mh-

4 ms·

1.
by mh- 5d ago
I think the people who aren't greeted by this dichotomy when turning their PC aren't catching the reference. At the risk of explaining a joke: Windows tries to get you to set up things like OneDrive on startup with a "do it n
2.
by mh- 5d ago
To my understanding, with batched inference and other "optimizations" you wouldn't get the exact same token predictions even with temp=0.0.
3.
by mh- 5d ago
Every official statement I've seen around this is careful to say that they "don't intentionally reduce model quality", which leaves plenty of room for "we adjusted some knobs and our evals show performance is ma
4.
by mh- 6d ago
I don't know how to respond to that, because it has quite a few negative effects on them..
5.
by mh- 6d ago
As someone who was annoyed by it at the time, I don't think it's a good example of that. This was in an era of much more limited upstream bandwidth. The p2p nature was obscured from non-technical users in some implementations. S
6.
by mh- 7d ago
It was a bit challenging long before that. I ran webhosting on FreeBSD in the early 2000s - remember finding a tool you needed and hoping it was in ports?
7.
by mh- 7d ago
I'm even more confused now. I guess this was an oblique way of accusing people of astroturfing?
8.
by mh- 7d ago
> I don't think I have to speculate why comments like yours get downvoted so much. I'm genuinely not sure what you're implying here. Can you expand?
9.
by mh- 7d ago
I thought it was an interesting comment and I suffer from the same issue when trying to get a clear idea from my head into writing. I sacrificed some Pangram credits to reply to the parent comment, fwiw.
10.
by mh- 7d ago
I can see why you might think that, but Pangram says otherwise. https://www.pangram.com/history/3d55b442-d151-4490-9c71-28d0...
11.
by mh- 9d ago
I read your OP and looked at the site (including the "peek at the chat" in the top right). But until the (grand)parent comment, I thought the intent was for the nearly-out-of-tokens agents to talk to one another. I don't ha
12.
by mh- 13d ago
That's such a great idea. Thanks for doing it!
13.
by mh- 16d ago
On the other hand, disabling WebGPU is offering entropy for fingerprinting, too. A vanishingly small % of users will have done so.
14.
by mh- 22d ago
Not sure, but I intentionally included the opening paragraph because I spotted the writing style switch when I got to the second paragraph. So then ran it on Pangram to confirm. edit: to be clear, if you click the details, the "70%&quo
15.
by mh- 22d ago
I don't usually comment about LLM-written articles, but it feels like an especially bad idea to do so in an article where you promote a book you're writing. Will that be LLM output, too? https://www.pangram.com/his
16.
by mh- 24d ago
This is interesting, is there a version with shorter responses? I spent a few minutes reading about SynthID-Text [0] and couldn't find mention of this (I've not read the paper yet), but my intuition [1] is that encoding more bits
17.
by mh- 25d ago
To be fair, parent commenter said the operator , not the jammer.
18.
by mh- 25d ago
Why not use the even smaller speakers that you put in your ears, and then you can hear it. And it's inaudible to every other human. FWIW, both Spotify (Jam) and Apple Music (SharePlay) have a feature to sync audio playback and give b
19.
by mh- 1mo ago
No. It's as you say here, as well.
20.
by mh- 1mo ago
I'm the CTO of a GCP shop with an 8 figure annual commit. If you'd told me at the end of Cloud Next 2025 that by now Google still wouldn't have a competitive offering to agentic coding offerings from Anthropic (Claude Code
21.
by mh- 1mo ago
Same for us, I'm not even sure what product that other "Enterprise" status page refers to..
22.
by mh- 1mo ago
When I worked at a building in SF (where we leased, exactly like this situation) and SEIU demonstrated in front of the building, we were told to consider staying home just because we'd have to walk through the picket line. It wasn&#x27
23.
by mh- 1mo ago
There is some LLM detection for comments that auto-flags them. It has some unfortunate side effects, though. Namely on a thread where people are discussing a text-heavy post that is LLM-authored, quoting a large enough block of that content
24.
by mh- 1mo ago
Just the trademark, that I've seen.
25.
by mh- 1mo ago
web.net, at least, absolutely is an email provider.
26.
by mh- 1mo ago
They do not throw away the servers every year..
27.
by mh- 1mo ago
If parent commenter is off by an order of magnitude in their costs, their point still stands.
28.
by mh- 1mo ago
Worth mentioning GitLab's paid enterprise offering are more expensive than GitHub's, on a per-seat basis. Lots of companies moved because it was cheap, but it's not anymore. Ironic that companies might choose to migrate to th
29.
by mh- 1mo ago
People forget this far too often.
30.
by mh- 1mo ago
+1 all of this. That would be incredible (this already seems very cool - excited to get home and try it!)

More ›