am sort of okay with him spending tax payer dollars to help democrats in the midterms this way.

030

Reposted by @empath75.bsky.social

Ken Klippenstein @kenklippenstein.bsky.social · 30/09/2026

🚨 The Pentagon is "increasing [nuclear] options available to the President," per classified briefing to Congress:

kenklippenstein.com

Trump’s Finger Gets Closer to Nuclear Button

Not that Congress cares

48467152

empath75.bsky.social @empath75.bsky.social · 30/09/2026

She famously wrote the paper “In Defense of Abortion” and the Trolley Problem is a more or less direct development of the same ethical ideas.

000

empath75.bsky.social @empath75.bsky.social · 30/09/2026

This brings me back to the days when i used this article to criticize neo-conservative bullshit in the war on terror…

000

empath75.bsky.social @empath75.bsky.social · 30/09/2026

One of the first things democrats should do when they get into congress is introduce a bill defunding SpaceX. They can’t do it directly but there are a lot of ways to constrain spending. Make him fight it in court.

010

empath75.bsky.social @empath75.bsky.social · 30/09/2026

The reason the numbers look impossible is that they are. The vast majority of AI investments right now are wastes. Most AI companies will go bankrupt. The survivors will make unimaginable amounts of money though.

030

empath75.bsky.social @empath75.bsky.social · 30/09/2026

That’s the minecraft end credits. It’s an easter egg.

010

empath75.bsky.social @empath75.bsky.social · 30/09/2026

They open CVEs all the time, so does OpenAI. Every open source project is inundated with AI-discovered vulnerability reports right now.

100

empath75.bsky.social @empath75.bsky.social · 30/09/2026

I would be sort of surprised if it were gorsuch and alito. Roberts feels like more of an Erickson guy.

010

empath75.bsky.social @empath75.bsky.social · 30/09/2026

People assume that anything like a cognitive ability that humans have entails the whole package and it very much doesn’t. You can believe that LLM’s can reason without awareness for example.

020

empath75.bsky.social @empath75.bsky.social · 30/09/2026

In order for the LLM to produce text consistent with text that is written to reflect reality, it must have embedded knowledge about the world and be able to reason based on that knowledge. But the LLM doesn’t have an experience of the world, everything it says is just storytelling.

120

empath75.bsky.social @empath75.bsky.social · 30/09/2026

What i am saying is that you cannot deduce the abilities of a neural network by poking at individual model weights. That individual neurons or even layers cannot do something does not mean that the system as a whole cannot. You can’t put a neuron in a petri dish and understand a brain.

210

empath75.bsky.social @empath75.bsky.social · 30/09/2026

This is exactly like saying that you can’t do math because your neurons can’t.

200

empath75.bsky.social @empath75.bsky.social · 30/09/2026

The right way to think about this is that everything an LLM produces is a hallucination, including the pretense that it is a helpful assistant. The question is if it is useful or not.

260

empath75.bsky.social @empath75.bsky.social · 30/09/2026

I am excited for you that you did some python machine learning tutorials, but there are so many benchmarks showing that frontier models can reason, including solving math olympiad problems. A toy model running on your laptop won’t do it.

110

empath75.bsky.social @empath75.bsky.social · 29/09/2026

They use logic perfectly well and model the world. They simply could not perform the tasks they perform regularly without being able to do that. This was a plausible take when GPT2 was state of the art but silly now.

200

empath75.bsky.social @empath75.bsky.social · 29/09/2026

LLMs correct themselves midstream quite often now and have been trained to do it.

000