eliotho

8 ms·

1.
by eliotho 4d ago
couldn't have said it any better
2.
Show HN: Throttling AI models under load can backfire and increase demand (SIM)(throttle.staffinganalytics.io)
2 points by eliotho 16d ago|0 comments
3.
Show HN: I built a tool showing how AI providers (should) throttle their models(throttle.staffinganalytics.io)
6 points by eliotho 20d ago|0 comments
4.
Show HN: I built a tool showing how AI providers (should) throttle their models(arxiv.org)
3 points by eliotho 22d ago|0 comments
5.
by eliotho 2mo ago
it came from the data (UN Comtrade) that notably doesn't include sanctioned oil flows (Iran, etc). On top of that, China's reserve levels are a state secret, yet, the model's framework is BYOD (Bring Your Own Data) which give
6.
by eliotho 2mo ago
Thanks! Two answers (repeated in other comments): UN Comtrade data doesn't include sanctioned oil (Iran), and nobody (except their government) knows China's actual stockpile. The model is more of a sensible baseline/stress te
7.
by eliotho 2mo ago
Thanks, added to list of possible UX improvements
8.
by eliotho 2mo ago
Yes, and the network setting is flexible enough to allow smaller supply chains where nodes don't have price setting power
9.
by eliotho 2mo ago
the models, proofs and writing are mine (with some AI assistance for proofreading/formatting). AI was mostly used in the visualization and the rerunning of multiple scenarios of the numerical examples
10.
by eliotho 2mo ago
This pipeline predates the data but its oil is routed to Bab-el-Mandeb which is in the network. That being said, the numbers are more of a sensible baseline as the actual flows for many countries are unknown/sanctioned.
11.
by eliotho 2mo ago
France imports 0 oil directly through Hormuz, but when the Strait closes, all other countries start raising their safety stock which increases oil prices. This makes France's stockouts very expensive mostly due to price (not flows). Wh
12.
by eliotho 2mo ago
this is an interesting market design question: the point is that tapering would require everyone knowing everyone else's depletion clocks. As they are unknown (sanctioned trade, non-public state figures), the opacity is the game itself
13.
by eliotho 2mo ago
This is an interesting insight, as usable reserves < reported reserves the clocks in the model are optimistic
14.
by eliotho 2mo ago
Fair catch, the data predates the reroute. Although this pipeline is routed to Bab-el-Mandeb which is in the network
15.
by eliotho 2mo ago
That's actually the intended behavior: while reserves are silently being spend is when the crisis is brewing. Although the main point is that nobody knows China's reserves, but this allows to calibrates scenarios based on what you
16.
by eliotho 2mo ago
Thank you! Much appreciated. On the suspense, we are all riding the same train :$
17.
by eliotho 2mo ago
And this is what the model price formation assumes, and in fact, the silent mechanism that makes the crisis worse. Reserves silently deplete for each country, and each epoch where they exhaust is when the price rebalancing occurs IN A SUDDE
18.
by eliotho 2mo ago
There are two versions of the model: -The one rendered by default (called Endogenous), with endogenous price that affects the demands according to the price elasticity slider. Reflecting how reactive are the nodes demand wrt price -With fix
19.
by eliotho 2mo ago
Thank you. Yes, in desktop there is one (that might still be going a little bit too fast), but in mobile it might not be that easy to find. Perhaps I will slow down the speed in mobile by default and change the location (should be in the bo
20.
by eliotho 2mo ago
This works as the same feature reversed. Will think of a way of adding something like this
21.
by eliotho 2mo ago
Fair point. Actually this is both a semantics oversight on my part and also expected behaviour. at 100% retained there's no shock, but the model's countries target consumption plus a large safety buffer, and some can't fill t
22.
by eliotho 2mo ago
Much appreciated, server seems to be (fingers crossed) going strong so far
23.
by eliotho 2mo ago
This is a nice idea. I will build this a feature on the scenarios menu. And it would give an estimate of how much time doing this buys China under different configurations
24.
by eliotho 2mo ago
Which is the interesting game-theoretic aspect of the whole conflict. China really hasn't revealed their hand
25.
by eliotho 2mo ago
Exactly, which is what makes the simulation interesting because nobody really knows China's reserves clock. But with the tool, you can put your guess number and see that the game of chicken also has an expiration date for them
26.
by eliotho 2mo ago
Thanks, the modeling is similar math to the financial banking networks paper by Eisenberg and Noe Systemic Risk in Financial Systems (used as a stress tool by regulators after the 2008 financial crisis). My adaptation is combining this with
27.
by eliotho 2mo ago
For the importer nodes those are IEA mandates. In the specific China case, the UN Comtrade data doesn't report sanctioned oil (Iran)
28.
by eliotho 2mo ago
and the interesting thing is that the common factor is that all these crises (oil, financial, gas) spread silently until a node collapses and there is a domino effect over the whole network
29.
by eliotho 2mo ago
That's kind of the point because it hasn't been totally 100% closed. There's both sanctioned and unsanctioned oil flowing, which is the point of the scenarios in the simulation. Also it's more of a simulation/stress
30.
by eliotho 2mo ago
>What concrete predictions does your model make? Well, the model is less of a prediction and more of a stress testing tool. But under the hypothetical closure scenarios it shows the timing the oil reserves of distant countries exhaust, a

More ›