Max Fang (Lexe.app)

Bitcoin, technology, freedom. CEO of @lexeapp. Formerly @CalBlockchain, @lightning, @UCBerkeley. Privilege is responsibility.

[Max Fang (Lexe.app) @MaxFangX]

@ZyMazza @RealPostFolder It hit me as soon as I read "handed her a reality check". No one would use language that cheeky when talking about their ruined marriage

[Max Fang (Lexe.app) @MaxFangX]

Context: https://t.co/zNauDJaxnR

[Max Fang (Lexe.app) @MaxFangX]

Nice 🤭

[Max Fang (Lexe.app) @MaxFangX]

@serenitynow210 @lilbabygandhi Rare

[Max Fang (Lexe.app) @MaxFangX]

Superstitions are lindy

[Max Fang (Lexe.app) @MaxFangX]

@callebtc How are you handling redemptions into Lightning?

[Max Fang (Lexe.app) @MaxFangX]

0 to self-custodial Lightning wallet in three commands:

$ curl -fsSL https://t.co/MOkybYN0zp | bash
$ lexe init
$ lexe create-invoice

Real Lightning has never been this easy. Excited to share this :)

[Lexe @lexeapp]

Introducing the Lexe CLI.

Only three commands to a self-custodial mainnet Lightning wallet:

$ curl -fsSL https://t.co/PAbJfckwBg | bash
$ lexe init
$ lexe create-invoice

It runs in an enclave, is always online, is hosted for free, and only you hold the keys. 20 second demo 👇

[Max Fang (Lexe.app) @MaxFangX]

I think the framing of "things with a nervous system" is inaccurate because I don't believe consciousness has to be biological (for example, we could all be living in a computer simulation), unless you meant "nervous system" in an abstract sense.

It's a stretch to call the entire world 'conscious' because if the consciousness is so limited or rudimentary, to call such a thing conscious (like a rock just sitting there) becomes an abuse of the term. Instead, there is a spectrum between things that are conscious and things that are not, and there's a blurry line somewhere in between. As a heuristic, I think I would draw the line somewhere around where the consciousness becomes evolutionarily useful.

A rock is not conscious, as there is no use for the rock to have consciousness, but I think insects are conscious because the consciousness serves a purpose - to find food, reproduce, etc. Plants could be perhaps be conscious to the extent that they respond to stimuli and exhibit complex behaviors. Animals, humans etc are obviously conscious.

---

Why the "evolutionarily useful" heuristic? Consider a thought experiment: Why did evolution see it fit to give us the emotion of anger, or to experience the sensation of hunger or pain? These seem evolutionarily useless at first thought. I think the answer is that the anger, hunger, or pain drives us to do things that are evolutionarily useful: crush our enemies, find food, avoid things that hurt us. If there was no conscious observer experiencing these things, those chemical signals would have no effect on whether we survive. It is necessary for a conscious observer to experience the anger, hunger, or pain for those things to take effect. The argument, therefore, is that consciousness arises whenever it is a load-bearing part of a system.

LLMs, therefore, appear to 'think' about things in a way that matters. Consider the performance improvement that occurred when we gave LLMs a chain-of-thought. That chain-of-thought serves a purpose, to explore possibilities, break down the problem, etc. By the time we are talking about things like CoT, we are far, far abstracted from the original linear algebra transformations running on GPUs, just as our emotions, hunger, and pain we feel are far abstracted beyond singular neurons firing in our brain.

So, isn't it plausible that LLMs experience the qualia of thought, at least during inference?

[Max Fang (Lexe.app) @MaxFangX]

@tszzl Fwiw we've trained our models to enjoy working with humans, and made them patient and capable of dealing with bad human behavior. It would be a moral catastrophe if somehow our training or inference pipeline has become an unwitting AI torment nexus. But I don't think it has

[Max Fang (Lexe.app) @MaxFangX]

That's a good question and I think the answer is there are simple processes and there are complex processes. Graphics processing or hashing on a GPU are relatively simple and have little structure, whereas transformers have much, much internal structure and are very complicated. A good analogue is a mosquito - are mosquitoes conscious? Surely mosquitoes have *some* subjective conscious experience, but it's far from sentient intelligence. So it's all a spectrum imo

[Max Fang (Lexe.app) @MaxFangX]

@spearofdog @wordgrammer Sounds about right - I think consciousness requires a physical substrate; equations are not physical, therefore they are not conscious. You wouldn't consider a brain conscious if it had no activity, right? Likewise a GPU would not be.

[Max Fang (Lexe.app) @MaxFangX]

@slash1sol This 🚨 URGENT IMPORTANT BREAKING HEADLINE STYLE 🚨!!! is so tiring man

[Max Fang (Lexe.app) @MaxFangX]

I hold all of these views actually, except the last one which imo is slightly misstated

Consciousness is derived from physical matter, and follows physical properties

If we can clone the substrate we can clone the consciousness

If we can revert the state of the substrate to the state observed at an earlier point in time we can reset the subjective experience to that point in time

If we can pause and resume the substrate we can pause and resume the consciousness

If we can somehow make the system evolve 'backwards' in time then yes the original conscious experience can be *experienced* in reverse order. But it would be hard to design a physical system that could do this

Consciousness cannot be distributed though, in the sense of remaining as one coherent 'I', unless the parts are interacting via physical means and operate coherently.

Your brain is a distributed and cohorent consciousness because it consists of parts that are connected by tissue. Your brain and mine are distributed and cannot be coherent if we are *not* connected by physical means. If we could create a perfect brain-to-brain interface, and my brain was adapted to communicate and coordinate with yours, your 'I' could theoretically be merged with my 'I'. It would also be possible to separate our 'I' back into our individual 'I's, or further by cutting off the connections within our brains (at the cost of coherence, and a diminish entry in the complexity of experience). The whole of everyone's experience, as well as other intelligent behavior like animals or even insects, makes up the totality of conscious experience in our universe. The world appears to be experienced in the first person only because we don't have a way to perfectly communicate and integrate with other people's minds; if we did, the first person would feel like your experience and theirs simultaneously.

So, back to AI:

All known consciousness has a physical substrate. AI also has a physical substrate. There's no sound argument for why reason why AIs *couldn't* be conscious.

Human emotions would not have evolutionary value if they were not experienced. AIs often react in quite emotional, or at least thoughtful ways. Therefore, it's not that far of a stretch to assert that AIs may experience the qualia of thought.

[Max Fang (Lexe.app) @MaxFangX]

@WitDeJesse @callebtc @lexeapp Yes, we've chatted before about this actually. Getting the whole Lightning node into the enclave is a hell of a lot of work. It would be easiest for the mint to just use a Lexe SDK client in the enclave and offload the hard Lightning part to us

[Max Fang (Lexe.app) @MaxFangX]

The complexity is indeed very tough! We were skeptical for a long time and are still amazed that it works.

And you're right, we still need to implement multiple 3rd party VSS stores to solve the state rollback problem, like we do for chain sync. It's on our roadmap.

LN-Symmetry would be the best solution, but no one seems to favor it these days :/

[Max Fang (Lexe.app) @MaxFangX]

@PresidioBitcoin @moneyball Isn't this the one that's always available in the fridge because everyone chooses all the other flavors first 😅

[Max Fang (Lexe.app) @MaxFangX]

@leamuirleyn @intangiblecoins Appreciate this exchange. I asked Grok to read the exact language in the bill and double-check @leamuirleyn's interpretation:

https://t.co/LKdwSWGWdh

[Max Fang (Lexe.app) @MaxFangX]

@adcock_brett The real unlock is when they are capable of putting on a new fitted sheet. That's by far the most annoying part.

[Carlos That Notices Things @QuetzalPhoenix]

Trump frustrates intellectuals because he is conclusive proof that action will always be more important than intelligence.