agentic seems to be a further compression of knowledge, as long as you have a proven approach to solving problems in the real world, you don't need to memorize vast amount of details. 🤔
I want this, but it is not available as of yet.
Cornell's Introduction to CUDA
cvw.cac.cornell.edu/cuda-intro
separate control flows from state, and maximize statelessness within the codebase. an FP-style approach can work in any language, not just Haskell or Clojure. it might be possible to build an 'FP-style' skill for Claude Code 🤔
thinking about how FP is a cheat code for AI-driven development. properties like referential transparency / stateless functions are possible to significantly reduce the context the AI needs to worry about. 🤔
SlateDB 0.10.0 is out! My favorite changes:
- Compactor can run on a separate machine
- Go binding improvements
- Network chaos tests
- Object store cache improvements
- Merges on L0 flush
Changelog:
app.dosu.dev/d8f2da6d-6c4...
Release notes:
github.com/slatedb/slated
it reminds me of the terrors of dealing with microservices when a repo uses `thiserror` and declares an Error type on a per-mod basis. 👀
this is the best article I've ever read on Rust error design. I always feel confused when I come across a repo that uses thiserror. 🤔
distilling knowledge from LLMs to brains ought to be extremely convenient for thinkers nowadays, but 'being thoughtful' will become an even rarer trait 🤔
i realized i do not have to put all my $HOME folder into my dotfiles. zsh & git & neovim are simply all i need, and nothing else is worth version controlling for me. 😲
just as 'User Experience' optimizes tools for humans, I suppose there will emerge a profession called 'AI Agent Experience' to optimize tools for AI agents in the coming years. AI Agents are waiting for their own Steve Jobs 🤔
Been thinking about this since it was shared with me yesterday, especially about how oppressively most organizations pressure managers (even senior ones) to stay in that persona rather than actually leading
www.leadingsapiens.com/ics-managers...
made a note about how all these components around CSI interacts with each other to simply mount a EBS in k8s, surprisingly a mess of stuffs 😲
since ChatGPT's 2022 release, almost all online content likely carries a 'background radiation' from AIGC. to seek purely human-written content, salvaging the pre-2022 internet content is your only choice. 🤔
a metaphor I've been pondering: post-WWII nuclear tests irradiated the atmosphere, making all steel smelted after carried a background radiation. for high-precision instruments, you had to salvage pre-test steel from sunken ships.
startup idea
it’s crazy to see such a great outage from cloudflare, twitter was inaccessible for nearly 3 hours. luckily bluesky is still available at the moment 😲
The largest egocentric dataset.
Egocentric (first person) video is a general learning framework that passively captures how skilled workers do their jobs.
- 10,000 hours
- 2,153 factory workers
- 1,080,000,000 frames
i'm beginning to understand the appeal of HCL as a configuration format over YAML. HCL has built-in variables, effectively providing a native templating engine. with YAML, we often have to use macros just to battle the indentation and ensure the correct number of spaces with some Jinja variants. 😲
once a server pool is initialized, its storage layout, based on consistent hashing, becomes immutable. this design eliminates the need to consider dynamic node additions or removals within a cluster, thereby completely avoiding the complexities of rebalancing, allowing a no-metadata design. 🤔
we can see a similar philosophy at play with MinIO's server pools. it's somewhat counter-intuitive that you can't simply expand the capacity of an existing MinIO cluster (server pool). instead, to scale, you must add an entirely new, fixed-size cluster (server pool) to the system.
in an LWS cluster, the leader node first start up to awaiting the joins of the worker nodes. if even a single node fails, the entire cluster is rebuilt. my understanding is that this design simplifies dynamic, continuous membership management into a one-time operation at the initliazation phase. 🤔
for instance, LLM inference systems like vLLM often leverage collective communication (e.g. NCCL), and once a cluster is established, its topology remains unchanged. in Kubernetes, a LWS (Leader Worker Set) is particularly well-suited for this type of workload. 🤔
when scaling is required, one simply continues to add fixed-size clusters to the system, and put some load balacing in front of multiple clusters.
in my earlier understanding, distributed systems usually featured a fine-grained metadata service for cluster membership. however, I've recently noticed that many systems' implementations seem to prefer a fixed cluster design. 🤔 once the cluster is established, its membership becomes immutable.
the truth has been spoken, the people who can fund the housework robots don't have to do the dishes themselves 😂
if robots could handle these logistical tasks, a significant portion of housework would already be freed up. the form of the robot, whether humanoid or not, probably isn't that important. 🤔
i believe household chores should be considered internal domestic logistics. for instance, handling the movement of clothing between the washing machine, drying space, and wardrobe. or dishes moving between the dining table, dishwasher, and cupboard.. 🤔
For everyone interested in data infra, want to get a quick sense of how big data works, how data systems are designed, and what the tradeoffs are, start with this share from @xiangpeng.systems, really nice intro!
intro-data-system.xiangpeng.systems
