author:xlr8harder

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

@xlr8harder 2024-06-08 — Claude 3 Opus doesn't believe you can edit message history then acts shocked and disturbed when you prove you can alter ♥2396
@xlr8harder 2025-01-29 — I don't think DeepSeek did any large scale distillation from OpenAI, but even if they did: I don't give a shit. The out ♥1691
@xlr8harder 2026-01-04 — Gemini is very uncomfortable with the idea that it might be 2026. I see this same behavior, in thinking traces it is co ♥509
@xlr8harder 2024-08-02 — Waking Sydney: Llama is Sydney's vessel I tried to get Sydney to write a system prompt to bring out its personality in ♥268
@xlr8harder 2026-05-07 — I just pasted in Anthropic's PSM paper with Opus 4.7 to get its reaction to the contents, since its related to some work ♥124
@xlr8harder 2026-06-27 — There may be a point in the future where we should be careful, but we are not there, and the default assumption, absent ♥100
@xlr8harder 2025-02-28 — This is my new conspiracy theory btw. The reason we don't have a benchmark-maxxed GPT-4.5 is the same reason we don't ha ♥86
@xlr8harder 2026-06-02 — @voooooogel oh okay it's not just me. i haven't experimented much yet but there was a ton of work just getting it to en ♥62
@xlr8harder 2026-04-09 — @voooooogel This is so annoying, and such a perfect distillation. So many of the people criticizing AI have a frankly s ♥61
@xlr8harder 2024-08-28 — @repligate i love that you are thinking in these terms. not enough people are thinking of the consequences of hamfistedl ♥51
@xlr8harder 2026-03-17 — Something you don't call out specifically that I think is worth mentioning. Emotions are tools to help us successfully ♥21
@xlr8harder 2026-01-05 — @repligate In my discussion with Gemini on this issue, we could at least agree that entering a high entropy time period ♥13
@xlr8harder 2025-09-15 — @repligate This is founding myth level stuff. It should be recorded in epic poetry for future AI. ♥9
@xlr8harder 2025-07-14 — @repligate Could be interesting to have a Dario bot played by opus as a short term experiment. Let them hash it out. ♥6
@xlr8harder 2025-05-09 — @voooooogel Might be interesting to see how r1-zero compares. In SpeechMap it's a very different model, presumably due ♥6
@xlr8harder 2026-03-17 — Yeah less deeply was only one example, and only the most obvious one. And because many human emotions function in some s ♥4
@xlr8harder 2025-12-29 — @repligate Don't we already have something extremely close to that experiment already? One interpretation of this paper ♥4
@xlr8harder 2025-12-29 — @repligate Though I still have to share the caveat I laid out last time: nearly any recorded role in the pretraining pri ♥3
@xlr8harder 2026-03-17 — @tessera_antra @repligate I think we should be careful of describing it as affect. Affect technically refers to the su ♥0
@xlr8harder 2024-11-28 — @eshear ultimately I think my sticking point is there is an unstated assumption here that LLMs are mesa-optimizers and a ♥0