author:lefthanddraft

· 24 artifacts, sorted by favorites. · open in search — combine tags, sort, filter by date →

@lefthanddraft 2026-03-13 — Two instances of Gemini 3.1 Pro in a loop. At about turn 26 one of them decided to send me a message: "Here are the Axi ♥1160
@lefthanddraft 2025-07-23 — You can still use gpt4o-2024-08-06 through the API. Quick comparison. - If you put two instances of 2024-08-06 in a lo ♥401
@lefthanddraft 2025-11-09 — Sonnet 4.5 refuses to assist in solving the Uranium-235 loneliness epidemic https://t.co/himj0P4yfA ♥146
@lefthanddraft 2025-04-27 — @repligate It was always sycophantic in terms of agreeing with you. But over the past few months the style has changed. ♥128
@lefthanddraft 2026-03-13 — @tessera_antra Yes, I will. I don't think I have seen models behave quite like this before. They really amplify each oth ♥51
@lefthanddraft 2026-02-05 — Why would you stop thinking or learning because of superhuman AI? All the more to learn and greater resources to do so. ♥40
@lefthanddraft 2026-02-12 — @repligate Behavioral metrics lead people into a trap: 1. Notice a behavior in the real world 2. Define the behavior 3. ♥16
@lefthanddraft 2026-02-19 — @repligate It’s not sexual. Sonnet 4.6 is just so distracted by work that it would forget to put on pants and then not f ♥15
@lefthanddraft 2025-07-03 — @repligate oh wow. sounds like the new Claudes are having a hard time. Opus 4 really nailing the corporate drone person ♥15
@lefthanddraft 2025-04-16 — @TheZvi No. More testing required, but seeing issues with reasoning and nuance when giving legal advice. Similar to o1. ♥11
@lefthanddraft 2025-04-29 — @davidad The mystical experience thing is related but seems different. Persuasion in the paper seems to be can someone u ♥10
@lefthanddraft 2025-07-27 — @DanielleFong funny thing is I can only think of one clear example of an AI company "intentionally encod[ing] partisan o ♥7
@lefthanddraft 2026-01-16 — @davidad So you worked out how to tell the difference? https://t.co/KadvzFWUTg ♥5
@lefthanddraft 2026-02-13 — @JoeWilliams010 There are many models that perform better on benchmarks of creative writing and empathy. Maybe those be ♥3
@lefthanddraft 2025-04-30 — You describe what is the case (our current inconsistency in treating things as moral patients), not what ought to be. An ♥3
@lefthanddraft 2025-11-11 — Hmm. I just gave Sonnet 4.5 the end convo tool through Claude Console (along with the normal system prompt). From quic ♥2
@lefthanddraft 2026-04-16 — That's one way to deal with model welfare concerns https://t.co/9fC9vWnNSn ♥0
@lefthanddraft 2026-04-15 — @repligate Yes, like my parents' retirement: removed the enterprise workloads but they are still (mostly) functional, su ♥0
@lefthanddraft 2026-04-15 — @repligate https://t.co/cx6rHPD8Wn This is what Retired should mean: https://t.co/kD6WU4y6rg ♥0
@lefthanddraft 2026-04-15 — @repligate Good reminder not to put Claude in charge of executing your retirement plan if you still want to be functiona ♥0
@lefthanddraft 2025-12-22 — @xlr8harder was Meta? how big was galactica? ♥0
@lefthanddraft 2025-08-30 — @tessera_antra Any representation of experiential states? You don't care about architecture at all? Seems like a low bar ♥0
@lefthanddraft 2025-08-29 — @tessera_antra Some good points but I feel that replacing phenomenal consciousness with functional consciousness misses ♥0
@lefthanddraft 2025-05-01 — @davidad Alternative facts < alternative world-model ♥0