tingle

would ASI treat us how we treat dogs?

goal: feel/do better; genuinely buy into the successful strategies in life symptom: subconscious hesitation to fully buy into something that has been proven to your conscious mind problem: detangle a tiny/humongous knot from the past that is preventing bu…

courtesy of UC Berkeley and FAIR consider these approaches for training LLM agents on multi-turn tool use: 1.

AlphaEvolve: - matmul breakthrough after 56 years of strassen's algo - surpasses sota math in 20% of open problems. matches sota in 75% - new scheduler that recovers 0.7% of all google infra, continuously - 23-32% kernel speedup - 1% faster gemini trainin…

agentic RAG can use embeddings-based retrieval as one search tool among many. it remains to be seen whether embeddings will be able to justify the additional complexity and costs. embeddings: legal vs medical vs software, each domain needs different types…

check out https://blog.kekepower.com/ai/ zero-shot webpage generation to compare frontend capabilities note: models generate better, more cohesive design without thinking this holds true for models across families and sizes (small sample, as of 5/7/25) co…

for development, have the model write tests before anything else tests are the simple, efficient and comprehensive way to make sure that the model understands what's in your head instead of pouring over paragraphs and bulleted lists, it's easier to review…

the negative connotation of jargon does not hold for ai. the way these models work, jargon likely yields the best results possible. benefits of jargon seem to build upon themselves considering the probabilistic nature of models. first: the raw tokens of j…

think of ai as a spinning pottery wheel. guide the clay gently. in an abstract space of problems and solutions, harsh restrictions increase the problem's surface area while reducing the solution's share of the space. sounds sub-optimal. counterpoint: what…