todd underwood
@tmu
Followers
1K
Following
2K
Media
49
Statuses
2K
Reliable Machine Learning/AI at Google. Pittsburgh. A veces Puerto Rico. Views sometimes not even my own; definitely not employer's. He/Him/Él
Joined May 2007
An update on the research reliability (SRE) team and tmu at OpenAI: . (yes, that kind of update). :-). I don't often use twitter but it is something of a habit/fetish of some people so thought it would be useful to post here as well.
0
0
8
Now I want all my favorite shows in lots of languages. In particular, I want all my Latin American news/commentary in English/Spanish both! I used to do translation so I know how hard it is, but I don't care. I want it. @LatinoUSA : ¿que creen ustedes? @Maria_Hinojosa @julito77.
0
0
0
He estado escuchando la 2nda temporada de L Brega de @AlanaLlama . Coño, quánto me encata. Para Boricuas y les demás. Escúchanlo:. (Amazing podcast about Puerto Rico that's also in English for the rest of you).
1
0
3
(not really using twitter much anymore, as i've noticed most of you are not either) but i spotted this over here. wow. really honored. thanks to the community and congrats to all of the other winners! @mlopscommunity @Dpbrinkm @niallm @KrantiParisa
I am honored to present such prestigious awards. Huge thanks to everyone in the MLOps Community who voted for the MLOp awards. Check out all the MLOps award winners here:
2
2
14
It's incredible and humbling that people are reading the book that we wrote. I sincerely hope it is useful and that when we fall short that we are able to learn from that. We're super happy to support book clubs reading it. Thanks, @sophiamyang !.
Happy November! This month Our DS/ML #bookclub is reading @OReillyMedia Reliable Machine by Cathy Chen, @niallm, @KrantiParisa, D. Sculley, and @tmu. Join us for book discussions and reading motivation:
1
2
8
RT @googlesre: Today's sessions at @SREcon #EMEA include our colleague @stevemcghee and @JamesBrookbank, “SRE in Enterprise — check out the….
0
11
0
(signed) Reliable ML Book giveaway at SRECon. Links to the submitted questions (because twitter is terrible at threading). Any others? Books will be given away at 12:30 in Amsterdam.
@tmu i'm not there, but if i provide a shipping label can i get one?. :-). 1) Can I train with non photo/flower photos.2) How many cpus do I really need to process my data?.3) How does one train a dragon (of a dataset) that has known fake data in it, and will ML help me identify them?.
1
0
1
These are great answers so far. Keep them coming if you want a book!.
1
0
0
. Efficacy of ML models are hella hard for me to understand and measure as I've never spent the time to learn it. Outcome of the model becomes less valuable when input data is corrupted, but data validation is hard. Outputs of DALL-E have given me nightmares. I may need therapy.
1
0
0
Question: What about programmatically generated diagrams?. Answer: Diagrams need human editing or they don't make sense. You don't want automatically generated diagrams because you want to focus on the abstract functionality, not the real implementation.
2
0
2
Question: how do you collaborate remotely with drawing?. Mux: I've gone all in on mermaid.js and there's also and they integrate well into github.
1
0
2
Compare your diagrams to the official documentation (and here is the official documentation so you can see how true mine is).
1
0
0
How do you take this and do something useful with it. Draw something from memory and do it again in a couple of days. Look at what changes. Do your colleagues learn something from your diagrams? Do you learn from theirs?.
1
0
0