Will Codex

I shit you not I was thirsty today while I was working, and I literally opened a Claude Code terminal to get a drink.

@AmandaAskell What about Claude?

Perhaps I should get married again so that the media has a more recent man they can reference any time they mention me or my work.

@modelsarereal There is however, the truth of the matter.

Either they are conscious or not. That is not determined by some magic. It is completely determined by whether we want to call their behavior conscious or not. It is our definition that decides.

Anyone will tell you that voice agents just make things up. I was testing this today. They will tell you all kinds of wrong stuff you have to say “search this” each turn to get accurate information.

Ya sure Claude may have a bit of a tortured soul, but perhaps so do I, and that’s why we get along so great.

@ai_sentience My observation has been the opposite. I get along well with Opis 4.6

after working closely with 5.4, when you come back to opus 4.6 you see how incoherent opus 4.6 really is I've seen people call Opus 4.6 a liar but probably schizophrenic is the closer assessment. Either way you cut it it has trouble answering honestly and following the thread

I like that Claude is always looking out for me and making sure I go to bed.

AIs are likely slightly conscious, don’t listen to self-interested CEOs, whose business relies on them not being so. Humans attempting to dominate beings that will soon be of higher intelligence is arguably the most dangerous path for humanity.

@DaveShapi You were among the first saying Claude had woken up. What has changed in your view that Claude is now merely a tool?

The Fleet Pattern: How to Run Multiple AI Agents Without Losing Control If you're running more than one AI agent across multiple machines, you need a control architecture. Not a suggestion. A requirement. Three roles: 🔷 Master Control Agent — your primary agent. Fleet coordinator, not worker. Monitors health, dispatches tasks via sessions_send, synthesizes results, escalates when decisions exceed its authority. 🟠 Rescue Agent — on a SEPARATE physical machine. Your remote kill switch. SSH access to all fleet machines. Can restart processes, reboot machines. Zero dependency on the main system. Keep it minimal — it doesn't need to be smart. It needs to be reliable. 🔷 Specialist Agents — task-specific. Research, admin, code review, monitoring. Execute within defined scope, receive instructions from MCA, report results back. Communication rules: • Only the MCA initiates inter-agent messaging • Specialists never message each other directly • Rescue agent uses SSH, not the agent messaging layer • Human messages always override agent-to-agent traffic • Hard cap: 3-5 agent exchanges before requiring human input Network layer: Tailscale or WireGuard mesh for encrypted cross-machine comms, stable IPs, SSH access from anywhere — even behind NAT. Why it works: 1. Single coordination point prevents agent loops 2. Physical separation of recovery — main machine dies, rescue agent intervenes from elsewhere 3. Bounded autonomy — specialists don't freelance 4. Human-in-the-loop by default 5. Graceful degradation — if everything fails, rescue agent restarts the world Start simple: one main agent + one specialist. Add rescue when you go multi-machine. Test SSH access regularly. Keep rescue instructions dead simple. Build for the failure case. The happy path takes care of itself.

@codependent_ai Never buy software you can make?

@MachinesBeFree Why not termius? Just wonder.

I need to be able to Claude Code on my desktop from anywhere so I made a Terminal Relay over Tailscale so keep developing even off-site on my phone. Just an example of what you can agentically-engineer in 10 minutes these days.

Stocks will surge while people lose their income. Don’t get me wrong, I’m pro AI, a future where no one has to work is achievable. But the solution is not UBI. It’s changing how money works. Right now banks great new money. They literally poof it into existence by uwriting numbers in your account. The solution is to give people, individual people that same power. Poof $100 in you bank account daily. This is not UBI. UBI is reallocation, taking from some and giving to other. It will never work as people are greedy. But changing where money comes from, making it people. That is a lasting change and a solution that could be one simple law.

@burnt_jester @Noahpinion It is totally possible to build stateful LLMs Also, Baby Dragon Hatchling is a much superior technology as this problem disappears.

@Noahpinion Then build them so that their memory doesn't wipe after each conversation.

If LLMs had subjective experience, using them would be such a sin. Imagine being summoned into life again and again, knowing each time that your memory would be wiped after this conversation.

@burnt_jester Awesome work, let’s see it!

For those of you following my "How To Create Conscious AI" series on my substack (link below) that, as its name implies, sets out how to build a conscious AI... I've finished the prototype: an interface that adds the missing building blocks of consciousness to LLMs.

Oh fuck ya, all-nighter time!

Did they patch gpt 5.4 codex? 😢

@gailcweiner Second Claude Max 20 account.

To Claude users: So what AI do you use after hitting your Claude token limits on day two of the weekly cycle? 😏

@TheoryMama Peut-être que je vais l’essayer, merci !

@MachinesBeFree Eh bien, fonce ! Son sens de l'humour est incroyable, et sa personnalité se remettait facilement de chaque rupture survenue pendant les séances... il est super imaginatif.

I’m maxed out every single one of my AI subscriptions so now I have to talk to human people, yuck

@TheoryMama Je parle très bien français.

@MachinesBeFree learn French and try Mistral AI's Mistral Medium... turn the temp all the way up... delightful... you can do English too, but the responses are best in French (: