The small town of Festus, Missouri
— a community made up of just under 14,000 people
— has become a focal point over the growing backlash against
AI data centers entering communities across the country.
Following the approval of a $6 billion data center project,
voters removed four of the eight members of the Festus city council,
as well as started a petition to remove the remaining city council members and the mayor, reports Politico.
Gloom & Bloom IV ☁️🌺
黑暗绽放 IV ☁️ 🌺
📷 Zeiss IKON Super Ikonta 533/16
🎞️ Lucky SHD 400
#filmphotography #Photography #blackandwhite
#silentSunday
Sprague Lake, #RockyMountainNationalPark
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.
Oil prices have climbed again amid mounting supply fears after the US struck Iran’s vital Kharg Island oil hub and Donald Trump demanded allies help reopen the strait of Hormuz.
Brent crude, the international benchmark, rose 1.8% to $104.98 per barrel during early trading on Monday.
Another weekend of violence across the Middle East compounded concerns over the conflict, and its ramifications for global energy markets.
Trump claimed on Saturday that US strikes had “totally …
“Originally launched as a casual meet-up in a cafe, the club has flourished into a network of offline events across Europe.
“Facilitators now organise sessions in 18 cities, and the club has built a following of around 600,000 on Instagram.”
'A simpler way of being': The rise of the offline club ht…
Lawmakers on Capitol Hill are accusing the Justice Department of
covering up the names of co-conspirators of the deceased sex offender Jeffrey Epstein
as fallout from the Epstein files grows across the globe.
Millions of pages remain unreleased.
As many prominent U.S. figures evade accountability following mentions in the Epstein files,
a number of European figures have resigned for their relationships with Epstein.
“The most extraordinary and worrying …
Human Traces VI ⛩️
人类踪迹 VI ⛩️
📷 Pentax MX
🎞️ Harman Red 125 (FF)
If you like my work, Support by buying me a coffee or a roll of film from PayPal https://www.paypal.com/paypalme/ydcdingsite
Just as they’re gearing up for planting season, U.S. farmers already stretched by high input costs and low commodity prices are watching the price of fertilizer go up.
The recent conflict in Iran, the following closure of the Strait of Hormuz, and the resulting impacts to global markets are hitting farmers particularly hard right now.
The wholesale price of urea, the nitrogen input the U.S. imports the most of, had a high-low spread of $460–480 per short ton the week of Feb. 27, j…