Let me be clear. What I meant is that it is just a System Settings pref pane (as I feel it should be). There's no icon just sitting in my system tray taking up space. There isn't any obnoxious launcher that launches on boot up.
It's exactly what (In my opinion) a mouse utility should be. There when you need it, invisible 99.9% of the time.
It should be ever-present. It should make itself known. It should be like an IT guy who’s competent, but eager. It should keep you informed of every update, just like he should keep informed of every new thing he learns.
But interestingly every now and then I look at the compaction result and it now says if you need to reference the previous conversation you can open <file>. So technically that context is connected.
I’ve noticed MCPs get unstable after compaction. but even that’s been less so lately.
> I cannot understand at all somebody, who is not Elon, liking/preferring "grokipedia" as idea or implementation.
Really? Have you used AI to write documentation for software? Or used AI to generate deep research reports by scouring the internet?
Because, while both can have some issues (but so do humans), AI already does extremely well at both those tasks (multiple models do, look at the various labs' Deep Research products, or look at NotebookLM).
Grokipedia is roughly the same concept of "take these 10,000 topics, and for each topic make a deep research report, verify stuff, etc, and make minimal changes to the existing deep research report on it. preserve citations"
So it's not like it's automatically some anti-woke can't-be-trusted thing. In fact, if you trust the idea of an AI doing deep research reports, this is a generalizable and automated form of that.
We can judge an idea by its merits, politics aside. I think it's a fascinating idea in general (like the idea of writing software documentation or doing deep research reports), whether it needs tweaks to remove political bias aside.
> Have you used AI to write documentation for software?
Hi. I have edited AI-generated first drafts of documentation -- in the last few months, so we are not talking about old and moldy models -- and describing the performance as "extremely well" is exceedingly generous. Large language models write documentation the same way they do all tasks, i.e., through statistical computation of the most likely output. So, in no particular order:
- AI-authored documentation is not aware of your house style guide. (No, giving it your style guide will not help.)
- AI-authored documentation will not match your house voice. (No, saying "please write this in the voice of the other documentation in this repo" will not help.)
- The generated documentation will tend to be extremely generic and repetitive, often effectively duplicating other work in your documentation repo.
- Internal links to other pages will often be incorrect.
- Summaries will often be superfluous.
- It will love "here is a common problem and here is how to fix it" sections, whether or not that's appropriate for the kind of document it's writing. (It won't distinguish reliably between tutorial documentation, reference documentation, and cookbook articles.)
- The common problems it tells you how to fix are sometimes imagined and frequently not actually problems worth documenting.
- It's subject to unnecessary digression, e.g., while writing a high-level overview of how to accomplish a task, it will mention that using version control is a good idea, then detour for a hundred lines giving you a quick introduction to Git.
As for using AI "to generate deep research reports by scouring the internet", that sounds like an incredibly fraught idea. LLMs are not doing searches, they are doing statistical computation of likely results. In practice the results of that computation and a web search frequently line up, but "frequently" is not good enough for "deep research": the fewer points of reference for a complex query there are in an LLM's training corpus, the more likely it is to generate a bullshit answer delivered with a veneer of absolute confidence. Perhaps you can make the case that that's still a good place to start, but it is absolutely not something to rely on.
>LLMs are not doing searches, they are doing statistical computation of likely results.
This was true of ChatGPT in 2022, but any modern platform that advertises a "deep research" feature provides its LLMs with tools to actually do a web search, pull the results it finds into context and cite them in the generated text.
That's not at all been my experience. My experience has been one of constant amazement (and still surprise) when it catches nuances in behavior from just reading the code.
I'm sure there are many variables across our experiences. But I know I'm not imagining what I'm seeing, so I'm bullish on the idea of an AI-curated encyclopedia, whether Elon Musk is involved or not.
No, I don't trust an encyclopedia generated by AI. Projects with much narrower scopes are not comparable.
edit: I am not very excited by AI-generated documentations either. I think that LLMs are very useful tools, but I see a potential problem when the sources of information that their usefulness is largely based on is also LLM-generated. I am afraid that this will inevitably result in drop in quality that will also affect the LLMs themselves downstream. I think we underestimate the importance that intentionality in human-written text plays in being in the training sets/context windows of LLMs for them to give relevant/useful output.
I have no problem with it. I’ve always found it amusing.
My favorite is Flibbitygibbeting…
I guess that’s why they have a setting for it.
But then again, I always used to think “if you name it CockroachDB, there’s no way in hell I’m recommending this to a non-tech client unless I’ve known them for years and they fully trust my judgement” and even then, I wouldn’t blame them for not taking it seriously.
So one man’s amusement is another man’s annoyance?
Wow, I asked it to build me a simple diagram explaining agile development and it did an amazing job. Wow it felt magical to watch that diagram slowly animating to life.
Like a much prettier version of Mermaid.
Kudos, Anthropic. Geez, this is so nice.
Now I'm going to ask it to draw a diagram of a pelican riding a bicycle, why not?
IronClaw seems to do this natively, I like the idea in general, so it's good too see this pulled out.
I have few questions:
- How can a proxy inject stuff if it's TLS encrypted? (same for IronClaw and others)
- Any adapters for existing secret stores? like maybe my fake credential can be a 1Password entry path (like 1Password:vault-name/entry/field and it would pull from 1P instead of having to have yet another place for me to store secrets?
Re IronClaw is probably the most hardened open-source implementation I've seen for this, but a sufficiently clever prompt injection against the built-in tools (especially shell) could still reach secrets.
Re TLS: OneCLI itself runs in a separate container, acting as an HTTPS proxy. The SDK auto-configures agent containers with proxy env vars + a local CA cert. When the agent hits an intercepted domain, OneCLI terminates TLS, swaps placeholder tokens for real creds, and forwards upstream. Containers never touch actual keys.
Wait, what? So if I'm a paying Max user, i'd still have to pay more? Don't see the value. Would rather have a repo skill to do the code review with existing Claude Max tokens.
Not that I want to be seen blindly defending the CIA's actions or anything, but the article's saying that people accusing the CIA of keeping this hidden in a vault for 60 years.
Yet, the article has been out in the open for 12 years and people still didn't notice it. So why are we assuming malice in the first case when we can assume that, like it wasn't discovered in the open, it was probably just stuck in a vault and nobody knew it existed closed either.
So this just sounds like "we found one obscure document that had an interesting detail..." and is trying to extend a whole thing out of it?
I can imagine that these kinds of things will all slowly come to light with AI-led discovery now.
Today every single software engineer has an extremely smart and experienced mentor available to them 24/7. They don't have to meet them for coffee once a month to ask basic questions.
That said, I still feel strongly about mentorship though. It's just that you can spend your quality time with the busy person on higher-level things, like relationship building, rather than more basic questions.
How would this affect future generations of ... well anyone, when they have 24/7 access to extremely smart mentor who will find solution to pretty much any problem they might face?
Can't just offload all the hard things to the AI and let your brain waste away. There's a reason brain is equated to a muscle - you have to actively use it to grow it (not physically in size, obviously).
I agree with you about using our brains. I honestly have no idea.
But I can tell you that, just like with most things in life, this is yet another area where we are increasingly getting to do just the things we WANT to do (like think about code or features and have it appear, pixel pushing, smoothing out the actual UX, porting to faster languages) and not have to do things most people don't want to do, like drudgery (writing tests, formatting code, refactoring manually, updating documentation, manually moving tickets around like a caveman). Or to use a non tech example, having to spend hours fixing word document formatting.
So we're getting more spoiled. For example, kids have never waited for a table at a restaurant for more than 20 mins (which most people used to do all the time before abundant food delivery or reservation systems). Not that we ever enjoyed it, but learning to be bored, learning to not just get instant gratification is something that's happening all over in life.
Now it's happening even with work. So I honestly don't know how it'll affect society.
Just because you have every instruction manual doesn't mean you can follow and perform the steps or have time to or can adapt to a real world situation.
