09:58:40	 From  Chris (@UncensoerdNews) : What a great talk.  Thank you so much.
10:12:07	 From  Dave Jaffe : https://vcfed.org/vcf-west-event-schedule/
10:14:28	 From  Brad Nelson : object oriented forth - dick pountain
10:14:43	 From  Chris (@UncensoerdNews) : Is dan golding doing a talk, or did I miss it?
10:14:45	 From  Juergen Pintaske : https://www.amazon.co.uk/Object-Oriented-FORTH-Implementation-Structures/dp/0125635702
10:14:56	 From  Chris (@UncensoerdNews) : Great.
10:18:19	 From  Bob Armstrong : The object it CoSy is list w header `( Type Count refCount )`
10:19:56	 From  Brad Nelson : A lot of deep learning actually doesn't need big floating point, in fact it uses float16:
https://en.wikipedia.org/wiki/Half-precision_floating-point_format
10:20:25	 From  Travis Bemann : it's dynamic range that matters, not mantissa size
10:21:08	 From  Travis Bemann : float16 is much more efficient for AI than float or double
10:21:34	 From  Travis Bemann : note that a lot of AI is now done on GPU's optimized for computing lots of "neurons" at once
10:21:59	 From  Juergen Pintaske : The question would be how much resolution the algorithms actually need? What in biology needs more than 32 bit?
10:23:30	 From  Travis Bemann : if anything, what is needed is being able to process large numbers of float16's in paralle
10:24:55	 From  Brad Nelson : My impression (not an expert) is that as Travis says, it's more about the dynamic range of floats, and actually not many bits needed. But then GPUs and CPUs have certain sizes they've been optimized to work well with for other purposes so that feeds into the sizes folks have chosen.
10:25:24	 From  Brad Nelson : Though there's also a difference between what you might want in terms of range to train the network vs use it.
10:25:26	 From  Chris (@UncensoerdNews) : Which is the recommended $30 board?
10:27:06	 From  M Edward Borasky (@znmeb) : Speaking of $30 boards, I have one! https://tinyvision.ai/products/upduino-v3-1
10:27:32	 From  M Edward Borasky (@znmeb) : I am looking for an open source Forth to run on it!
10:28:23	 From  Liang Ng : Jones FORTH??
10:29:08	 From  Brad Nelson : Also this paper claims you can squeeze weights down to 8-bit fixed point or even 3-bit log scale:
https://arxiv.org/pdf/1603.01025.pdf
10:29:33	 From  Chris (@UncensoerdNews) : Thank you.  I will mention that in their issues page on githuyb.
10:30:52	 From  Chris (@UncensoerdNews) : A most interesting paper.
10:34:58	 From  Bob Armstrong : neural nets don't need high precision , The ` multiplicative fn  simpler than sigmoid which apparently works better . In CoSy :
: ai_mult .+ 0>i *i ;
10:36:00	 From  Brad Nelson : Yeah ReLU has actually been apparently very effective and suffers less from vanishing gradient problem.
10:43:03	 From  Liang Ng : I suspect neural network can make up high precision number by simply combining multiple low resolution neurons.
10:43:29	 From  Liang Ng : They exist in large number anyway, don't they?
10:44:30	 From  Travis Bemann : you can't get higher precision values through combining lower precision values, but neural nets don't need them in the first place
10:54:20	 From  Don Golding : https://www.facebook.com/groups/1304548976637542
10:54:39	 From  Dave Jaffe : there are hearing aids that allow the user to setup the audio graphic equalization and has several setups for different audio environments
10:55:06	 From  M Edward Borasky (@znmeb) : What is the CORE I GitHub repo??
11:09:39	 From  Chris (@UncensoerdNews) : https://github.com/angelus9/AI-Robotics
11:12:19	 From  M Edward Borasky (@znmeb) : Thanks!!
11:13:43	 From  Chris (@UncensoerdNews) : KISS
11:15:01	 From  M Edward Borasky (@znmeb) : James Bowman did an FPGA Forth if I recall correctly
11:17:53	 From  Kevin Appert : James used his FPGA Forth Engine in several products...
11:18:34	 From  M Edward Borasky (@znmeb) : Is that J1A / SwapForth? 16 bits?
11:19:13	 From  Kevin Appert : GameDuino and his HDMI "Dazzler"
11:25:36	 From  M Edward Borasky (@znmeb) : Found it … the one that uses the Pico is sold out! :-(
11:26:20	 From  M Edward Borasky (@znmeb) : But yes, it's SwapForth / J1A
11:26:27	 From  Kevin Appert : that may be another word for 'discontinued'
11:28:18	 From  Travis Bemann : it depends
11:28:27	 From  Kevin Appert : James Bowman did the J1 Forth Engine and others have done variants.   More at his website Excamera.com
11:28:49	 From  Travis Bemann : the RPi4's (not RPi Pico) are sold out... but not because of being discontinued but rather because of being bought up perpetually by scalpers
11:29:17	 From  M Edward Borasky (@znmeb) : Yeah … "sold out" is capitalist-speak for "not profitable"
11:30:07	 From  Chris (@UncensoerdNews) : So who was the person interested in building hearing aids?  What is his email addrss?
11:31:07	 From  M Edward Borasky (@znmeb) : @Travis You can buy spools of Picos :-)
11:31:24	 From  Chris (@UncensoerdNews) : I suspect the the web server application makes it load different memory banks.  Killing the cache.
11:31:57	 From  Juergen Pintaske : I somehow cannot understand, that it seems that only these people here know how to do a better PC. Try to run ExCel or Word on this Core1. If you just run programs that fit in - fine.
11:37:18	 From  Travis Bemann : I think the thing is that if you can't buy something at all, it's really discontinued, whereas if you can buy it, but you're paying > $100 it's the scalpers at fault
11:42:02	 From  Kevin Appert : there are quantities of discontinued chips available for big usd
11:42:15	 From  Travis Bemann : Just a shameless plug, but I have a new Facebook group for zeptoforth, https://www.facebook.com/groups/zeptoforth
11:42:55	 From  M Edward Borasky (@znmeb) : We need to have a Mastodon Forth presence!!
11:43:34	 From  Travis Bemann : I just created a new Mastodon account for myself yesterday or the day before
11:45:46	 From  M Edward Borasky (@znmeb) : My Mastodon invite: https://ravenation.club/invite/sQTUA3op
11:47:07	 From  Chris (@UncensoerdNews) : https://mastodon.social/tags/forth
11:47:19	 From  Travis Bemann : I'm https://mstdn.social/@tabemann
11:47:58	 From  Chris (@UncensoerdNews) : I am. https://mastodon.social/@PythonLinks
11:48:46	 From  M Edward Borasky (@znmeb) : I'm https://ravenation.club/@AlgoCompSynth
11:51:35	 From  Dave Jaffe : the buttons could have been a Chuck Moore keyboard
12:00:29	 From  Dave Jaffe : Obituaries - http://www.forth.org/svfig/memory.html