1,060 karma · joined November 15, 2022
I've always appreciated Solid's DX/mental model (I _never_ have to think "when does this re-render?") so v2.0's continued wins on DX are remarkable. I never thought a release could be about how much can we _cut_. Primitives at work!
IDK why it's so much more chill than the "child abduction" alerts I get. Those are straight up adrenaline-inducing so I've disabled them.
The lawnmower can't hate you. Don't anthropomorphize the lawnmower, no matter how clever it sounds. (Actually not that hard with Opus5)
In starker terms: if you're ok with "vibe principles", "vibe invariants", "vibe lint-rules", "vibe skill updates", and "vibe agent-tooling"... well. That's your call.
Thankfully after a dozen or so retros the low hanging fruit starts to disappear. But until then, you've no idea what's there.
> Let's conclude with a retro. Did you run into any issues during this session that you think could be improved? Any failed tool calls, confusing docs/prompts, or tricky wording that took you effort to figure out, etc? Any final thoughts that you want to raise? Anything minor you didn't mention? Help make this codebase easier for the next agent to work in.
It has ideas for coding changes, spots unmentioned small bugs, suggests invariants (I have a high-level list of invariants to keep), principles to adopt, lint rules, tooling tweaks (I've built SO MUCH agent-tooling; so many useful bash scripts), skill-file updates, follow up work, all kinds of stuff.
Listen to your agent's whinging.
> AI "Proves" Collatz Conjecture with Lean 4 Bug
/s
Yep, AFAIK. I have a statusline that tells me how far I'm into the cache lifetime. I tend to have dense prompts and carefully read its outputs, so by the time this outage occurred when I was ~50m into it. Cue frantically spamming "esc esc up enter", and 10m later it's dead Jim.
Edit:
> We are currently investigating this issue.
> Posted 1 hour ago
There goes everyone's cache.
Yes Boris, tell me more about how "coding is solved".
Edit: literally just now, while writing a claude code hook, opus5 gave me this LOC because apparently the changelog and what's in the transcripts _differs_; great docs d00ds.
# Present only for a subagent's calls. Both spellings accepted: the changelog
# names `agent_id`, transcripts use `agentId`, and we need not care which lands.
agent=$(printf '%s' "$input" | jq -r '.agent_id // .agentId // empty')I fucking _loathe_ Trump. I still hope he makes America better. But god I fucking hate him. I think his policies are garbage, but I'd love to be proven wrong. All I see is bribery and nepotism... but yknow, he (finally) nominated Jared Isaacman. I'll fucking take it.
Musk is anathema. But yknow, I'm glad he's started the companies he has.
> It's theoretically possible for a bad actor to embed hidden adversarial behavior in a model. But if this happens, it serves the interests of responsible actors to find these exploits as soon as possible, and the best way to do this is to let anyone who wants to inspect them.
This is a bad argument. It isn't trivial to tell if the weights have poisoned:
``` https://www.thedeepview.com/articles/microsoft-how-to-spot-a... https://futurism.com/future-society/easy-poison-open-weight-... https://semgrep.dev/blog/2026/ai-supply-chain-problem/ ```
I'm not arguing in favor of closed-AI; I'm simply saying poisoning may be subtle.
I was gonna say that at least the frontier labs may be motivated to not-poison their own models, but then Anthropic just attempted to poison Fable's LLM training capability so... sigh. I'll try not to derail.
I'm off to Codex land I guess, which is sad because I do appreciate Anthropic's spine when they told off the DoW.
I'm just saying there's a predictable result when you express it with the level of detail and amount of effort that you _did_. And frankly, your comments are no better than pre-reasoning era LLM rage-bait.
What level of engagement are you looking for here; support for your lack of citations, or "yeah that's also my personal experience rah rah"?
I'm having a meta-level discussion, if you can't tell. I'm not "putting words in your mouth", I'm trying to discuss: the quality of your discussion. I'm discussing the quality of your arguments and your evidence. If you think that "racial profiling" is too hot, substitute in something else; that's not the point.
Everyone has an annecdote. "it's common enough to be recognized as a trend" is equally justification for racial profiling, and at least racial crime statistics are easily citable. And you still haven't even put forth that modicum of effort.
Has anyone found success with memory and Claude Code? I found Opus's 4.8 memories largely lacking value. Memories are too verbose/specific/yet-generic for the value captured.
Instead, I've been holding "retros" with the agent immediately after a session, and _those_ responses have been typically fantastic. It has ideas for coding changes, spots unmentioned small bugs, suggests invariants, principles to adopt, lint rules, tooling tweaks, skill-file updates, follow up work, all kinds of stuff.
Gonna take a moment to share my own generic "retro" prompt, which has found many areas of improvement IME.
> Let's conclude with a retro. Did you run into any issues during this session that you think could be improved? Any failed tool calls, confusing docs/prompts, or tricky wording that took you effort to figure out, etc? Any final thoughts that you want to raise? Anything minor you didn't mention? Help make this codebase easier for the next agent to work in.
It's somewhat doc-focused since I'm currently working on fairly dense design docs... but you can easily customize it for your own needs.
This prompt reveals how absolutely _ass_ the Claude Code harness is (so many stupid tool call failures), but not much I can do about that.
1st video is ~1minute long and puts you in the right mindset.
Do not anthropomorphize the lawn mower. It will cut off your foot, given the chance.