mr. TIM

AI Architect | North Carolina | AI/ML, IoT, science

WARNING: I talk about kids sometimes

🌉 [bridged](https://fed.brid.gy/bsky/timkellogg.me ) from 🦋 [timkellogg.me](https://bsky.app/profile/timkellogg.me ), follow Bridgy Fed for the Atmosphere (Bluesky) (npub1pqr…2sck) to interact

Web site: https://bsky.app/profile/timkellogg.me
Link: https://timkellogg.me

Last Notes

2026-10-19 15:24:17 UTC

Fable 5.5 incoming
and its strength is just more of what made the original Fable 5 so good
[www.theinformation.com/newsletters/...](https://www.theinformation.com/newsletters/ai-agenda/anthropic-readies-model-reads-users-intentions-better )
https://lepista.us-west.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:ckaz32jwl6t2cno6fmuw2nhn&cid=bafkreihamsbnahuvdmwwjbis77zylde5yjlm7fdgpcywsinbyeomjn5zru

2026-10-02 18:58:23 UTC

Chris Olah & Anthropic bitterly disagree with Pope Leo on the possibility that Claude may be conscious
[www.thelettersfromleo.com/p/nyt-anthro...](https://www.thelettersfromleo.com/p/nyt-anthropic-nearly-walked-out-on )
https://lepista.us-west.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:ckaz32jwl6t2cno6fmuw2nhn&cid=bafkreia65nefqkozznbgltjgab644kni42zfhltraijjj7wdlteif4n3lm

2026-09-27 12:35:02 UTC

- reply

with that said, the “full steam ahead despite the risks” is kind of a dumb take, equivalent to ignoring your monitoring
imo Jensen’s framing is probably closest — treat it as an engineering problem, and stop/bail out if your monitors are blaring failures

2026-09-27 12:31:32 UTC

- reply

that’s not guaranteed to work, but if you have enough such early signals, you can also detect if it’s not working and pivot
so i have split perspectives on this
1. the public should be aware of x-risk
2. we should push forward anyway, bc that’s how engineering is done

2026-09-27 12:29:39 UTC

- reply

my engineering instinct is to approach x-risks through monitoring — detect extremely early signs and try to reduce them
for LLMs in particular, those early signs are also signs of the product being worse
e.g. Opus 5 was not aligned, and therefore Opus 5.5 fixed the bug
[bsky.app/profile/timk...](https://bsky.app/profile/timkellogg.me/post/3mwdyk7sxqk2h )
RE: https://bsky.app/profile/did:plc:ckaz32jwl6t2cno6fmuw2nhn/post/3mwdyk7sxqk2h
#note1mup…9g2w

2026-09-27 12:20:57 UTC

i would say, in most contexts it is probably unwise to focus on risks you’ve never experienced
e.g. in software engineering i’ve developed a bit of a gag reflex whenever an engineer is solving a problem that doesn’t exist yet
trouble is the severity of x-risks puts them in a unique position
https://lepista.us-west.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:ckaz32jwl6t2cno6fmuw2nhn&cid=bafkreiholsbzl5hgmqail6tbtgq2wbkcghihsiscvnt4qmkdhj2dybxkty

2026-09-27 12:20:57 UTC

- reply

i don’t think we, as a society have figured out what to do with x-risks
we have a long history of worrying about x-risks that turned out to be duds. then again, if we’re wrong merely once we get wiped out