krallistic

6 ms·

1.
by krallistic 1y ago
Various reasons Some people just believe there is no innate knowledge or we dont need it if we just scale/learn better (in the direction of Bitter Lesson) (ML) Academia is also heavily biased against it due to mainly two reasons: - Its
2.
by krallistic 1y ago
Seems like a band-aid solution for a broken system. But in general science will have to deal with that problem. Written text used to "proof" that the author spend some level of thought into the topic. With AI that promise is bro
3.
by krallistic 1y ago
Its quite funny to see that LLMs reviewed interest in KnowledgeGraphs/Reasoning/Triple Stores etc... since (on a high level) they both are often pitched to solve the same goal. (E.g. Ask an AI about a topic...)
4.
by krallistic 1y ago
And sampling from a (now fixed) distribution can be made deterministic... So the total generation of text from an LLM can be made fully deterministic. The problem for scientists is that we cant do that in the deployed systems...
5.
by krallistic 2y ago
"The goal of Automated driving is not to drive automatically but to understand how anyone can drive well"... The goal of DeepBlue was to beat the human with a machine, nothing more. While the conquest of deeper understanding is us
6.
Proof of Thoughtfulness: Writing in the Age of AI(robmoore.tech)
1 points by krallistic 2y ago|0 comments
7.
by krallistic 2y ago
Start-ups deliver something much more complicated/different than these research projects. If the whole research project at the end actually delivers a somewhat coherent prototype, it's seen as a huge success. Most start-ups start
8.
by krallistic 2y ago
Teaching is not really relevant in the hiring process of professors. I saw several committees for prof position and teaching is treated like a checkmark. You should done it and provide a small sample lecture (which you prepare much more tha
9.
by krallistic 2y ago
"Deep learning is hitting a wall now with just scaling" "Deep learning is only good for perception" (with language one of the areas where its not good)
10.
by krallistic 2y ago
But in this example, many of the programmers know the alternative (i.e they learned c++, java etc) and still prefer something else...
11.
by krallistic 3y ago
That point is also a bit of a stretch. While there are a few papers on using KG/Ontologies to enhance training, this is really far from the mainstream, and i would be surprised if it would be used anywhere (outside of a research paper)
12.
by krallistic 3y ago
Death at Via Ferratas are extremely rare (with proper gear ofc). But the falls are notoriously hard. Force can get really high and the chance of injury is high, even in cases where the fall zone is free of any metal etc..
13.
by krallistic 3y ago
This. In the second article, the author touches on this a bit. With a local setup, I often think, "Might as well run that weird xyz experiment over night" (instead of idling) On a cloud setup, the opposite is often the case: &quot
14.
by krallistic 3y ago
"Search" has been the name for the A* space. So it makes absolute sense. "Planning" means in the symbolic AI space often something quite in the direction of PDDL/STRIPS/ICAPS planning.
15.
by krallistic 3y ago
> The biggest thing any ML practitioner realizes when they step out of a research setting is that for most tasks accuracy has to be very high for it be productizable. You can do handwritten digit recognition with 90% accuracy? It's
16.
by krallistic 3y ago
> gives me the impression that anything other than grid search hyper parameter optimization is a fools errand. This would give credence to the notion that hyper parameter tuning really is akin to just re-rolling a character sheet until y
17.
by krallistic 3y ago
I would guess because China and Russia are relevant markets for tesla...
18.
The N Implementation Details of RLHF with PPO(huggingface.co)
1 points by krallistic 3y ago|0 comments
19.
by krallistic 3y ago
"Inference" - getting the predictions out of the model. While training you need to run: Input -> Model -> Output (Prediction) - Compare with True Output (Label) -> Backpropagation of Loss through the Model. Which can hig
20.
by krallistic 5y ago
If claims to have good knowledge in TS, but had not played around with Prophet (which is dead simple, takes a few hours - or 15 minutes for you - to see some problems), how interested/knowledgeable are they in TS? (I assume that they l
21.
by krallistic 5y ago
Its easy to point at the bookshelf example and say "Haha AI is stupid", but its actually quite impressive. One could easily argue that most human interviewers have similar bias, and that it can detect such complex signals (books,
22.
by krallistic 6y ago
For must use-cases the performance cost of loading the model are pretty low compared to either their training cost or making thousands of inference calls (when used in an API). Mayebe they matter if you do AWS Lambda with ML Models, but mos
23.
by krallistic 6y ago
"Among young men [...] and the growing percentage who coreside with their parents all contribute significantly to the decline in casual sex. The authors find no evidence that trends in young adults’ economic circumstances, [..] explain
24.
by krallistic 6y ago
Sure if no methods/work exists. In this case there already has been some work done, to mitigate these issues. So either mention them in related work and/or highlight why theses methods arent enough.
25.
Symbolic Behaviour in Artificial Intelligence(arxiv.org)
2 points by krallistic 6y ago|0 comments
26.
by krallistic 6y ago
Keep the music - excellent branding.
27.
by krallistic 6y ago
The comparisons to NLP presents a good view on the problems. Its "easy" to write some logic rules to parse input text for a 50% demo. But then you want to improve & scale, and suddenly all the nuances, bites you. The rules get
28.
by krallistic 6y ago
Maybe a good indicator that there is only minor (industry) need/benefit. The "biggest" Knowledge Graph is Google, but it is unclear, how much there is actually Semantic Web and how much search, ML, NLP etc.. They are all nice
29.
by krallistic 6y ago
Its nice to see Nepal with this first winter books finally in the history books. They done so much for mountaineering.
30.
by krallistic 6y ago
An average 10km run burns 600cal. Sure just living burns 2000cal, but its a noticeable change. In Addition, for most people its much easier to track their exercise, than pedantic tracking calories.

More ›