jononor

50 ms·

901.
by jononor 4y ago
Who would be good enough?
902.
by jononor 4y ago
My go-to for embarrassingly parallell problems that are still single-computer scale, using the multiprocessing backend. Like computing features across a bunch of inputs. Does the job in a simple and pain free manner. Have got some wrapper f
903.
by jononor 4y ago
That is very interesting. So you have any resources to learn more about this approach to modelling?
904.
by jononor 4y ago
Why do you say that? There are microcontrollers announced with neural network accelerators integrated, and low power FPGAs with accelerators on the market, along with many chip startups aiming to provide milliwatt scale dedicated neural chi
905.
by jononor 4y ago
You do not have to cast pandas DataFrames when using scikit-learn, for many years already. Additional in recent version there has been increasing support for also returning DataFrames, at least with transformers and checking column names&#x
906.
by jononor 4y ago
The mingw toolchain is for cross compiling to target Windows. It is very mature and are used by many open source projects, especially desktop software which is popular on Windows but mostly developed by people running Linux.
907.
by jononor 4y ago
Title is not very good. The reason is that the grid operators have a policy that earlier valid applications for more power get priority over later ones. And the TikTok data center was applied before the Nammo manufacturing expansion was. Qu
908.
by jononor 4y ago
Cool to hear people interested in audio. If you also are interested in Machine Learning, consider joining the Sound of AI community. https://thesoundofai.com/index.html#about
909.
by jononor 4y ago
How much data do you need to move in/out? How much variation is there in demand?
910.
by jononor 4y ago
Which compiler flags do I use to turn off UB?
911.
by jononor 4y ago
Does anyone know of time-series data with Twitter incidents/up/downtime? Would be interesting to do an analysis of the service level and change point analysis wrt the firing.
912.
by jononor 4y ago
Quite rare in practice though. Only case I have had that is gh-pages branch in a GitHub project repository, which contained only generated artifacts (generated by code in master). Which I believe there are better ways to do now. At least wi
913.
by jononor 4y ago
If you would put the most evil genius human into Bing LLM, how are their chances for ending the civilisation? I think pretty poor, because the agency of a chatbot is quite low. And we have good chances of being able to shit it down. The com
914.
by jononor 4y ago
Yes a virus that is very deadly, or one that renders the persons infertile, could be one. Or a bacteria that take everything of a critical resource, like oxygen of the atmosphere. But my argument for safety here is that Bing bot is severely
915.
by jononor 4y ago
Ok, I'll bite. If an LLM similar to what we have now becomes conscious (by some definition), how does this proceed to become potentially civilization ending? What are the risk vectors and mechanisms?
916.
by jononor 4y ago
Two laptop model with the same CPU can have big differences in thermal design, which impacts a lot the fan noise but battery efficiency only in a marginal way.
917.
by jononor 4y ago
Do some something that sounds similar but writes differently? Like a rhyme. Knacknus Saur, Binackus Naurackus, ?
918.
by jononor 4y ago
For ChatGPT and similar LLMs there is the loss function does not reward accuracy/truthfulness/correctness. But instead answers that seem correct to a superficial human check. These rewards making something plausible up instead o
919.
by jononor 4y ago
I think that learning to acquire new/additional training data would be a better first step towards learning agents, than trying to mutate its structure/hyper-parameters.
920.
by jononor 4y ago
What is actually the fraction of human knowledge available on the web? How does the trajectory look like? What is the fraction for knowledge produced the last 5 years?
921.
by jononor 4y ago
How many would buy a gaming-first phone (by Valve)? I don't quite know what such a product would mean in practice, or if there is a killer "app" (game) for phone. But it would have higher probabilities of success than "a
922.
by jononor 4y ago
Not sure about 200B+ models on phone hardware in 10 years. But I think we'll be able to deliver the same predictive performance with models 1/10 the size and those will fit. That is what happened with CNN vision models over the la
923.
by jononor 4y ago
Stack Of never paid anyone though, so it seems possible for Google to launch a service where people answer questions to feed the AI and give back similar tokens as SO. I mean, they could make a game, where people have to try to beat the AI
924.
by jononor 4y ago
No, all programming philosophies is a about composition? What units one expresses a program in, and how they can be combined together.
925.
by jononor 4y ago
Which barriers do you considered to be the most problematic (and why)? I think that using todays CPU/GPU on a modern phone, one would be able to run a model capable of a Word Error Rate of under 10%. I mean, it would take up maybe 1GB
926.
by jononor 4y ago
For a smartphone to be fundamentally unexplainable by relevant experts one would have to go back more than 100-200 years, I think. The first trans Atlantic radio transmission was in 1902, for example. Though I guess even the experts of the
927.
by jononor 4y ago
Yes, it works fine since years. I have made around 200 objects (and many iterations on each) using the Part Design workbench, that have been 3d printed with Cura. STL export has never been a problem.
928.
by jononor 4y ago
Blender as an open source project was borne out of a company that ceased to exist. It has had multiple people working on it full time with a product mindset from the very onset. This is very different from largely volunteer/spare time
929.
by jononor 4y ago
A high-end phone is quite capable of doing automatic speech recognition continuously, as well as NLP topic analysis. The last years voice activity detection has moved down into the microphone itself, to enable ultra low power always-listeni
930.
by jononor 4y ago
Interesting. 2.2 million seems to be around the top 10% in the US, and around 20x the median (Federal Reserve data from 2016). It also places one in the global top 1%.

More ›