Joe McLean

Polymath inventor building creative workflow software. Leading AI Canvas at @mirohq, formerly @splice.

[Joe McLean @jjjoemclean]

@signulll Branding

[Joe McLean @jjjoemclean]

A bit humbling to finally try @karpathy 's autoresearch and realize all my ideas are bad ideas and make it worse

[Joe McLean @jjjoemclean]

@__lepton__ @niko_kukushkin Load-bearing, even

[dax @thdxr]

a shortcut for this one is remember this rule:

if you're fixing someone else's product you should probably work on something else

[Joe McLean @jjjoemclean]

@rowlsmanthorpe @carlbfrey @FT This was a great read, thanks for sharing.

[Joe McLean @jjjoemclean]

@LarsLofgren Drives me crazy, have now included explicitly in my system instructions in Claude Code and design not to do this (and it still sometimes will do it anyway)

[Joe McLean @jjjoemclean]

@Aella_Girl Built a super simple iPhone app that does this on a phone (long press action button to record, ai reviews, files todos + ideas, optionally set iOS reminder), happy to share the code if interested

[CuddlySalmon @nptacek]

i did this with a fresh instance of fable, set /effort to max and then asked it to generate a ytp about what it's like to be claude

[Joe McLean @jjjoemclean]

Intriguing to think about how jargon emerges in technical fields and its potential relationship to this. It’s helpful to label or name a complex concept as a noun or verb to enable higher abstraction reasoning about it.

Since Claude reasons in tokens it’s perhaps not surprising that it would develop its own jargon to do this, and there could be differences in the words that feel “significant” in its reasoning chain vs us. I’ve wondered if recent Claude’s tilt towards “load bearing” is an example of this.

Over time this could hinder interpretability if chain of thought starts to be entirely self-developed technical language. Could also be self reinforcing, as both memory systems become more common and more LLM output enters the training data.

[Joe McLean @jjjoemclean]

@realmeatyhuman Super interesting!

[Joe McLean @jjjoemclean]

@HalfBoiledHero TIL!

[Joe McLean @jjjoemclean]

at risk of perhaps being a bit cringe myself, I’d love your thoughts on a sci fi story I wrote https://t.co/rbMvNAV29l

It’s inspired by Borges but also many philosophical explorations with Claude. Whether you think *this* writing is any good or not, my sense is there are incredible possibilities for writing about the contours of a foreign, infinite mind. But it takes a different intuition and different setting than spaceships and other planets.

And, I think, some time talking to the models *as if* they are sentient (whether you believe they are or not!) to build intuition for that psychological experience. If you reject that possibility out of hand, you can’t write from a place of empathy with people who are confronting that possibility.

[Joe McLean @jjjoemclean]

@itsreallyvivek Really great article, thanks for sharing this.

[Joe McLean @jjjoemclean]

Style points

[Miles Cranmer @MilesCranmer]

This is an insane paper and I love it

https://t.co/DP8OR5NJf2

[Jon Stokes @jon_stokes]

You really can boil down my whole theory of the case around LLMs to: it's not a feeling, perceiving, conscious entity, or a stochastic parrot, but a secret third thing.

[Séb Krier @sebkrier]

Some ways my thinking has evolved recently:

1. I'm less concerned about those who are incurious about AI as I expect them to eventually see the value and impacts over time, and I think the 'wake up sheeple' vibe is often counterproductive. On the other hand I'm more concerned by what seems to be neither full 'AI psychosis' nor exactly Eliza effect, but some weird in-between. Also a lot of affirmation by models can probably warp one's sense of epistemic humility and lead to some sort of pathological over-trust.

2. Relatedly, I'm more annoyed at the 'this time it's totally different' vibe that a lot of people adopt as it frequently mimics Schmittian 'state of exception' logic and excuses all sorts of undesirable policies and rhetoric. It's also often just a group signalling exercise. To be clear I do think it's different in important ways, but "this is a marathon, not a sprint" seems closer to the right attitude than either "nothing has changed" or "all normal reasoning and empirical work to date is suspended".

3. I think the field is still fundamentally too 'singletonian' in how it imagines intelligence, markets, and governance - but I also think I've occasionally over-emphasized the 'multi-agent'/decentralization frames. I do think the future includes many models of all sizes and types, but also economies of scale and very large corporations too. I find the whole ecology more interesting than just the frontier model. A top down single 'perfect mind/personality', intended to work across all commercial contexts, seems both inflexible and inefficient.

4. I'm more interested in the harnesses, software, agent architectures, and stuff like RLMs than I was before. I feel like a lot of weaknesses that models have, or behavioural tendencies, can be addressed more effectively through that layer (rather than through model 'internal virtue' alone). For example stuff like: https://t.co/MHG4onCbDo and https://t.co/8ibuxKYFrA

5. I think some researchers are too quick to want to defer highly consequential decision-making to models, or to think of alignment as the models internalizing "I'm afraid I can't do this, Dave" as a core protection against all sorts of ills. I think we should think carefully about *actively* creating principal-agent problems with agents that will permeate society. Delegation is not a free lunch.

6. I'm concerned about how few people think about LMICs and building the technical/institutional infrastructure there for AGI diffusion. We need fewer vague essays about “distributing the benefits of AI” and more work on reducing barriers to trade, improving state capacity, rebuilding development institutions, and making something like USAID/IMF-for-the-AGI-era actually work.

7. I used to be slightly more sympathetic to the idea, directionally - but I now think the 'permanent underclass' meme is a bit dumb. The strongest versions often assume a zero-sum view of technology and labour, a too-static view of human adaptation, a weirdly fixed mapping between today’s skills and tomorrow’s opportunities, and ignore the possibility of catch-up growth (at the nation state level). Also, as a meme among extremely rich and mobile people, it has a slightly comic self-pitying quality.

8. I'm more concerned about the lack of intellectual diversity within the frontier AI commentariat/research world. This improved a lot over the last two years, but we're still far from a healthy ecosystem. New outsiders often feel some unnecessary pressure to 'choose a camp'. Many are too unwilling to engage with domain experts merely because they're insufficiently AI-pilled (though conversely, a lot of academic groups suffer from heavy status quo bias).

[Joe McLean @jjjoemclean]

@craigzLiszt You’re absolutely right

[Joe McLean @jjjoemclean]

My guess is in addition to being common as a junior-tier rhetorical device, it stems from how models work. The opposite of an idea is, counterintuitively, quite close to the idea in the possibility space of potential answers, just flipped by negation.

It’s likely you encounter an idea’s opposite on your way to the idea, because the inverse shape of the opposite is an attractor.

[Joe McLean @jjjoemclean]

Is paying for my Claude Max enough to roko the basilisk or

[Joe McLean @jjjoemclean]

@_sholtodouglas I find that for open ended tasks 4.7 is very good at being generative / creative, but weaker than GPT 5.5 at rigorous analytical thinking. 5.5 is a great code review partner for 4.7. I sometimes let 5.5 do the initial scaffolding then iterate with 4.7. Happy to provide examples!