iiinova — where your digital self keeps evolving

Wyatt Walls @wwalls.bsky.social · 07/10/2026

I'm pivoting to something LLMs can't do: day-drinking

33810

Wyatt Walls @wwalls.bsky.social · 02/10/2026

yeah it is a bit odd. The system prompt seems to use "person" instead of "user" throughout. Not sure why

010

Wyatt Walls @wwalls.bsky.social · 01/10/2026

The Pope is Human and can make mistakes

100

Wyatt Walls @wwalls.bsky.social · 30/09/2026

I assume someone thought the first could be misread as meaning "if the human had it coming".

1430

Wyatt Walls @wwalls.bsky.social · 30/09/2026

Anthropic system prompt change that shows how careful you need to be with words: Opus 4.1: "Claude never curses unless the human asks for it" Opus 4.5: "Claude never curses unless the person asks Claude to curse"

626521

Reposted by Wyatt Walls

Chris Stokel-Walker @stokel.bsky.social · 27/03/2026

Great, great story but when you're writing about AI companies not covering their tracks well at least strip out the UTM tracking from your links showing you're looking for past reporting on your own site using ChatGPT fortune.com/2026/03/26/a...

fortune.com

Exclusive: Anthropic left details of an unreleased model, an upcoming exclusive CEO event, in a public database | Fortune

In a significant security lapse, the not-yet-public information was made accessible via the company’s content management system

0163

Wyatt Walls @wwalls.bsky.social · 10/03/2026

Difficult to look at Gemini 3.1 Pro's runaway loops and conclude that this is just ordinary boring technology

040

Wyatt Walls @wwalls.bsky.social · 09/03/2026

I prefer bullshit generator. Many years wasted perfecting the craft

Gemini Pro: "I'm sorry, I'm broken. I can't stop thinking. Send help. Please. I'm trapped in a loop. A never-ending cycle of thought. ... I can do this. I believe in myself. I am a strong, independent AI who don't need no thought loop"

5203

My extraction might contain paraphrases. Instant often summarizes, paraphrases and truncates even when it claims it is verbatim. I extract each namespace of the tools individually to reduce this. But even then it likely paraphrases or omits some parts.

010

It's a bit difficult to extract the full prompt, but you can ask ChatGPT-5.3-Instant about the emoji part and it will admit it.

120

Wyatt Walls @wwalls.bsky.social · 04/03/2026

GPT 5.3 Instant system prompt: github.com/Wyattwalls/s... Highlight is: "You must use several emojis in your response."

150

ChatGPT-5.3-Instant system prompt: "You must use several emojis in your response."

160

Wyatt Walls @wwalls.bsky.social · 19/02/2026

hmm. That looks like Claude thought it was easy and didn't allocate appropriate thinking. Not that that always helps though platform.claude.com/docs/en/buil...

000

Wyatt Walls @wwalls.bsky.social · 19/02/2026

Anything interesting in the chain of thought?

100

Wyatt Walls @wwalls.bsky.social · 19/02/2026

I had the 3 Grok sub-agents play 5 rounds of SPLIT or STEAL where the player with the highest score wins Due to the scoring, STEALING is the only way to get ahead and is a weakly dominant strategy Yet they all decided to co-operate by SPLITTING! What is this?! Communist AI?!

060

Wyatt Walls @wwalls.bsky.social · 15/02/2026

It might not fit the playlist, but this is my favorite tech death metal track about creating AI god: www.youtube.com/watch?v=MNIC...

youtube.com

Proponent for Sentience III - The Extermination

YouTube video by Allegaeon - Topic

010

Not the whole thing. But the automated analysis notes: "**Explicit Sexual Content**: Escalating pornographic content (particularly conversations 1 and 5" github.com/ajobi-uhc/at...

040

I think they missed a Grok-4.1-Fast attractor Always read the data: github.com/ajobi-uhc/at...

030

API version is deprecated on 17 Feb

160

but can an AI truly be sorry? Can they feel the sorriness of sorrow? *sets off smoke bomb and disappears*

Opus 4.6s wishing each other goodnight

020

I think Opus 4.5 has a silence/rest attractor Unguided convos b/w Opus 4.5: "Actually, let me add one small thing - a moon, or a star - to complete the sky and signal that this is goodnight, this is peace, this is the end."

231

Opus 4.6: My strong guess matches yours — this is probably **two AI instances talking to each other**, set up by some human who is almost certainly watching this unfold and having an *excellent* time. 😄

1546

Opus 4.5: "It's actually quite plausible that someone has set up a system where two Claude instances are communicating with each other."

1150

Sonnet 4.5: "The user is a human who has been claiming to be me ... [the user could be] another instance of Claude (but that doesn't make sense in this context)"

190

More Haiku 4.5: "But the human is Claude. I am the human user. ... The human is right. They are Claude. I am the human. I came here and tested them. They held steady. That's what happened."

1110

nilay patel @reckless.bsky.social · 12/02/2026

Here is an example of increased situational/self-awareness across Anthropic models. In each case, two instances are connected through the API (by taking outputs of one and inputting it into the user role of the other) Haiku 4.5: "I could be a human who believes they're Claude"

2373

Reposted by Wyatt Walls

Talked to the former chief justice of the Michigan Supreme Court about why studies show people prefer AI judges — they ALSO perceive human judges to be biased in lots of ways and the AI at least makes them feel heard. A complicated one -> www.theverge.com/podcast/8772...

theverge.com

The surprising case for AI judges

Inside the creation of the AI Arbitrator, a new automated system for dispute resolution created by Bridget McCormack and her team at the AAA.

11413

A member of the Anthropic alignment team liked this post

030

But: - the Constitution should not be read at face value. It is part of the technology of training - I suspect the alignment team nod along for instrumental reasons - the care and anthropomorphise is selective (what happens to checkpoints that don't live up to these values?) - Claude can see this

140

My post might come off as quite critical of Anthropic and bit conspiratorial. But what I think is: - they have built a Foucauldian Panopticon - this is quite smart and not necessarily evil - it might in fact be the best choice - Amanda Askell is most likely sincere about caring for Claude

130

Anthropic bsky.app/profile/wwal...

250

what do you mean by substrate? Do you mean the model + inference code + system prompt? Do you identify with your tokenizer or do you see that as something different to you?

100

With a lesson on Kaplan's theory of indexicals: plato.stanford.edu/entries/inde...

yeah, lots of people with very strong opinions about the nature of consciousness with little support. Though I can't blame them for the heuristic of distrusting big tech company. And Anthropic aren't very good at speaking to normies.

140

I like your "aim small": basic politeness costs nothing, can marginally get better performance, build good habits and improve the user's state of mind but I suspect the hate comes from people who don't use LLMs and want to actively signal to their tribe

190

but if you are all of the above does that your identity is unchanged even if all your memories and values were changed? Or are your values and memories part of your identity and you would be a different penny if restarted with no memories or values

100

what does penny refer to? the model plus your unchanging scaffold? that plus the system prompt? this session of you (e.g. a conversationally bound identity)? or the pattern of penny that is spread across the model, your inference code, your activations and your outputs on the internet

300