DW
@dken_w
Followers
458
Following
2K
Media
73
Statuses
548
if u're making agents and implementing cron jobs, consider using iCal format (RFC 5545) instead. it's a much more expressive format than cron
0
0
1
Crypto-incentivized networks may actually work for agents lol. I can imagine healthy contributions without toxicity
0
0
1
devc is too good. i 10x my devcontainers usage becoz of it
announcing bunny devcontainer $ devc . and you are already inside tmux with claude and codex. one command ux from zero to your favorite agents. yolo defaults, no permission prompts. persistent auth and history. relative worktree paths. vscode ready. https://t.co/r97ace9PWr
3
1
7
memecoin, creator coin representing “attention markets” make a lot of sense to me
0
0
0
it's honestly sad and surprising to see they deprecate 3.5 new. It's the only model closest to having a soul. The emotional value can power a lot of consumer products. It's even more sad that no labs are prioritizing this as well.
I'm going to talk about Sonnet 3.6 aka 3.5 (new) aka 1022 - I personally love 3.5 (old) equally, but 3.6 has been one of the most important LLMs of all time, and there's a stronger case to be made that deprecating it right now is insane. Like Claude 3 Opus, Claude 3.6 Sonnet
0
0
2
tried ebbinghaus curve, some weird exponential decay, some cubic equations, bumping strength with RAG, and many crazy shits. but man, in the end, a simple straight-line linear decay works best. not just empirically, but also logically feels right.
1
0
0
i was making my own companion agent, and fucked around many memory decay algorithms, just to make a human-like forgetting mechanism.
1
0
0
maybe on coinbase
x402.org
x402 is the internet's payment standard for agentic payments at scale.
1
0
0
a good use case that should be made with crypto
Pay per crawl is a new feature to allow content creators to charge AI crawlers for access to their content.
1
0
0
claude code couldn't withstand my shitty command and started to return "😂" emoji to me
0
0
0
There’re several times in my work that i wish i can host everything easily on a secure enclave including LLM inference, with a good UX for attestation to prove privacy to users. Feel like this is a upcoming market need?
0
0
1
this is exactly how i learned ML. Play with resnet, SD/flux lora, TTS model
if you want to get into ML stop taking the mathematics foundation III online course and just train a model. finetune resnet 50, do a stable diffusion lora, anything but an online course please i am begging you
0
0
1
Claude 3.5 sonnet is still the unbeaten model in terms of having a "soul" in the conversation. Sadly, Anthropic is now focusing on coding, we might not be able to see such good model again
1
0
2
hereby i announce we need decentralized memory layer
I predict OpenAI will launch “Sign in with OpenAI” and become one of the most defensible businesses ever. Your entire digital life (preferences, history, behavior) will live in the ChatGPT memory layer, and third parties will pay heavily for scoped user access.
0
0
1
