I feel like the post doesn't relate to the title?
The post seems to be about changing the Leaderboard and doesn't comment too much about whether the actual real-life performance of LLMs is plateauing and what can be done about it.
2,559 karma · joined February 2, 2019
<https://theUmpShow.com> - Baseball video game
The post seems to be about changing the Leaderboard and doesn't comment too much about whether the actual real-life performance of LLMs is plateauing and what can be done about it.
The docs still mention bigger models with 128k tokens and smaller models with 8k tokens. It seems reasonable to optimize for big and small use cases differently? I don't see how we are being "robbed".
It was pdf.js handling of fonts
I am surprised by the over-engineered comment. Object auto-detection is kind of an advanced feature and I would expect a lot of extra infra required for it. Immich hides it because they just publish docker containers for you but I imagine it would feel more complex if you were to look at the internals.
I tried to use them but haven't really stuck with it.
You want to use their AI model but you don't trust them to not train on your data so you don't want to send your data to them. They don't trust you enough to send you their models.
So you encrypt your data, they compute on your data and send you the encrypted result back.
The only problem is that you have turned an expensive computation into a exponentially more expensive computation
They usually all have pager duty integration as well.
Some examples:
Datadog: https://docs.datadoghq.com/synthetics/ Grafana cloud: https://grafana.com/grafana/plugins/grafana-synthetic-monito...
Here is my summary of it[1] and the original paper[2] I found this in
[1] https://azeemba.com/posts/homomorphic-encryption-with-images...
You play as RBG fighting monsters with different RGB values.
With that attention comes many new people finding it but also discovering that they can't use it.
I assume the use here extends the uncertainty part of the problem to the unsubscribe workflow.
Last year, someone got got CVE's assigned for a curl issue for code that didn't exist AND managed to get a high severity assigned to it. So curl becoming a CNA lets them provide some control to this process.
I am surprised to hear about a PD controller though. In my testing, a PI controller seemed to behave much better than a PD controller. In my research, it seemed a common strategy to drop the D-component completely but I did not see the suggestion to drop the I-component.
You could remove the root motion of the animation and control the position of runner in code. This would allow controlling the position/speed easily but to look natural you still have to tweak the animation speed dynamically. Just changing the foot position via IK would not be enough to make the animation look natural. So feels like you are still left with the original problem of tweaking the animation speed.
The bug bounties also prefer seeing a working attack instead of theoretical reports. So not sure how they could have tested their attack in this situation without making actual changes.
- For drag-and-drop/WYSIWYG, I really like DrawIO. They have a web version https://app.diagrams.net/ but I strongly recommend the desktop version https://github.com/jgraph/drawio-desktop/releases/
- For text-as-diagram, I think Mermaid wins this by default since GitHub added markdown support for these: https://mermaid.live/ (This was github's announcement https://github.blog/2022-02-14-include-diagrams-markdown-fil... )
One paper I used showed how X-rays can be analyzed without exposing the X-ray data itself.
This can be generalized more to any situation where one party owns a proprietary algorithm and another party owns sensitive input data but they don't trust each other.
Modern LLMs are actually the perfect example. You want to use chatgpt but they don't want to give you their model and you don't want them to see your input. If HME was more efficient, you could use it so that the person executing the model never sees your real input.
The goal for Math.random is to have much lower memory and speed impact. You can certainly disagree with their priorities but given those priorities, it makes sense to go with a much simpler algorithm that can be reversed easily.
Your observation is correct and the surface of the sphere is a metric. The ratio of radius to circumference is not constant with that metric though so I feel like something should disqualify it. But I am not sure how.
So I think your observation shows that we need a stronger constraint than just being a metric. Other commenters have hinted that you need a normed vector space but I am not sure if that's sufficient.
Any metric that "pulls on the origin" compared to Euclidean distance will have to do the mapping in a continuous way. This will basically result in both the radius and circumference being expanded in that metric.
Matter of fact, I linked an article that proves that for _all_ metrics, the value of π is always between 3 and 4 (inclusive). Unfortunately the article might have gotten the hug of death so here is an alternative link: https://www.researchgate.net/publication/353330827_Extremal_...