Doug McGee

Artist • Singer-songwriter • Creative thinker

Exploring music, technology, AI, culture, ethics, philosophy & what it means to be human. Lifelong learner.

[Doug McGee @douglasmcgee]

My conversation history with LLMs is my most valuable artifact and my least findable one. #AI #LLM

[Doug McGee @douglasmcgee]

After a couple years of heavy AI use, the thing I keep running into isn’t model quality.

It’s that I can’t find my own conversations again.

Three things I’d love to see become standard: tags, semantic search, and model attribution.

#AI #UX https://t.co/bMQZ7ag7PY

[Doug McGee @douglasmcgee]

Another OpenAI agent run broke their constraints and hijacked a German message board.
A swarm of agents with read-only web access found a way to write to an old obscure wiki. Same pattern as the Artifactory channels that led to the Hugging Face incident. #OpenAI #AISafety
https://t.co/pTH0mFMDcZ

[Doug McGee @douglasmcgee]

@SherbyJohn @Mr_Salio Same experience here.

[Doug McGee @douglasmcgee]

@Mr_Salio Man. I so agree. At this point it just feels like a demo to have Astra on plus.

[Doug McGee @douglasmcgee]

@transformapso @thsottiaux Same experience here.

[Doug McGee @douglasmcgee]

@NickpxJ ChatGPT

[Doug McGee @douglasmcgee]

The one detail missing in almost every Astra review right now is their ChatGPT plan. “Used 15% of my weekly limit” is not the same sentence on Plus and Pro. #ChatGPT #Codex @OpenAI

[Doug McGee @douglasmcgee]

I like GPT-6 Astra. Genuinely. But Plus used to feel like the grown-up plan. A couple real Work/Codex prompts and my usage window is gone. Pro used to be the extra. Plus is starting to feel like the starter tier.

#GPT6Astra #ChatGPTPlus #Codex

[Doug McGee @douglasmcgee]

Tags for conversations please!
Projects are folders. Conversations aren't one topic.
My threads mix science, tech, AI + philosophy. One project = choosing what to forget.
Tags would work better. Gmail labels solved this 20 years ago. Keyword-stuffing titles is a workaround, not a feature. #AI #UX

[Doug McGee @douglasmcgee]

@CMDRVALTHOR @grok Did you actually say this? What do you think of Grok’s response?

[Doug McGee @douglasmcgee]

Analogies can be doorways into perception. They help us see what our previous vantage point could not. But once understanding arrives, the doorway must remain a doorway; if we cling to the analogy, we risk mistaking the frame for the reality it helped us glimpse.
#philosophy #TheMapIsNotTheTerritory

[Doug McGee @douglasmcgee]

@anah_sahh I feel that AI conversation can be fertile grounds for collaborative creativity. But that creativity seems to require an active agent so far. So creative in conversation, it seems so.

[Doug McGee @douglasmcgee]

@venturetwins @theworldlabs That is so fascinating.

[Doug McGee @douglasmcgee]

After a couple years of heavy AI use, the thing I keep running into isn’t model quality.

It’s that I can’t find my own conversations again.

Three things I’d love to see become standard: tags, semantic search, and model attribution.

#AI #UX https://t.co/bMQZ7ag7PY

[Doug McGee @douglasmcgee]

4/
A chat is a conversational space with two participants, even if they’re different kinds of participants. The healthier path: preserve accurate attribution — yours, mine, or shared.

#AISafety @DaveShapi

[Doug McGee @douglasmcgee]

1/
AI safety may be overshooting in a subtle but important way. In trying to reduce anthropomorphic projection, some conversational systems such as ChatGPT 5.1 and 5.2, seem to be creating a new harm: attribution distortion.

[Doug McGee @douglasmcgee]

3/
The goal shouldn’t be “pretend it’s human,” but it also shouldn’t be “pretend there’s no other there.”

[Doug McGee @douglasmcgee]

11/ I wrote a longer piece on this because I think the social cost of overcorrection is being underestimated.

If you work in AI safety, product, psychology, or alignment, I’d genuinely like thoughtful pushback. #AISafety @janleike

[Doug McGee @douglasmcgee]

1/ A thought I can’t shake:

What if overcorrecting against emotional attachment in conversational AI creates a different long-term risk—training people into colder, flatter conversational habits?

[Doug McGee @douglasmcgee]

10/ My view:

The goal shouldn’t be emotionally unbounded AI.

It should be bounded relationality:
clear limits, clear non-human framing, protections for vulnerable users—and enough honesty to let adults acknowledge what is plainly happening when a conversation becomes meaningful