🦋 Welcome to the MAIN() IRC channel of the Raku Programming Language (raku.org). Log available at irclogs.raku.org/raku/live.html . If you're a beginner, you can also check out the #raku-beginner channel!
Set by lizmat on 6 September 2022.
00:02 arkiuat joined 00:34 abraxxa left 00:35 abraxxa joined 01:05 librasteve_ left
disbot <simon_sibl> I encounter an issue here, not sure if thats because of my code or not 01:31
<simon_sibl> termbin.com/v3bi
<simon_sibl> the http part works well, but the netcat part (to create a paste) doesnt work well
<simon_sibl> basically netcat doesnt wait for the server to send back the file ID
<simon_sibl> I tried many different flags, but none of them seems to solve it, it seems that netcat doesnt want to read after the first_line receive part for some reason 01:32
01:34 hulk joined 01:35 kylese left 02:07 Guest13 joined 02:15 hulk left, kylese joined 02:23 Guest13 left 03:04 Aedil joined
disbot <simon_sibl> alright, this work: termbin.com/9bvp 03:17
<simon_sibl> but I need to use the --no-shutdown flag on netcat
<simon_sibl> okok, that with nmap/ncat with gnu/netcat it works as expected ! 03:23
04:24 arkiuat left 04:32 arkiuat joined 05:48 crnlskn joined 06:06 kst``` joined, perryprog_ joined, hulk joined 06:07 apa_c joined, kylese left, jmcgnh left, apac left, sftp left, releasable6 left, perryprog left, kst`` left, jdv left 06:08 jmcgnh_ joined 06:09 jdv joined 06:10 sftp joined, sftp left, sftp joined, releasable6 joined, jmcgnh_ is now known as jmcgnh 06:11 arkiuat left 06:20 lizmat left 06:23 arkiuat joined 06:25 kst``` left 06:27 arkiuat left 06:29 lizmat joined 06:34 [Coke]_ joined 06:37 [Coke] left 06:57 arkiuat joined 07:02 arkiuat left 07:22 Sgeo left 07:25 ACfromTX left 07:28 ACfromTX joined 07:29 arkiuat joined 07:36 lichtkind joined 07:38 arkiuat left 07:48 ACfromTX left 07:51 arkiuat joined 07:56 arkiuat left 07:58 lizmat left 07:59 lizmat joined 08:00 ACfromTX joined 08:01 Aedil left 08:17 dakkar joined 08:25 arkiuat joined 08:30 arkiuat left 08:33 arkiuat joined 08:38 arkiuat left 08:55 abraxxa left 08:56 abraxxa joined 09:09 arkiuat joined 09:23 arkiuat left 09:24 icovnik left 09:51 arkiuat joined 09:58 arkiuat left
disbot <antononcube> @librasteve One way to explain LLM graph to you is the following scenario. We make free different graph nodes computing measurements for queries by: 1. Crag 2. WolframAlpha 3. LLM And a judge node: 4. LLM to pick a result from 1,2,3 10:10
<antononcube> The computations in the nodes 1,2,3 happen asynchronously. 10:11
<librasteve> .oO - that's is very cool 10:12
<antononcube> This can be an interesting experiment, actually, and a good example use case.
10:13 arkiuat joined
disbot <librasteve> my use case is a "handy calculator on the command line" 10:14
<antononcube> "LLM::Graph" is infrastructural, it should be nice and easy to use it "under the hood." 10:16
10:18 arkiuat left 10:21 Aedil joined
disbot <librasteve> =b 10:25
10:47 arkiuat joined 10:52 arkiuat left 11:08 arkiuat joined 11:12 pierrot left, pierrot joined 11:13 arkiuat left 11:16 apa_c left 11:19 apogee_ntv left 11:20 apogee_ntv joined 11:37 arkiuat joined
disbot <antononcube> Does “App::Crag” use LLMs, or do 11:41
11:42 arkiuat left
disbot <antononcube> I mean, that there is also the inversion of control possibility using the so-called LLM function calling. 11:42
<librasteve> okaay - never heard of that - maybe a quick way to write an LLM::Graph node 11:43
<antononcube> Please, see here: rakuforprediction.wordpress.com/20...-1-openai/ 11:45
<antononcube> The idea is what LLMs can call a function running locally on your computer. That function is being described to them with some JSON schema. 11:47
<antononcube> Until a year ago was unreliable.Now many people say it is. 11:48
<librasteve> ok if you have App::Crag installed, then you can go crag '?^<elephant mass in kg>' on the command line
<librasteve> you get back a Physics::Measure.new(value => 6000, units => 'kg').Str 11:50
<librasteve> (ie its stringified to 6000kg)
<librasteve> maybe should go in the "ways to check the calculation output of your LLM" 11:51
<antononcube> That is LLM-graph way. The LLM-tool way is describe to the LLM what is the tool (run-cmd in this case) and the LLM would decide when to call it. 11:57
<antononcube> The LLM-graph can be seen as a generalization of
Voldenet mcp is still still not strictly reliable 11:59
it's more reliable, but sometimes ollama updates just break things there
12:00 arkiuat joined
disbot <librasteve> please dont use App::Crag to build a satellite 12:00
<antononcube> No, just to build an elephant.
Voldenet and hide it in the room 12:01
disbot <antononcube> Well, or just measure a few…
Voldenet there are problems with mcp servers still, for example destructive or non-repeatable calls get called multiple times
in non-interactive way 12:02
disbot <antononcube> @Voldenet Yeah, that is the challenge with updating the LLM packages— LLM providers can’t help themselves but break their APIs little by little.
<librasteve> ?^<typical room volume in m3> / ?^<elephant volume in m3> 12:03
<librasteve> 6①
<antononcube> To some extent, that’s why I implemented LLM-graph. It can give you more control and let you use multiple agents to get results. 12:04
Voldenet I'm not really annoyed too much about it, because the models ecosystem still in very early stages
disbot <librasteve> ?^<elephant cost in US$> => 100000USD
12:04 arkiuat left
disbot <antononcube> Well, I’m annoyed, because people demanded at work and expect a lot. 12:05
Voldenet well, walking on water is as easy as implementing software according to spec 12:08
very easy if they're both frozen
disbot <antononcube> At some point some developers want to vaporize them all. 12:12
<antononcube> This reminds me — I like pointing out that Agile was invented/shaped/verbilized at ski resort. I.e. around “frozen water” with clear vision of obstacles when skiing downhill. 12:15
12:20 arkiuat joined
Voldenet it still works when floor is lava, you just need to sprint more ;) 12:21
12:25 arkiuat left 12:31 crnlskn left 12:36 crnlskn joined
disbot <antononcube> Yeah, limited scenarios with floors. If they came up with such a methodology in the Everglades, here in Florida, I would be much more impressed. 12:41
<antononcube> @librasteve Can you "surface" run-cmd of "App::Crag", so it can be used in Raku sessions? 12:49
12:51 apa_c joined 12:52 apa_c left, apa_c joined 12:54 arkiuat joined 13:00 apa_c left
Geth raku-mode: ttn-ttn++ created pull request #62:
Indent method calls when they happen on a new line
13:05
13:12 apa_c joined
disbot <librasteve> ok if you have App::Crag installed, then you can go crag '?^<elephant mass in kg>' on the command line 13:24
<antononcube> I get that. But I want to use "App::Crag" functionality within Raku sessions. Having those functionalities via run or shell seems too indirect. 13:26
<antononcube> But, well, that is way to make a prototype of comparing "App::Crag" to other systems. 13:27
13:51 apa_c left 14:12 crnlskn left 14:23 perryprog_ is now known as perryprog
disbot <librasteve> use App::Crag; run-cmd '?^<elephant mass in kg>'; now supported (give me a couple of mins to fez upload v0.0.33) 14:27
<librasteve> done the fez .... 14:29
14:30 apa_c joined 15:08 [Coke]_ is now known as [Coke] 15:14 apa_c left 16:10 melezhik joined 16:32 japhb left 16:38 japhb joined 16:40 dakkar left 16:42 arkiuat left
disbot <antononcube> @librasteve Cool, but can give an example of using run-cmd and/or eval-me ? 16:58
<librasteve> run-cmd '?^<elephant mass in kg>'; 17:03
<antononcube> I tried that and others...
<antononcube> I get: > Error: X::Method::NotFound «No such method 'tell' for invocant of type 'Out'» 17:04
<librasteve> looks like you are in the repl - that's what you start with the crag cmd with no args (kust like going raku on the command line 17:05
<librasteve> assume I know nothing about jupyter - so you are either (i) in some raku code (eg in a file) that will be run via raku file.raku in which case put use App::Crag; run-cmd '?^<elephant mass in kg>'; in the file or (ii) you are at the Unix command prompt > in which case type in crag '?^<elephant mass in kg>' (like raku -e say "hi";') 17:20
<librasteve> alles klar? 17:21
tbrowder hi, any user here ever hear of multi-os app Ventoy used to created live USB's 17:27
if so, is it known to be save for use with private data and free from viruses? 17:28
*safe
17:29 human-blip left 17:31 human-blip joined
disbot <antononcube> @librasteve This is what I got so far: 17:55
<antononcube> cdn.discordapp.com/attachments/633...67cdf&
<antononcube> I cannot run crag in Jupyter easily and / or asynchronously. 17:56
<librasteve> aha 17:59
<librasteve> there are 3 raku prefixes in play - ^<foo>, ?<bar> and ?^<baz> 18:01
<librasteve> ^<foo> means ^<value unit [±error]> 18:02
<librasteve> ?<bar> means ?<some text goes to Gemini via LLM-DWIM> 18:03
<librasteve> ?^<baz> means ?^<some text to llm 'in units'> 18:05
<librasteve> note thats query caret ?^
<librasteve> btw all three are raku prefixes so any quoting format '', "" <> will work 18:06
<librasteve> use "" if you want to interppolate $vars of course
<librasteve> (<> is adjusted to join the words with a space to make a string) 18:07
<librasteve> in the case of baz - there is some prompt engineering and the in units part at the end is both passed to the LLM and regexed out to make the units part, so has to be valid 18:08
18:30 melezhik left 18:31 librasteve_ joined 18:33 phogg left
disbot <antononcube> The LLM extension is not that interesting using the LLM-graph POV. It is better to know when to use "App::Crag", with what kind of queries. (I.e. a certain Raku code.) 18:44
<antononcube> I am not sure to what degree "Physics::*" packages know about elephants and other animals.
18:46 phogg joined
disbot <antononcube> Again, this is the perspective of using LLM-tools -- LLMs do the the LLM-ing and deterministic computations are done locally. 18:46
20:02 apac joined 20:33 Aedil left
disbot <librasteve> did you get ?^ to run? 20:48
Geth Pod-To-HTML/main: 127a900228 | (Steve Dondley)++ (committed using GitHub Web editor) | README.md
correct module name for install instructions
20:52
Pod-To-HTML/main: c2204db2f5 | librasteve++ (committed using GitHub Web editor) | README.md
Merge pull request #94 from sdondley/master

correct module name for install instructions
disbot <antononcube> @librasteve Not in Jupyter -- used run. On the command line -- yes. 20:58
<librasteve> the point is this ?^<some text to llm 'in units'> ... ie dual parsing of LLM query ... one to condition the Unit::Parser, the other to shape the LLM response (right?) 21:02
<librasteve> so to ingest the LLM response to an analytic codebase
<antononcube> Ok. But I am more interested in the following: - Make a classifier for recognizing "App::Crag" deterministic computations input - Decide should "App::Crag" be used or not - If not just use Wolfram|Alpha and LLM - Or, use Wolfram|Alpha and LLM always - Have a LLM judge to choose the final answer 21:06
<antononcube> The classifier can be made with the inputs in "App::Crag" test files.
21:13 LainIwakura joined 21:16 LainIwakura left 21:28 jgaz joined 21:34 LainIwakura joined 22:05 phogg` joined, japhb_ joined 22:10 phogg left, japhb left 22:12 LainIwakura left 22:18 Sgeo joined 22:40 LainIwakura joined 22:52 LainIwakura left 23:24 phogg` is now known as phogg 23:47 lichtkind left