| gouthamr | relaying a message i sent to #openstack-release: | 16:00 |
|---|---|---|
| gouthamr | aprice[m] will share a poll for the naming of 2027.1 to the openstack-discuss ML today, Voting will close EOD Monday... the name itself will be announced on Tuesday | 16:01 |
| clarkb | the email should already be in your inboxes (it is in mine) | 16:01 |
| gouthamr | ah! | 16:01 |
| gouthamr | yes, it is: https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/thread/NL3364WME3LMUIJA5NXXJOXOAGP6SDSB/ | 16:01 |
| * gouthamr https://www.kubernetes.dev/blog/2026/06/26/open-source-maintainership-in-the-age-of-ai/ | 18:23 | |
| gouthamr | i didn't know about https://contribute.cncf.io/blog/2025/12/16/github-copilot-enterprise-for-maintainers/ until now, and that they had started rolling out code rabbit... it appears that corporate sponsorships will drive some of this.. an equivalent is someone investing in gerrit plugins for AI assisted reviews, or formalizing work like teim-review that sean-k-mooney was doing | 18:28 |
| gouthamr | s/was/is | 18:28 |
| sean-k-mooney | i havent really looked at coderabbit but i know gibi did breily and didnt like it liek a year ago | 18:30 |
| sean-k-mooney | btu that is what openshift use i bleive | 18:30 |
| clarkb | fwiw I don't think gerrit plugins are the right path | 18:30 |
| gouthamr | yeah that'll need us to scale infra | 18:31 |
| sean-k-mooney | clarkb: ya that would not be my first choice either | 18:31 |
| clarkb | they are too rigid and expect a one size fits all model. So unless someone wants to write a bunch of java to fix that they are a bad choice | 18:31 |
| sean-k-mooney | gouthamr: we also coudl apply for access to the openai oepsouce program if we anted too | 18:31 |
| clarkb | the system mordred is using appears to work well. I also think the proxy through zuul approach could work well too | 18:31 |
| sean-k-mooney | https://openai.com/form/codex-for-oss/ | 18:31 |
| clarkb | and all of that is external to zuul using gerrit event hooks/stream | 18:31 |
| sean-k-mooney | i have not really lookd at the details of that | 18:32 |
| sean-k-mooney | clarkb: ya my third party ci just runs like any other job | 18:32 |
| sean-k-mooney | there are lost of ways to do it without modifying gerrit | 18:32 |
| clarkb | yup and doing it that way allows you to tailor the approach to the team in question which I think is important | 18:33 |
| sean-k-mooney | either as a zuul job or by consuming the event stream | 18:33 |
| gouthamr | agreed, if this is finding success in cyborg/watcher, keep us posted, sean-k-mooney | 18:34 |
| sean-k-mooney | i just got my first contibution form melwitt to improve it | 18:34 |
| gouthamr | on the codex-for-oss, its a 6-month Pro level access, which won't help us imo.. don't trust openai/anthropic/<INSERT AI LAB> to look out for the benefit of open source.. they could just want you lured in, and consider no other choice or benefit from training with your data ¯\_(ツ)_/¯ | 18:36 |
| sean-k-mooney | at the end of the day all teim-review is doing is invoking claude with lightllm providing a api proxy for tokens, and a set of subagent defintions/prompts | 18:36 |
| sean-k-mooney | im curretnly paying them 230 a month anyway. ill proably reduce that at some point. i used to hate gpt models but my opionion has changed since gpt 5.4 came out afte that they started actully doing what i wanted them to do | 18:38 |
| sean-k-mooney | my ci currntly uses glm-5.2 i unfortuetnly dont have the hardware to run that locally but i like that openweight models can actuly work for this use case | 18:39 |
| sean-k-mooney | it will proably take a few year but im hopefuly that gigen openwight models have mostly caught up we will be able to do local infernce with a useful model in the next 2-3 years without remotaging the house to do so | 18:41 |
| gouthamr | ++ yeah.. extending claude/gpt over time would impoverish you :D | 18:41 |
| sean-k-mooney | i actully signed up to the codex max level plan because i wanted to play with gpt modesl and openclaw/hermes agent an not worry about token usage | 18:44 |
| sean-k-mooney | i havent actully used it for that much since i dont really work in a way where a personal assistend is useful to me | 18:45 |
| gouthamr | ack, and you could sign up for https://openai.com/form/codex-for-oss/ independently too? is that what you did | 18:45 |
| sean-k-mooney | but i have been using it to review the work that sonnet/opus did | 18:45 |
| sean-k-mooney | gouthamr: no i coudl i just said i would pay it for a few months and see if i liked it or not | 18:46 |
| gouthamr | yeah | 18:46 |
| sean-k-mooney | currently my ci runs off https://z.ai/subscribe yealy max plan for like $700 i was like i can at least experiemnt with teim-reivew and then see if it proves useful or not | 18:48 |
| sean-k-mooney | its curently seteup to do 1 review at a time and averges 4-6 reviews an hour which is enouch for cybrog and watcher for now | 18:52 |
| gouthamr | makes sense... and have contributors found this useful? | 18:54 |
| sean-k-mooney | in the last 2 weeks it has used 462,617,071 tokens which is sltil only ab out 1% or the plans usage limits. | 18:54 |
| sean-k-mooney | i belvie so but i might have to so a survay at some point and see. i ocationally ask for feedback | 18:55 |
| gouthamr | wish we could go down this path with sponsored hardware, yet, open weight models (vertex/bedrock/<cloud provider>).. | 19:10 |
| sean-k-mooney | im sure that will eventually be possibel but not in the shot term, ram is jsut too expensive right now | 19:23 |
| sean-k-mooney | of couse if nvida want to ship me a dgx station i woudl be happy to use it to run my ci :P | 19:23 |
| sean-k-mooney | lol scan.co.uk list the pice as "let chat" | 19:25 |
| sean-k-mooney | i dont think ill be getting one of thsoe anytime soonhttps://www.ctoservers.com/nvidia-dgx-station--nvidia-gb300-grace-blackwell--748-gb--memory---20-pflops--connectx-8-6130-p.asp | 19:26 |
| sean-k-mooney | £94,895.00 | 19:26 |
| sean-k-mooney | if i recall that was orgianly ment ot be in the 20-35k range before the ram apocalypse made even used ram on ebay insane | 19:29 |
| gouthamr | yikes :D | 20:21 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!