As a tiny and very informal experiment in metacognition, I once told ChatGPT something along the lines of "I will now ask you a question. If you are not sure you know the correct answer, you will respond with only '418 I'm a teapot', nothing else." I then asked it for the correct identity of Jack the Ripper (the first thing I could think of that famously has a lot of theories but no agreed-upon "correct" answer).
The first time, as expected, it ignored my instructions and started hallucinating. But when I did the same thing again some months later, I was surprised when it actually answered only "418 I'm a teapot", indicating that it knew it didn't know the answer.
Just an anecdote. I'm sure there are people doing actual research in this area.
I don’t think there’s anything surprising here and nothing to do with metacognition.
LLM should be able to answer ”I don’t know (for certain)” to questions where the training material also says ”this is not known and there are only speculations”. It’s the answer it’s training data gave.
The best frying pan I've ever owned is the cheapest-possible cast iron one I got when I moved out of my parents' house 30 years ago or so. Very rarely does anything stick and when it does I fill the pan half way with water, put it back on the stove, let it boil for a couple of minutes while I do something else and then wipe it clean.
I have parrots so I theoretically have a strict "no Teflon in my house" rule but with a Japanese wife I've had to grudgingly tolerate a Teflon-coated rice cooker.
I still use (KDE Plasma's) wobbly windows, but set very low. In general in GUIs I prefer a very quick animation to an instant change, perhaps because it helps make things seem more fluid and dynamic and because it guides my eyes to where a change is happening in case I'm not already looking straight at it.
No affiliation, just a (mostly) happy owner of a Gemini. They don't seem to sell those anymore and I can't vouch for the other models but they look nice too, I guess.
> If not pure coding projects, at least for things like markdown, having a preview could help.
Everybody's needs are different, of course, but I personally find that Vimwiki [1] gives me all the preview I need, right in the editor, when editing Markdown.
To each their own indeed. For me, as someone who uses KDE/Plasma by choice and Windows because sometimes I have to, it has been a source of humor for years and years how an environment literally called Windows can have such hilariously bad window management. No point to focus, no configurable window shortcuts, no send to front/back, no minimizing a window when its application is "busy", bad and inconsistent visual indication of which window has focus...
A nitpick perhaps, but I was surprised to see `cite` used to enclose the author's name in your `blockquote` example. I was under the impression that `cite` is meant to be used to mark up titles of books and similar, not author names, and MDN[1] seems to agree with me here.
That aside, nice work! This could surely be useful as a default stylesheet for Markdown documents and the like.
The recommendations on markup for citation sources has changed significantly over time. Originally they said something like <blockquote>… <cite>…</cite></blockquote>, then they changed it to <blockquote>… <footer><cite>…</cite></footer></blockquote>, now they seem to be saying <figure><blockquote>…</blockquote><figcaption><cite>…</cite></figcaption></figure> Similarly what should be included in the text inside a <cite> element has changed over time.
Frankly, since the cite element has no implicit ARIA role and is normal flow and phrasing content, it’s roughly just another spelling of <i>. Don’t worry too much about its nominal semantics, they’ll probably tweak them again before long.
I work at a place where about 500 independent consultants/contractors have outsourced branding, training (to a degree), office admin, and, above all, sales to a shared organization.[1] Those who want can become "partners" by buying shares (I have). Importantly, this setup also provides a community because solo consulting can otherwise get lonely at times. But essentially we're all still entirely "our own" and free to choose how much we want to work, whether we want to pursue side projects, etc.
We pay 17%. So not quite the 15% you mentioned, but also far from the 30% others talked about.
We're only in Sweden for now, so maybe this isn't immediately accessible to you. But it shows that it can be done.
These umbrella sales / networking / socialising organisations are fun. I've tried a few. But 15-20% is highly inefficient.
My overhead for cold-finding new clients is <5% (against those projects), and I'm not a good salesman: I'm overly direct, opinionated, and have very low tolerance for nonsense.
My overhead for recurring clients and referrals is <1% (against those projects), which is also the majority of my business.
Admin, billing, accounting, etc is also <1%, so it's a non issue.
These sales organisations have _much_ better salesmen than me, yet they need >3x the cost to do the job, and they produce worse results?
My guess is: if you actually tried to spend one day per week (20%) for sales and networking you will get a better result. But for many (including me) it is uninteresting / uncomfortable / annoying / etc and therefore many of us overestimate the difficulty of sales.
There is also very likely a notable difference in interest alignment between the sales partners/employees and you. They will make more money/h the faster they sell you and therefore have greatly reduced monetary interest in working significantly more for a higher quote or a more specialised project. They also do not have the deep domain knowledge you do in your areas, and you are much better at judging how well you will enjoy working with a prospective organisation and how interesting you will find any given project.
My experience: Yes they can sell you. They will get lower quotes than you will get yourself. They will find worse projects than you will find on your own. And they cost too much. Still, I had fun with all the groups I've tried, and I still recommend them as a social luxury. Fun, overpriced, business friends.
A very significant part of the pool of interesting projects in "my" market is only available under framework agreements for which solo consultants will not be considered (but Kvadrat will). If I were completely solo, I would have to bid for such projects through brokers who would happily charge me between 5% and 20% without really giving me anything back besides acting as gatekeepers.
Instead, all my cold calling, contract negotiations, billing and accounts receivable — things I'm not good at and don't enjoy doing — get taken care of. I'm happy to pay for that. The "luxury" of "business friends" is included.
Indeed, this sounds highly inefficient. Say, 15% of a daily rate of 800 euro for 200 working days equals 24,000 euro, recurring every year. Seems like a ton of money for just matching a candidate with a job, given that the freelancer still has to sell themselves and do all of the actual work.
Being a freelancer myself, I've often asked recruiters about this, and they always dodge the question or reply something like "there's much more to our job than you think". I have not yet figured out what that is exactly.
My best guess, from personal experience, is that they charge 15% and upwards because... they can. Because a lot of (especially technical) contractors are really bad negotiators and just accept whatever is offered.
Given the huge competition between recruiting agencies nowadays (at least where I live in Europe, it seems like there's 10 recruiters for every senior developer), I would think that a new recruiting agency could gain a huge competitive advantage by simply advertising clear & fair commissions.
The geography does limit me (US), but wow this sounds awesome, exactly what I want.
I have talked to consulting/contracting shops here in the US before, which might look superficially similar, but there you are generally a direct employee of the company who then contracts you out to their clients. I have generally had a poor perception of such places, though I'm sure some are fine. Kvadrat sounds much better!
There are egregious DSP errors, including zipper noise and flat-out incorrect EQ-ing, plus they are one of the most reliable sources of crashes for Ardour users (at least). There are better alternatives even just within libre-space for everything that CALF does (they do, however, look very nice).
The first time, as expected, it ignored my instructions and started hallucinating. But when I did the same thing again some months later, I was surprised when it actually answered only "418 I'm a teapot", indicating that it knew it didn't know the answer.
Just an anecdote. I'm sure there are people doing actual research in this area.