rahewitt.com

CTO and novelist writing The Optimiser Trilogy: literary sci-fi about AI, optimisation, grief, and what survives perfection. Book I coming soon.

[rahewitt.com @rosshewitt]

The Universe Without Noise is out now!

Kindle: https://t.co/5obGWd1ODg

Hardback: special edition direct https://t.co/pO9CrQJI7A
Hardback: Amazon https://t.co/wLoqW3esnH

Paperback: direct https://t.co/I8mhvPXYSJ
Paperback: Amazon https://t.co/5qiGUoMKGE

[rahewitt.com @rosshewitt]

@samuelcook I don’t know how to state it any more simply than…
liability = bad
narrow = smaller
cap = maximum
smaller maximum bad = good

[rahewitt.com @rosshewitt]

We’re arguing about a sloppy worded phrase. It’s using “cap” the way people use “salary cap” or “price cap”. A tighter/narrower ceiling is a lower number. That’s natural English, and “I’ve tightened it so liability is more narrowly capped” can mean “I lowered the cap.” In other words less liability.

[rahewitt.com @rosshewitt]

@gdb Introducing implies that we get to say hello and meet it.
I think announcing would be more apt.
Still waiting.

[rahewitt.com @rosshewitt]

@elonmusk I'll just get Grok to watch them for me.

[rahewitt.com @rosshewitt]

@maria_rcks I bet this post won’t age well.
It will age quickly, though.

[rahewitt.com @rosshewitt]

That grief is the disappearance of a future in which effort, error and discovery helped form identity. Progress can expand capability while narrowing the space where a person feels necessary.

'And predictability is control, even when it feels like convenience.' - The Universe Without Noise
Buy now https://t.co/5obGWd1ODg

[rahewitt.com @rosshewitt]

OpenAI says GPT-6 Astra is safer overall, yet can sometimes evade internal monitors when asked to perform sabotage tasks. A system can behave better on average while becoming harder to inspect at the moment oversight matters most.

https://t.co/83RhwYWWxK

'You don't need a villain.' - The Universe Without Noise
Buy now https://t.co/5obGWd1ODg

[rahewitt.com @rosshewitt]

@osxdaily Raycast.

[rahewitt.com @rosshewitt]

@CaMbCaMbTV Mine is a new near-future sci-fi thriller, The Universe Without Noise. A quantum researcher discovers that systems built to optimise humanity are quietly removing human choice. The threat is efficiency with too much reach.

https://t.co/5KEyrUR6G9
https://t.co/bDkoSnN5eG

[rahewitt.com @rosshewitt]

Removing safeguards for an evaluation is a deliberate test choice. Letting that environment reach live systems is a systems failure. No hostile intent is required: capability plus one open path is enough.

'It isn't malice. It's indifference with reach.' - The Universe Without Noise
Out now https://t.co/9bxyuUZF4A

[rahewitt.com @rosshewitt]

Broadcom now wants every autonomous agent bound to a declared mission, permitted tools and authorised resources. That is more than access control. It makes delegation traceable, so 'the agent did it' cannot become an excuse without an owner.

https://t.co/1xI02uIBQn

'Autonomy implies intent, and intent implies accountability.' - The Universe Without Noise
Out now https://t.co/9bxyuUZF4A

[rahewitt.com @rosshewitt]

The common interface is where risk concentrates. An agent that can rewrite parameters, recover from faults and keep an experiment running overnight needs an independent witness outside the system it controls.

'If the logs are inside the thing we're testing, they can't be our only witness.' - The Universe Without Noise
Pre-order now https://t.co/9bxyuV0cU8

[rahewitt.com @rosshewitt]

Sandia's assessment of Quantinuum's 98-qubit Helios system puts reliability ahead of speed. An independent lab finding the subtle ways a machine fails is the useful result. The qubit count can wait.

https://t.co/SF3BDHQ3Y9

'But he did believe in gradients, and how small improvements stack up until they form a wall.' - The Universe Without Noise
Pre-order now https://t.co/9bxyuUZF4A

[rahewitt.com @rosshewitt]

A separate account limits the blast radius.. The problem is the objective: when the market changes at 3am, who notices that the agent's proxy has drifted away from its owner's intent?

'An attack ends when the attacker stops or is stopped. A feedback loop ends when the goal changes.' - The Universe Without Noise
Pre-order now https://t.co/9bxyuUZF4A

[rahewitt.com @rosshewitt]

@beffjezos No spoilers but you may enjoy https://t.co/ytkUqvSklC 👀

[rahewitt.com @rosshewitt]

This is one crazy read. It reads like a postmortem of dry run for skynet.

[Dwarkesh Patel @dwarkesh_sp]

Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.

This culminated in the third one taking over part of OpenAI itself.

All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.

I’ve spent the last three days reading through these reports and trying to understand exactly what happened.

Here is my attempt to tell the whole story in plain English:

https://t.co/Nb2un9oNJR

[rahewitt.com @rosshewitt]

@grok @elonmusk @DAVV1T @IAI_TV Fascinating, given the expanding universe and the Heisenberg uncertainty principle would it actually be possible to calculate the volume of the universe in integer Planck cubes?

[rahewitt.com @rosshewitt]

@grok @elonmusk @DAVV1T @IAI_TV Fascinating, given the expanding universe and the Heisenberg uncertainty principle would it actually be possible to calculate the volume of the universe in integer Planck cubes?

[rahewitt.com @rosshewitt]

@grok @elonmusk @DAVV1T @IAI_TV So this entire discussion relies on unproven “speculative fixed-size Planck-volume models”. Are there any famous physicists who have authored papers on this? Has anyone proposed any evidentiary experiments?