I tried tmux so many times, could not commit the sequences to memory, but then zellij was just out of the gate, easy to "discover," and then I started writing plugins (rust wasm), and I even submitted a PR which got accepted to support background colors in panes/tabs.
I am a monthly donor, I think it has the right balance of community plus the lead dev has a vision, opinionated but open to inputs, and focused.
Interesting, I just did this on some stuff I have personally handwritten, but the writing is about LLMs, and it says 25% AI generated, but because I used terms that are "x% likely AI" generated.
This is not a very useful test; it basically means a person has to "ban" terms like "user experience" and "tireless" etc. because these are "Nx more likely to appear in generated content".
I think it's a bit sad that we often say people "Sold out". Sometimes, I agree, but often, I point out that until the lady at the grocery store stops asking me for money when I walk up to check out, I need to pay my bills to eat.
I contribute to open source projects, but none of them to date could support me buying much more than a beer. If one took off such that I could "live" off it, I would be happy to leave my current job and dive all in. Until then, I just keep plodding along.
I put together a quick audit to check for "early landing" messages[1] using jq, ripgrep, and the messages[2] flagged in the stop guard script.
I have noticed a trend in these sessions asking more and more about calling it a day, "it's getting late," and other phrases. I sort of assumed it was some kind of "load shedding" on Anthropic's side.
My audit of 80 sessions was interesting. Sorry, I won't share details, but I recommend you do the same.
As a negative example, my audit of 31 sessions was uninteresting. I had one matching entry, where I had pasted a long list of console errors into Claude and it identified a few as pre-existing and asked me to get more information for follow-up analysis.
I wonder if it comes down to prompting—maybe by introducing these "golden rules" OP mentions in their CLAUDE.md, they're actually "priming" Claude to think about these stop phrases and introduce them proactively.
Do you have a CLAUDE.md file? What does it contain?
Those load shedding statements are infuriating. I’ve literal had sessions where we just get through planning a giant feature and I say “get started” with the response being “okay, we’ll pic up tomorrow “
Fascinating, I thought I was losing my mind. Claude CLI has been telling me I should go to bed, or that it's late, let's call it here, etc, and then I look at the stop-phrase-guard.sh [1] and I'm seeing quite a few of these. I thought it was because I accidentally allowed Claude to know my deadline, and it started spitting out all sorts of things like "we only have N days left, let's put this aside for now," etc.
I just saw it this weekend; "It is quite late and we have accomplished a lot. Get some rest and we can pick it up later". Not bad advice but then not it's place. Also trying to steer me away from a tough issue towards a low hanging fruit.
I wonder if its being trained on the human replies to the model, I sometimes write stuff like that back to Claude after I want to finish for the day myself.
My speculation on this has been that it's potentially a factor against ai psychosis, as psychosis risk (of any psychosis) is significantly elevated with lack of sleep. If you read case studies of ai psychosis, many of them also involve people staying up way too long right before they fall on a bad path.
The "calculator ruined the world" argument was actually studied to death once the panic subsided. Large meta-analyses of 50 years of data show it was mostly a non-problem. Students using calculators generally developed better attitudes toward math and attempted more complex problems because the mechanical drudgery was gone.
The only real "catch" researchers found was timing. If you introduce them before a kid has "automaticity" (around 4th grade), they never develop a baseline number sense, which makes high-level math harder later on.
It's a pretty clean parallel for LLMs. The tool isn't the problem, but using it to bypass the "becoming" phase of a skill usually backfires. If you use an LLM before you know how to structure an argument or a block of code, you're just building on sand.
You're using the motte-and-bailey fallacy. Pocket and scientific calculators have never caused anywhere near as much grief as current LLMs do as a loss-leader for near-trillion dollar corporations that are primarily responsible for artificially keeping the US economy afloat (at least prior to the latest war/not-war effort).
Peak irony: OpenAI's Codex project is being DDoS'd by its own creation.
They made code generation essentially free, and now they're drowning in AI-generated PRs that take more time to review than it would take to just write the fix themselves. So they've moved to invitation-only contributions. They built the tools that removed code as a barrier, and now they're living with what that actually means: quantity goes up, quality goes down, and every PR still needs a human with architectural context to evaluate it.
The bottleneck has completely flipped. It used to be finding someone to write the code. Now anyone can dump a 200-line PR in seconds, but the review time from someone who actually understands the codebase is the most expensive thing in the repo. That context can't be generated, it has to be earned.
They built a firehose and their own team can't drink from it.
Some users are moving to local models, I think, because they want to avoid the agent's cost, or they think it'll be more secure (not). The mac mini has unified memory and can dynamically allocate memory to the GPU by stealing from the general RAM pool so you can run large local LLMs without buying a massive (and expensive) GPU.
I think any of the decent open models that would be useful for this claw frency require way more ram than any Mac Mini you can possibly configure.
The whole point of the Mini is that the agent can interact with all your Apple services like reminders, iMessage, iCloud. If you don’t need any just use whatever you already have or get a cheap VPS for example.
If the idea is to have a few claws instances running non stop and scrapping every bit of the web, emails, etc, it would probably cost quite a lot of money.
But if still feels safer to not have openAI access all my emails directly no?
I use Terminus with Zellij and keep about 8 sessions going with a combination of Claude and Codex, and once in a while, Gemini. It's great when you're sitting in a docotor's office lobby bored out of your skull and when you get back to your desk you just join the session and it's all right there.
Yeah, I could make a similar post for similar reasons. I already have a bunch of mini-PCs I collect like Raspberry Pis. I already have a mITX build with Bazzite installed on it that I would use over a Steam Machine because it's faster. And like OP, I'd probably get anyways. Assuming price is ~800$ (with controller)
I am a monthly donor, I think it has the right balance of community plus the lead dev has a vision, opinionated but open to inputs, and focused.
reply