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 harder to publish, since if you learn Task X with innate Knowledge, its not as general, so reviewer can claims its just (feature) engineering - Which hurts acceptance, so people always try to frame their work as general as possible
- Historical reasons due to the conflict the symbolic community (which rely heavily on innate knowledge)
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 broken.
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...)
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...
"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 used for a lot of research, most AI (read modern DL) research is not about understanding human intelligence, but automatic things we could not do before. (Understanding human intelligence is nowadays a different field)
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 with a proof-of-concept prototype to transform it into an economically viable product.
So, comparing these success rates does not make sense. Multiple research groups can deliver rough prototypes at the end and celebrate their "huge success". In most fields, there can be only a few economically surviving startups...
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 than your average lecture) and don't have to suck at it. After this checkbox, the differentiating factors are about citations and how much grant money you can/could/do have... (Western Europe, maybe somewhere else it's different).
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)
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..
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: "Do I really need that experiment or can I shut down the sever to save money?".
Makes a huge difference over longer periods.
For companies or if you just want to try a bit, then the cloud is a good option, but for (Ph.D.) researchers, etc., the frictionless local system is quite powerful.
"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.
> 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 way more nuanced than this. Of course, you need a decent "accuracy" (not necessarily the metric), but in many business cases, you don't need high accuracy. But you need a solid process: you can catch errors later, you can cross references etc, you need to failsafe, you need to have post-mortem error handling, etc...
I shipped stuff (classical ML) that was nothing more than "a biased coin flip," but that still generates value ($) due to the process around it.
> 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 you get one that is over powered.
The visualizations only show that at the Border there are a lot of fractals, not in every part of the space.
(Although the highest performance is often achieved close to the border.). I would not state hparam search as bad as that..
"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 highly batched & pipelined. (And you have to batch to train in any reasonable amount of times, and GPUs shine in batch regime)
When a single user request comes in, you just want the prediction of that single input, so no backprogation and no batching. Which is more CPU friendly.
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 look for candidates knowledgeable in that field, and just playing around with prophet dosnt make you knowledgeable)
Especially in TS is saw many PhDs, really focussed on one specific method, not looking left and right and neither are pragmatic about their choices. Which is essential to solve anything in real life. Filtering candidates out, based on criteria like "played around with different stuff" is a good indicator for open-minded people IMHO.
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, glasses etc) IS impressive.
The problem is this case is the data and/or wrong objectives, but the "AI" here has a lot of awareness, just on the "wrong" signals.
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 mostly pickle performance is absolutely fine.
BUT the security problems still remain and weigh much higher
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 harder to publish, since if you learn Task X with innate Knowledge, its not as general, so reviewer can claims its just (feature) engineering - Which hurts acceptance, so people always try to frame their work as general as possible - Historical reasons due to the conflict the symbolic community (which rely heavily on innate knowledge)