--- Log opened Fri Apr 14 00:00:32 2023
02:16 -!- codaraxis [~codaraxis@user/codaraxis] has quit [Ping timeout: 260 seconds]
02:17 -!- Llamamoe [~Llamamoe@46.204.72.182.nat.umts.dynamic.t-mobile.pl] has joined #hplusroadmap
02:53 -!- Ludack [~kjlk89@user/Ludack] has quit [Remote host closed the connection]
02:53 -!- Ludack [~kjlk89@84-27-225-150.cable.dynamic.v4.ziggo.nl] has joined #hplusroadmap
03:15 < hprmbridge> eleitl> Interesting comment https://old.reddit.com/r/mlscaling/comments/12l08y1/amazon_ceo_on_llm_scaling_the_really_good_ones/
03:26 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
04:21 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
04:21 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
04:32 < hprmbridge> Perry> if that were true, OpenAI could not exist, so it does not pass the smell test.
05:36 < hprmbridge> eleitl> With comment I meant the Reddit comment.
05:53 -!- Ludack [~kjlk89@84-27-225-150.cable.dynamic.v4.ziggo.nl] has quit [Remote host closed the connection]
05:53 -!- Ludack [~kjlk89@84-27-225-150.cable.dynamic.v4.ziggo.nl] has joined #hplusroadmap
06:01 < hprmbridge> Perry> What reddit comment?
06:01 < hprmbridge> Perry> You didn’t link to a comment.
06:03 < hprmbridge> eleitl> In case it's unclear which, this specifically https://old.reddit.com/r/mlscaling/comments/12l08y1/amazon_ceo_on_llm_scaling_the_really_good_ones/jg6dksg/
06:05 < hprmbridge> Perry> The person making that comment does not understand start ups. There is no way you can burn through half a million without thinking about it if you want to keep your company.
06:22 < muurkha> my experience with startups is that founders usually don't keep their company
06:23 < muurkha> I mean that's kind of what taking funding is, in theory and usually also in practice
06:24 < muurkha> and once you've taken funding there's a lot of pressure to use it, to find a direction to send that money in that has a potential high payoff
06:25 < muurkha> if you stop growing you're dead.  if you run out of money, on the other hand, you can probably raise another round of funding if you're growing rapidly.  at least that was the thinking
06:25 < muurkha> it's maybe not very good thinking for the current high-interest-rate environment
06:38 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
06:38 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
08:07 -!- test_ is now known as _flood
08:30 < kanzure> gptpaste: what is optical data tape?
08:30 < kanzure> https://www.biosearchtech.com/products/pcr-reagents-kits-and-instruments/pcr-instruments-and-software/array-tape-ultra-high-throughput-platform/nexar
08:31 < nsh> (.gpt)
08:31 < nsh> .gpt what is optical data tape?
08:31 < gptpaste> ​Optical data tape is a type of storage media that uses optical technology to read and write data. It consists of a long, thin strip of plastic or metal coated with a special material that can be written to and read using a laser. Optical data tape is commonly used for long-term archival storage of large amounts of data, such as in libraries, archives, and data centers. It is al - http://sprunge.us/93kt4G
08:35 < kanzure> spool tape?
08:35 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1096458789835784253/nexar.png
08:36 < hprmbridge> kanzure> bad image of array tape but it's something https://cdn.discordapp.com/attachments/1064664282450628710/1096459065414123650/image.png
08:39 < L29Ah> .gpt what is optical data ball?
08:39 < gptpaste> ​As an AI language model, I do not have any information on an "optical data ball." It is possible that the term is a misspelling or a reference to a specific product or technology that I am not familiar with. Can you please provide more context or details about what you are referring to? - http://sprunge.us/spcBr7
08:40 < L29Ah> i wonder how did they make the thing to admit it doesn't know about a subject
08:41 < L29Ah> alpaca readily generates a plausible poppycock as an answer to this question
09:38 < muurkha> this GPT-3.5 chatbot seems to be a lot better at avoiding that
09:39 < muurkha> .gpt Write a scene in a novel where two characters, Bryan and Lauri, are discussing something called an "optical data ball".
09:39 < gptpaste> ​Bryan and Lauri sat across from each other in the dimly lit coffee shop, their eyes fixed on the small, metallic sphere that lay between them. Bryan picked up the ball and examined it closely, turning it over in his hands as he spoke."So, this is the optical data ball?" he asked, his voice tinged with excitement.Lauri nodded, her eyes sparkling with enthusiasm. "Yes, it's a - http://sprunge.us/Prk5NZ
09:40 < muurkha> I think Dan Gelbart's first startup in the 80s was an optical-data-tape startup
09:41 < muurkha> .gpt Write a scene in a novel where two characters, Bryan and Lauri, are discussing something called an "optical data ball" and explaining how it works and what its capabilities are.
09:41 < gptpaste> ​Bryan and Lauri sat across from each other in the small coffee shop, their laptops open and scattered with notes and diagrams. In front of them sat a small, spherical object that glowed with a soft blue light."So, this is the optical data ball," Lauri said, gesturing to the object. "It's a revolutionary new technology that can store and transmit massive amounts of data throug - http://sprunge.us/ijjXxR
09:42 < kanzure> muurkha: yes that is the one fenn was mentioning
09:42 < kanzure> (that person's optical tape)
09:43 < muurkha> eleitl: what do you mean by "scaling limits up to WSI"?
09:44 < muurkha> kanzure: I didn't know fenn had mentioned optical tape
09:45 < muurkha> L29Ah: often you can get it to include knowledge in these fictional segments that it refuses to provide directly
10:03 < hprmbridge> nmz787> Wafer scale integration... See cerebras
10:04 < hprmbridge> nmz787> In my opinion WSI is more of a stunt than something economically cost effective for the future of scaling, as compared to 3D IC packaging advances
10:05 < hprmbridge> nmz787> Getting to sort test the die for yield provides a LOT of economic benefit
10:08 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
10:11 < hprmbridge> nmz787> see slides 6, 10 https://slideplayer.com/slide/18101798/
10:14 < hprmbridge> nmz787> starting DD (defect density) for a given process might be 5E7 /cm^2.... and target is around 0.1 /cm^2.... there's still plenty of yield in the middle times, but if your wafer IS the die... then depending on your architecture it may or may not be usable (or sellable)
10:15 < hprmbridge> nmz787> being able to cut up a wafer, test die, throw away bad ones, and then re-join dies effectively solves the loss during HVM ramp up (process debug)
10:19 < muurkha> oh, wafer-scale integration, thanks
10:19 < muurkha> amusingly, wafer-scale integration was the rule throughout the 01960s
10:19 < muurkha> I think?
10:20 < muurkha> nmz787_: you might be able to design your architecture to tolerate high defect densities
10:23 < hprmbridge> nmz787> this says dicing dates back to at least 1955 https://www.chiphistory.org/documents/dicing.pdf
10:23 < hprmbridge> nmz787> I think back then they may have just cleaved the wafers, rather than sawing them
10:23 < hprmbridge> nmz787> (and now plasma dicing them)
10:40 < muurkha> they do cleave pretty astoundingly well
10:41 < muurkha> even if you don't scribe them
10:42 < muurkha> I think it's still the case that ICs in that time period were not diced
10:42 < hprmbridge> nmz787> I'm not sure I believe you at all
10:42 < muurkha> I could be wrong!  I've only seen a few photos to base this belief on
10:43 < hprmbridge> nmz787> unless you're arguing some difference between cleaving and dicing
10:43 < muurkha> no, no
10:43 < muurkha> I mean the ICs were actually round discs
10:43 < hprmbridge> nmz787> just look at the size of transistors in the 60s
10:43 < hprmbridge> nmz787> TO-5 package, for example
10:44 < hprmbridge> nmz787> tube packages were also round, I think that's the reason TO (transistor outline) packages followed the same outline
10:44 < hprmbridge> nmz787> flatpaks and DIPs were a BIG leap in the industry (I just learned this yesterday, actually), not without pushback
10:44 < muurkha> yeah, I readily believe that people were dicing up transistor wafers by the 01950s
10:44 < muurkha> well, late 01950s
10:44 < muurkha> hmm, I should know what flatpaks were
10:47 < hprmbridge> nmz787> possibly there was some early technology where the PN junction was formed during the ingot production process... and in that case a "wafer" ended up being the die... but I have a feeling that wasn't the bulk of devices produced (though I'm not history buff)
10:47 < muurkha> I don't think I've ever seen these flatpacks
10:47 < hprmbridge> nmz787> it seems at one point dicing was also referring to cutting a crystal into wafers
10:48 < hprmbridge> nmz787> https://en.wikipedia.org/wiki/Flatpack_(electronics)
10:48 < muurkha> maybe, though this VLSI Research document from 01987 is just referring to cutting a wafer into dies
10:48 < muurkha> yeah, that's what I'm looking at
10:52 < hprmbridge> nmz787> dicing 1957 https://sites.google.com/site/transistorhistory/Home/us-semiconductor-manufacturers/ti
10:53 < hprmbridge> nmz787> 1956 circular ultrasonic cutting of wafers to form round dice https://sites.google.com/site/transistorhistory/Home/european-semiconductor-manufacturers/philips
10:54 < hprmbridge> nmz787> 1958 https://sites.google.com/site/transistorhistory/Home/us-semiconductor-manufacturers/western-electric-main-page
10:54 < hprmbridge> nmz787> there's probably a more coherent review article out there
11:51 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Quit: :traffic tickets are what you buy to watch traffic without sneaking in 2023-04-14 18:51:53:533]
11:52 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer]
12:28 -!- Ludack [~kjlk89@84-27-225-150.cable.dynamic.v4.ziggo.nl] has quit [Quit: Leaving]
12:38 -!- flooded [~flooded@146.70.202.51] has joined #hplusroadmap
12:42 -!- _flood [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 264 seconds]
13:09 < hprmbridge> nmz787> ChatGPT says "The broader term that includes any method of converting a semiconductor wafer into smaller functional items is "wafer singulation". This term encompasses not only wafer dicing, but also other processes such as breaking, sawing, or laser cutting. The goal of wafer singulation is to separate the individual components on a wafer into functional units that can be used in electronic
13:09 < hprmbridge> nmz787> devices."
13:09 < hprmbridge> nmz787> I thought dicing was equivalent to wafer singulation
13:20 -!- hplusbot [~skybot@user/hplusbot] has joined #hplusroadmap
13:23 -!- hplusbot [~skybot@user/hplusbot] has quit [Remote host closed the connection]
15:00 < kanzure> https://www.newsdrum.in/international/creating-and-implanting-synthetic-monkey-embryos-could-pave-the-way-to-stem-cell-babies  but it doesn't mention renewal.bio for some reason
15:00 < kanzure> why would they report a failure for synthetic monkey embryos to implant?
15:58 -!- Llamamoe [~Llamamoe@46.204.72.182.nat.umts.dynamic.t-mobile.pl] has quit [Quit: Leaving.]
16:48 < muurkha> nmz787_: I thought so too
17:20 -!- codaraxis [~codaraxis@user/codaraxis] has joined #hplusroadmap
17:20 -!- codaraxis__ [~codaraxis@user/codaraxis] has joined #hplusroadmap
17:21 -!- codaraxis__ [~codaraxis@user/codaraxis] has quit [Remote host closed the connection]
17:24 -!- codaraxis [~codaraxis@user/codaraxis] has quit [Ping timeout: 255 seconds]
18:14 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
18:15 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
18:56 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
19:00 -!- flooded [~flooded@146.70.202.51] has quit [Ping timeout: 260 seconds]
19:02 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
19:10 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 240 seconds]
23:41 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
--- Log closed Sat Apr 15 00:00:33 2023