@Techmeme@techhub.social2026-04-15 00:25:51
Mintlify, which uses AI to help companies generate software documentation, raised a $45M Series B led by a16z and Salesforce Ventures at a $500M valuation (Rashi Shrivastava/Forbes)
https://www.forbes.com/sites/the-prompt/20
@rafa_font@mastodon.online2026-03-15 13:36:37
Why are we having fewer children?
(Interview with Berkay Ozcan, Professor at LSE)
- Couple formation happens at later age
- Women are choosing "careers" and not just "jobs"
- More people choose not to have kids at all
What else is going on? Short anser: we don't know yet
Even in countries providing a lot of support to parents, fertility rate has still declined
Immigration is no silver bullet. It's part of the solution, not th…
@cdarwin@c.im2026-03-16 15:07:26
In another round of late-night social media posts, Donald Trump attacked the Supreme Court for not overturning his loss in the 2020 presidential election
— and tried to pressure the court to rule his way in future election cases.
Flurries of angry social media posts are not unusual for Trump,
but this one was particularly alarming for its scope and the criticism it directed against the nation’s highest court,
where six out of nine justices are conservatives and three …
@jtk@infosec.exchange2026-01-16 19:25:18
The title of the blog post blog was clickbait for me. It argues #Thinkpad fandom boils down to price, build quality, and repairability.
There is one glaringly obvious omission any longtime user should spot immediately. There is no mention of the unique trackpoint-based keyboard design that you can't find anywhere else.
I've stopped using ThinkPads as my primary platform f…
@krone@frawas.de2026-03-16 14:30:12
Angriff nach Videodreh - Seiler: Justiz sieht genug Verdacht für Verfahren #News #Nachrichten
@heiseonline@social.heise.de2026-04-09 14:45:00
@Simone21@mastodon.social2026-04-13 10:00:40
"Leo betonte, dass er keine direkte Attacke auf Trump oder andere Personen beabsichtigt habe, als er eine «Wahnvorstellung der Allmächtigkeit» kritisierte, die den Krieg gegen den Iran und andere Konflikte weltweit befeuere."
Aber wem der Schuh passt, der zieht ihn sich an. 🤭
Quelle mit Bezahlschranke:
@cdarwin@c.im2026-03-16 21:20:49
Donald Trump wondered if the U.S. “shouldn’t even be there”
when answering questions about the war on Iran on Sunday,
claiming that Iran’s military is already totally obliterated
as the U.S. and Israel’s bombardments enter their third week with no end in sight.
A reporter asked Trump on Sunday about his demands that other countries aid him in trying to force transit through the Strait of Hormuz while it’s closed by Iran.
In response, he said that other countries…
@hex@kolektiva.social2026-04-08 11:45:24
Willem challenged us to ask ourselves what we would do if we were living under Nazi occupation. Before all of this, I doubt anyone thought they would be complicit. I doubt anyone said to themselves, "nothing. I would cower in fear and do nothing."
But for 4 years or so we all answered that question again and again with our lives. Now here we are, answering it again... Every day. But it's no longer "what would you do during the rise of Hitler?" It's now, "what would you do after the invasion of Poland," and "what would you do after you knew about the concentration camps?"
For some people, the answer is still, "nothing."
But a lot of people have been brave in the face of it all. A lot of people have died, and a lot more will die. He will die, perhaps after a ruling by some court or other but, honestly, probably not. That's just how these things work out. Lots of people die, some for no reason, some because they stood up against injustice. A whole lot of people do nothing, until it's safe to claim victory... Until it's no longer safe to be on the other side.
That's just how these things go. Fascism is self-defeating, but it causes incredible harm on it's path of self-destruction. The more people who stand up, who risk themselves, the faster it collapses and the fewer it can hurt. That's also just how these things go. It's incredibly dangerous for everyone until enough people take some extra risk and make it safe for everyone again.
But that question still stands... Which one of those groups are you in? Are you proud of what you are doing, or will you look back with shame? Some of y'all have a lot to be proud of, but, if you're not, it's never too late to earn your way into that proud group.
@aredridel@kolektiva.social2026-04-14 14:22:42
So to follow up on this, I've caught it in action. Models, when quantized a bit, just do a bit more poorly with short contexts. Even going from f32 (as trained) to bf16 (as usually run) to q8 tends to do okay for "normal" context windows. And q4 you start feeling like "this model is a little stupid and gets stuck sometimes” (it is! It's just that it's still mostly careening about in the space of "plausible" most of the time. Not good guesswork, but still in the zone). With long contexts, the probability of parameters collapsing to zero are higher, so the more context the more likelihood you are to see brokenness.
And then at Q2 (2 bits per parameter) or Q1, the model falls apart completely. Parameters collapse to zero easily. You start seeing "all work and no play makes jack a dull boy” sorts of behavior, with intense and unscrutinized repetition, followed by a hard stop when it just stops working.
And quantization is a parameter that a model vendor can turn relatively easily. (they have to regenerate the model from the base with more quantization, but it's a data transformation on the order of running a terabyte through a straightforward and fast process, not like training).
If you have 1000 customers and enough equipment to handle the requests of 700, going from bf16 to q8 is a no-brainer. Suddenly you can handle the load and have a little spare capacity. They get worse results, probably pay the same per token (or they're on a subscription that hides the cost anyway so you are even freer to make trade-offs. There's a reason that subscription products are kinda poorly described.)
It's also possible for them to vary this across a day: use models during quieter periods? Maybe you get an instance running a bf16 quantization. If you use it during a high use period? You get a Q4 model.
Or intelligent routing is possible. No idea if anyone is doing this, but if they monitor what you send a bit, and you generally shoot for an expensive model for simple requests? They could totally substitute a highly quantized version of the model to answer the question.
There are •so many tricks• that can be pulled here. Some of them very reasonable to make, some of them treading into outright misleading or fraudulent, and it's weirdly hard to draw the line between them.