Robert Monarch

Looking forward to talking later today with @AndrewYNg @BDataScientist & @IvaGumnishka about how AI can impact the world positively and how you can start to evaluate whether or not AI can help in a particular problem in social or environmental impact eventbrite.com/e/ai-for-good-…

Launching today: AI for Good! 🌎 Designed in collaboration with Microsoft’s AI for Good Lab and instructed by @WWRob, it will empower you to tackle real-world challenges using AI, and how to determine if AI is the right tool for the job. Enroll now: hubs.la/Q01WgbtZ0

@dromanocpm @AndrewYNg Thanks, I’ve worked with/at the UN for ~20 years in disaster response, so it informs the course. We intro the Sustainable Development Goals in the course but qualify that they are more useful as a shared language than as a definitive set or division of objectives for impact

@AndrewYNg @WWRob This is what AI for Good at the UN / ITU stands for also. I like the ITU for setting that framework with the WHO.

Our new specialization, AI for Good, teaches a framework for using AI when stakes are high and human lives may hang in the balance. @WWRob presents a practical framework for ML in socially important projects (and products of any kind), and illustrates it with real projects in improving climate change, disaster response, and public health. AI for Good is designed to be useful whether or not you have coding experience. It does include Python code examples that you can execute and interact with to gain deeper insight into different applications. Please check it out! deeplearning.ai/courses/ai-for…

Not every ML project is bound for success. @WWRob, ML engineer & disaster responder, shares a story of a failed system for maternal health and how it paved the path to create more succesful projects. Enroll in AI for Good to learn from real-life cases: hubs.la/Q021xy170

Twitter just updated its privacy policies to permit it to infer and share your identity with anyone. Like @washingtonpost reports, I believe this will enable oppressive regimes to execute people with “anonymous” accounts with only 10 followers. washingtonpost.com/opinions/2023/…

@LeonDerczynski I agree, but your now deleted tweet was more interesting bc it felt like you were throwing down a gauntlet

automatic llm red teaming is not the right tool for discovering new failure modes, only for discovering new ways of provoking known failure modes. discovering new failure modes needs creative humans

@LeonDerczynski Why can’t red teams be creative?

@AnimaAnandkumar Yes, RLHF seems like branding to cover up and differentiate from established practices in annotation. I’ve already seen multiple researchers abandon RLHf to use more established and more mathematically well-founded approaches to human feedback for LLM tuning

My hypothesis is that more RLHF and fine-tuning destroys the calibration of pretrained models. It is a shallow fix that encourages the models to convince the user that they are following instructions, rather than actually following them.

@JWGrieve @yoavgo 2003’s “Probabilistic Linguistics” mitpress.mit.edu/9780262523387/ @rensbod @jbhay & Jannedy (eds) best shows the link, but in 2003 it was more common to cite Firth’s student, Halliday. The chapters by #NLProc people like @chrmanning & @jurafsky would have been read by word2vec authors

@JWGrieve @yoavgo No, I regularly talked about work on Firthian linguistics at Stanford and Silicon Valley more broadly in the 2000s, including to authors of the lda & word2vec papers. This might be a case of genuinely thinking they invented something they heard - I was an outlier at that time

Hi @yoavgo I’m pretty sure it was you on here who once said the development of word embeddings had no direct link to Firth — this was a link made retroactively. Like Mikolov et al had never heard of him. Don’t doubt you’re right — but do you have a reference? Thanks!

@ryandcotterell @adinamwilliams @ZeerakTalat If you haven’t already, you will be approached by people who profit from extractive industries and use NLP to identify political “enemies”, because this use case has always been NLP’s greatest impact on humanity. I support a new publication option that excludes these people.

Our next meetup introduces two new @DeepLearningAI_ courses! In partnership with @cfregly & @anbarth at AWS and @WWRob of @BayAreaNLP, join us next week to learn more about new courses on AI for Good and Generative AI with LLMs #NLProc meetup.com/bay-area-nlp/e…

@mayhewsw I kept telling people this when I was CTO of the largest data labeling company *and* that paying people fairly was more cost effective in the overwhelming majority of use cases. It would be a net economic benefit to introduce legislation to enforce fair pay for annotators

My favorite takeaway from this presentation was that quality of the dataset is correlated with amount paid to annotators. Water is wet etc but nice to see it said out loud. #acl2023nlp

What a way to kick off this amazing coursework @WWRob @AndrewYNg 🙌🙌 #AIforGood

@DynamicWebPaige SF is a magical place to raise kids

I ❤️ working with machine learning research teams, and can't really imagine doing anything else (other than perhaps running a late-night coffeeshop + computer lab in a town that needs it 😅). But at some point, I really want to raise a family; and there is no way in heck I would feel safe or supported doing that in San Francisco. We desperately need a second large AI community hub that's located somewhere a bit less Mad-Max-y.

@milindjagre The course emphasizes the importance of the human side of ML: we’re fortunate that you’re working on this for us @milindjagre !

Can’t wait to go though this, instructed by my boss @WWRob 😎

I'm teaching new @DeepLearningAI courses on AI and Public Health, Climate Change & Disaster Response. Delighted to highlight: @onucharlesc @BDataScientist @felipeoviedop @IvaGumnishka @jack_kelly @LesterMackey @priyald17 @sarameghanbeery @shahrzaadghlm & @TapiwaChiwewe!

@random_walker It’s only millions FTE on outsourced platforms. More people want to work but there isn’t the demand, esp on platforms that pay fairly. But most annotation is internal to organizations and not necessarily the main job that someone performs, so that part is hard to estimate

Billions of people doing clickwork for AI sounds more dystopian than most of the AI safety risk scenarios that have been proposed tbh

To everyone coming to Dolores park for @TransMarch, my kids are sleeping nearby so … make as much noise as you want. Angry, Happy, Both, any sound of protest or celebration. Especially after this year’s attacks on trans/ngc folk 🩵🩷🤍