yorwba

42 ms·

841.
by yorwba 1y ago
It's mostly a consequence of the laws of physics having the Markov property. So the time evolution of any physical system can be modeled as a Markov process. Of course the corresponding state space may in general be infinite.
842.
by yorwba 1y ago
People who work at Alibaba (yes, the same company that released Qwen weights for free) told me that they were given access to Claude. (In a restricted environment that I presume is intended to prevent leaking too much of Alibaba's code
843.
by yorwba 1y ago
The argument isn't that anything is possible for LLMs, but that representing LLMs as Markov chains doesn't demonstrate a limitation, because the resulting Markov chain would be huge, much larger than the LLM, and anything that is
844.
by yorwba 1y ago
Did you ever manage to improve your classifier beyond what you described in that post or is getting past ≈0.75 AUC ROC simply too difficult.
845.
by yorwba 1y ago
From the article: > The state's interior ministry has pointed out that candidates from other parties, including the Greens and Social Democrats, have also died.
846.
by yorwba 1y ago
It's an off-the-shelf camera: https://www.viewproonline.com/products/a40-pro-40x-optical-z... It's hardly surprising that Russian drones use cameras made in China. For comparison, Ukrainian drone manufacturer
847.
by yorwba 1y ago
You might be able to start with German, since German-Yiddish cognates tend to have fairly regular spelling correspondences (not exactly one-to-one, but often few-to-one). So given a Latin-script token from a model that does OK in German (bo
848.
by yorwba 1y ago
I upgraded Harbor before, it was a pain. I think you're encouraged to use their official Helm chart and then it's supposed to be fairly seamless https://goharbor.io/docs/2.13.0/administration/upgrade
849.
by yorwba 1y ago
Previously: https://news.ycombinator.com/item?id=45057561 (88 points, 2 days ago, 20 comments)
850.
by yorwba 1y ago
Plenty of research involves small models trained on small amounts of data. You don't necessarily need to do an internet-scale training run to test a new model architecture, you can just compare it to other models of the same size train
851.
by yorwba 1y ago
The dashed brown line in figures 3, 4, and 6 is labeled GTE-ModernColBERT. So better than all single-vector models, but worse than BM25.
852.
by yorwba 1y ago
When you truncate Matryoshka embeddings, you get the storage benefits of low-dimensional vectors with the limited expressiveness of low-dimensional vectors. Usually, what people look for in sparse vectors is to combine the storage benefits
853.
by yorwba 1y ago
I'm not following your construction. In the k=2 case, how do you construct your 4-dimensional query vector so that the dot product is maximized for two different angles theta and phi, but lower for any other arbitrary angle?
854.
by yorwba 1y ago
I don't think Mixture of Logits from the paper you link circumvents the theoretical limitations pointed out here, since their dataset size mostly stays well below the limit. In the end they still rely on Maximum Inner Product Search, j
855.
by yorwba 1y ago
It's motivated reasoning in order to end up at Dunbar's Number: https://en.wikipedia.org/wiki/Dunbar's_number
856.
by yorwba 1y ago
The most annoying bit might be that they use different, though equivalent, definitions of the property, so you would also need to formalize the proof of the equivalence of definitions.
857.
by yorwba 1y ago
Also, frost will trigger the process, so nowadays you don't need to wait so long for the fruit to decide to be edible if you put it in the freezer for a bit first.
858.
by yorwba 1y ago
Lots of things sound like good ideas that aren't. China has fewer checks and balances, so lots of things get built that sound like good ideas. This impresses people! Especially when they hear the huge numbers involved. E.g. the author
859.
by yorwba 1y ago
> Many recent CPUs eliminate NOPs without executing them The elimination will still take some amount of time, and the smaller this amount is, the better, because it allows dialing in the slowdown more precisely. Of course how many NOPs y
860.
by yorwba 1y ago
It also applies to infitives and participles and the verb in nominalized noun-verb compounds. So the rule is closer to "the verb is at the end of its grammatical unit, except for the finite verb in a main clause, which appears in secon
861.
by yorwba 1y ago
It totally makes sense that people don't request things they don't expect to be possible.
862.
by yorwba 1y ago
The flaw isn't just in the design, it's in the requirements. People want an AI that reads text they didn't read and does the things the text says need to be done, because they don't want to do those things themselves.
863.
iDAI.field: software for the documentation of archaeological fieldwork(github.com)
1 points by yorwba 1y ago|1 comments
864.
by yorwba 1y ago
There's no need to tell the difference between those, because the CUII has neither given up nor have they said that they have given up. They'll continue to block the kinds of websites they've been blocking and add new ones to
865.
by yorwba 1y ago
That comparison doesn't make sense. FISC can only be a rubber stamp because only the US government can bring a case before the court. Any rando can sue German ISPs for not fulfilling their DNS blocking duties under copyright law, but c
866.
by yorwba 1y ago
Hopefully nobody told you that "this trial shows that substance X is not harmful" while pointing to this study, which is not a trial, but a cohort study making use of the fact that different birth years in Denmark received differe
867.
by yorwba 1y ago
Migrating might be more of an option than you think. If you can get a job offer in IT and have a university degree or significant work experience, EU countries typically make it relatively easy to get the corresponding work visa. E.g. infor
868.
by yorwba 1y ago
The title is "Vector of Change" and the description is "Your challenge is to make something as close as the expected shader." and that's all the context you get. It doesn't seem like a very well-designed exerci
869.
by yorwba 1y ago
Numbeo is entirely based on self-reports by the kinds of people who use Numbeo. For price comparisons you can at least argue that although the prices listed are much too high, it's a realistic representation of how much you're lik
870.
by yorwba 1y ago
Yes, and 31.3% is given in the announcement as the performance of the new v3.1, which would put it in sixteenth place.

More ›