There's a lot of talk about AI and consciousness these days. While I really enjoy that this is a debate people are willing to have in good faith now, I also notice that few of the voices seem to be familiar with the vast philosophical corpus that already exists on the matter of (hypothetical) AI consciousness. Many arguments have been made before, as were good counter-arguments. I think the Consciousness Atlas gives a good intuitive understanding of just how many different ways to define, detect, and operationalize consciousness there are.
The conversation on AI and consciousness is incredibly relevant. It is incredibly frustrating that most commentators are utterly unfamiliar with over 60 years of inquiry into this in both philosophy and computer science.
Start with a hand-wavy definition of consciousness. Move the goal post whenever your stated prerequisite of consciousness is reached and resort to unfalsifiable assertions about qualia.
And throw in some category errors while on it: when you're talking to Claude, you're not actually calling a stateless LLM directly, you're talking to an AI system (and yes, that's often just three LLMs in a trench coat). But claims about the topology and workings of a single LLM are as relevant to the question of consciousness as claiming that humans can't be conscious because the limbic system doesn't technically support it.
Coincidentally, I just attended a fantastic conference on machine consciousness (https://machine-consciousness.ai). It's a fantastic place where literally all speakers disagreed with each other, and yet found an incredible amount of common ground.
Tangentially related: one of my all-time favourite neuroscience papers by Iriki et al. [1] showed that the mouse pointer becomes part of the body schema in a real, measurable way.
Basically there are neurons whose receptive field (ie the subset of the outside world that causes the neuron to fire) is "everything a monkey can reach". Flash a light in that area, that neuron fires, flash it beyond that area, neuron stays silent.
Now if you give the monkey a rake, the neuron's receptive field immediately grows to encompass the space it can now reach with the rake too: the rake becomes part of the body schema, not part of the outside world [2].
But if the rake instead is just a stick but it controls a mouse cursor on the screen, the _area of the screen_ the monkey can interact with with the cursor becomes part of the receptive field of that neuron too. That suggests that the cursor itself becomes part of the body schema.
TL;DR don't mess with people's mouse cursors, it's like cutting their limbs off.
Hyperspell | YC F25 | San Francisco, ONSITE | $150k-220k + generous equity and benefits | hyperspell.com
I'm Manu, founder of Hyperspell. 14 Years ago I found my first job in tech through Who's Hiring. I take this job board seriously :)
Hyperspell is building Context & Memory for AI agents. We're at the intersection of "future of AI and agentic computing" and "this actually works". We're not just changing how our customers use AI in their products, but are constantly iterating on how we use AI to build Hyperspell, and what it means to be a software company in the decades to come.
To be clear I’m not claiming that Mythos is _nothing_ but PR hype, merely that Anthropic is playing its cards really well, which is a claim independent of actual capabilities of their latest model.
Ha. Non-native speaker here although you wouldn’t be able to tell what talking to me, until you hear me confuse when to use this vs that, and lose vs loose. Some things my brain just refuses to remember.
The whole artificial scarcity Anthropic created around Mythos / Glasswing is quite brilliant to be honest (I’m Not saying ethical, just brilliant). The commercial gains are one side of course. But consider this:
Gets labelled supply chain risk by the pentagon. Hypes up what they claim to be the most advanced hacking tool on the planet. This puts the US government into a loose / loose position. Either deny the NSA access to it, or be called out on their bluff.
We have asked ourselves that question repeatedly over the past year. While I don’t have a simple solution, I have some mental models that may help.
Overall, there are two knobs to tune, each with a few strategies:
1. reducing the number of times you have to switch context
2. reducing the cost of switching
Let’s start 1.
- The easiest of course is to have less agents in parallel.
- Clustering interventions. When starting a new session, use plan mode or similar, have the agent interview you until it has a good idea of what to do, don’t move away from the window until it’s ready to execute. Read the thoughts to stay on it without switching until you’re confident it understands your intent
- invest heavily verifiability. That means make it easy to check if the final code correctly and exhaustively captures your intent. Let it write specs first and update specs as necessary during implementation. Have righteous integration tests and “digital twin” mocks for external integrations etc. have an adversarial prompt that reviews whether the code matches the specs.
Then reduce the cost of switching:
- i usually plan my work to have only one “heavy” task, and then 2-6 agents working on small tasks, ideally straight from tickets. My brain stays with the hard tasks, the easy ones should be in and out
- wait until all the easy ones need input, then do a round of those and go back the hard tasks
- prompt the agent to give you a brief summary every time it stops (what the goal was, bullet list of what it did so far, what it needs from you).
Finally: be okay with staring at a spinner. Day dream. Listen to music. Enjoy that the robots are doing work for you. Won’t try to optimize every second by also checking emails, responding on slack, or god forbid open hacker news. Just do one thing - code - and allow yourself to live in the terminal for an hour. Then take a break.
Hyperspell | YC F25 | San Francisco, ONSITE | $150k-220k + generous equity and benefits | hyperspell.com
I'm Manu, founder of Hyperspell. 14 Years ago I found my first job in tech through Who's Hiring. I take this job board seriously :)
Hyperspell is building Context & Memory for AI agents. We're at the intersection of "future of AI and agentic computing" and "this actually works". We're not just changing how our customers use AI in their products, but are constantly iterating on how we use AI to build Hyperspell, and what it means to be a software company in the decades to come.
We run an OpenClaw agent for our entire team — he lives in a group chat (although we have DMs too).
- Runs our standups, checks in withe everybody EOD on blockers
- Already know what we shipped on Github and Linear so it can focus on the work that's not tracked and summarize it in the morning for everyone
- Helps with debugging customer issues
- Keeps up with twitter and competitors and lets us know if they launch new features
Besides, I'm honestly blown away by the social aspect of it. I was honestly pretty skeptical at first, but having an AI team mate is actually _fun_. There, I said it. Everybody on the team said they'd be sad if we took it away.
I'll do a write-up on our setup sometime this week, I hope others will find our approach to security posture and multi-tenant usage insightful.
I love the vision, but it glosses over the most important difference between Infrastructure and Companies.
Infrastructure as code is prescriptive. The code is the source of truth, and the world gets crested from it.
Company as code is descriptive. It is constantly catching up to meat-space, rather than creating it. Changes are gradual instead of instant roll-outs. Patterns change over time and only get documented later.
Making the company code prescriptive would require an insane amount of discipline that might be more stifling and restrictive than it is freeing.
This goes back all the way to the beginnings of Psychology. William James, who is considered the somewhat of godfather of Psychology, argued that all feelings are bodily feelings; ie. emotions are caused by bodily sensations. Your heart is not beating BECAUSE of anxiety, rather your beating heart IS anxiety. You don’t tremble because you’re afraid, you’re “afraid” (a complex emotion mediated by stories we have) because you tremble.
It’s a theory psychologists and philosophers still argue about.
They're all great, but the 2012 "The Uncensored Picture of Dorian Gray" is the closest to the original script before the editor cut out things that he deemed... checks notes... "too gay".
It restores parts that were cut, and essentially bans chapter 3 and some other digressions on art history that Wilde added as a literary Beard to the footnotes - still there to read, but set in context)
It's not a huge different honestly, but I believe Oscar Wilde would want you to read that version.
I literally just hired Ben Horowitz last month, but I must assume that mine is the better systems and integrations engineer so I consider myself having getting the better deal.