mh-

21 ms·

451.
by mh- 1y ago
Agreed, this is an instant turn-off for me when I realize this in e.g. an RTS game. Red Alert or C&C come to mind on higher difficulty, can't remember which.
452.
by mh- 1y ago
As you identified, not paying for it is a big part of the issue. Running these things is expensive , and they're just not serving the same experience to non-paying users. One could argue this is a bad idea on their part, letting peopl
453.
by mh- 1y ago
I prototyped this a couple months ago using OpenAI APIs with structured output. I had it consume a "deep thought" style output (where it provides inline citations with claims), and then convert that to a series of assertions and a
454.
by mh- 1y ago
GoodRx is genuinely a good deal. There's a paid offering (called Gold, I think?) that makes things cheaper, but whether it pays for itself will depend on what you're filling.
455.
by mh- 1y ago
Yes, but don't miss what I said about the other setting. You can't see what it's using from past conversations, and if you had one or two flippant conversations with it at some point, it can decide to start speaking that wa
456.
by mh- 1y ago
Turn off the setting that lets it reference chat history; it's under Personalization. Also take a peek at what's in Memories (which is separate from the above); consider cleaning it up or disabling entirely.
457.
by mh- 1y ago
> Makes me wonder if people spoke this way about “using computers” or “using the internet” in the olden days. Older person here: they absolutely did, all over the place in the early 90s. I remember people decrying projects that moved t
458.
by mh- 1y ago
I hope everyone realizes these comments just discourage companies from letting their employees do their work in the open, in collaboration with the community. If you want them to wait until everything is super ready (or dead) and then &quot
459.
by mh- 1y ago
It clearly (to me) shows that it's providing a docx for background knowledge, the way all of the existing providers allow you to upload a file as part of creating a new prompt. It's in a <document> tag, under a <hint>
460.
by mh- 1y ago
Hmm, yeah, that's a great suggestion, thanks!
461.
by mh- 1y ago
I was an infra guy early in my career, and I'm still savvy, and I still prefer using Tailscale. It's very polished and reliable. But personally, I'm past the point of wanting to fiddle with things like this and would much pre
462.
by mh- 1y ago
Yeah, I have mixed feelings about CT (certificate transparency) for this reason. Folks are just consuming the firehose and scanning. And in this case, if the thing you're funnel'ing is on your residential connection, it basically
463.
by mh- 1y ago
Oh this is brilliant . Why haven't I thought of this? I travel a lot for work and always take a pic of my parking space number. A few weeks back I forgot to, realized I forgot before I got in the security line and was like.. nah, I wo
464.
by mh- 1y ago
Yes. And batched inference is a thing, where intelligent grouping/bin packing and routing of requests happens. I expect a good amount of "secret sauce" is at this layer. Here's an entry-level link I found quickly on Goog
465.
by mh- 1y ago
> prompts in alt First pic (blonde woman with eyes closed) has alt text that begins: > Extreme close-up portrait of a black man’s face with his eyes closed copypasta mistake or bad prompt adherence? haha.
466.
by mh- 1y ago
o/t: your astrophotography LoRA is very cool, I came across it before. thanks for making it! (for others: https://civitai.com/models/890536/nasa-astrophotography-or-f... )
467.
by mh- 1y ago
Yeah, big +1 on this tip. I have AirTags on my bags themselves as well as some other things. Don't have them on my key fob, but you may have inspired me to attach one haha. The map thing when you're nearby and it goes into the son
468.
by mh- 1y ago
I'd expect you to have more control over it, however.
469.
by mh- 1y ago
Haha, I heard this a lot growing up. And now I have kids of my own..
470.
by mh- 1y ago
Thanks for that pointer. The abstract of this paper seems interesting: https://arxiv.org/html/2407.01085v3 > use of [LLMs] as judges [..] reveals a notable bias towards longer responses, undermining the reliability
471.
by mh- 1y ago
Lots, yes. The fine tuning may attempt to introduce concepts that were intentionally omitted from the training data for safety* reasons. Maybe nothing wrong with that, but it might mean that the perceived weaknesses don't generalize to
472.
by mh- 1y ago
I haven't tried your exact task, of course, but I've found a lot of success in using JSON structured output (in strict mode), and decomposing the response into more fields than you would otherwise think useful. And making those fi
473.
by mh- 1y ago
This is totally something I'd do. I'm very organized when I travel for work and everything has a place. If I absentmindedly slip something into the wrong part of my bag, it might as well be invisible..
474.
by mh- 1y ago
To be clear, Claude doesn't provide that either. You can get "usage limited" off of Opus on the $200/mo plan.
475.
by mh- 1y ago
> unsure how best to overlap aliases and configs that are sensitive to my workplace I have a .zshrc that sources .zshrc_mh, .zshrc_$employer, etc. That way my .zshrc is always a shareable config of sane defaults, and weird/opinion
476.
by mh- 1y ago
> I have 9 YOE now so I've done the usual gauntlet hundreds of times by now. I'm not sure how to parse this.. you've done hundreds of interviews, as the candidate, in 9 years of experience?
477.
by mh- 1y ago
> They serve the full article to Google, which allows Google to show you excerpts that you have to pay for. I'm old enough to remember when that was grounds for getting your site removed from Google results - "cloaking"
478.
by mh- 1y ago
There's nothing requiring that image to be unique. Lots of sites just provide a higher solution favicon.
479.
by mh- 1y ago
Thanks.
480.
by mh- 1y ago
Why not wait to release them until enough data has come in that it's settled? Serious question, what's the downside?

More ›