Adverb

they mainly tested this for simple LoRAs/finetuning but this should be even more useful for folks wanting to do modality adapter retrofitting without losing timestep distillation arxiv.org/abs/2605.05204

D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Models arxiv.org

v good reporting www.wired.com/story/ai-pr-...

Ed Zitron Gets Paid to Love AI. He Also Gets Paid to Hate AI wired.com

arxiv.org/abs/2601.18053 Neat! Ask for a random Shakespearean play before you give your prompt to improve output diversity.

Addressing LLM Diversity by Infusing Random Concepts arxiv.org

Would guess gemini is more anguished. More seriously, arxiv.org/abs/2510.24797 hqas a table on this

Large Language Models Report Subjective Experience Under Self-Referential Processing arxiv.org

axios had an article on how Build American AI is affiliated with that Leading the Future pac www.axios.com/2025/10/29/a...

axios.com

thinkingmachines.ai/blog/defeati... from Horace He

Defeating Nondeterminism in LLM Inference thinkingmachines.ai

not thinking LLMs are interesting is just a mix of tech illiteracy & millenial tech lash. It's honestly that simple

people who'd list "anti-AI" as part of their personality will typically put as much thought & due dilligence into who they boost & listen to as your boomer grandpa who has fox news playing 24/7

So many baby-brained takes on generative art by people who apparently exist in a world without poetry or conceptual art.

Lol literal openai pac psyop on twitter that gets such a fuckton of engagement posting what amounts to bluesky takes

People proclaiming on whether "AI" can be used to create novel information could instead take a deep breathe & simply ask any chess player or any academic studying protein folding

ugh i have finally reached "this was a much faster & more effective experience than google search" using an LLM. pain

millenials nodding along to a pr expert spreading misinfo re LLM determinism on podcasts is really just the circle of life. a lot of people in every generation will listen to their generations' huksters on talk radio eventually

mmm people have an elaborate way to prompt chatgpt to generate horror images but if you just politely ask it, it'll give you horror images. I guess the bizarre aspect is that vaguely asking it to make up a "maybe disturbing" image generated literal realistic gore for me immediately

why's the pope post about ai more than me

i had this fun notebook called "fight club" that ended up having the funniest reward hacking behaviour & I think that some API LLMs are finally good enough that I'll be at least printing out interesting jailbreaks for cheap if I run the notebook using them as judges