MadeWithML – A practical approach to learning production machine learning
github.com
github.com
Disclosure: I'm a contributor in their open-source community activities and also in the ML space. Feel open to reach out to me if you have any questions.
I put together content around threat modeling, infrastructure attacks as well as ML specific attacks with practical examples (like using adversarial robustness toolbox from Linux foundation).
Details here: https://github.com/wunderwuzzi23/mlattacks
While some lessons can stand to be more well connected to the end-to-end project, the sheer amount of detail on some of the lessons simply don't exist anywhere else. The testing (https://madewithml.com/courses/mlops/testing/) and monitoring (https://madewithml.com/courses/mlops/monitoring/) lessons are clear examples of this.
I don't see anything nefarious. Seems to be a pretty straight-up effort to improve the world.
He says these are his reasons:
While this content is for everyone, it's especially targeted towards people who don't have as much opportunity to learn. I believe that creativity and intelligence are randomly distributed while opportunities are siloed. I want to enable more people to create and contribute to innovation.
It's good to get an overview over the various things one needs to learn. But I cannot imagine being ready to put anything in production after reading through this, so no MLOps exactly.
> data: not enough of the right data.
> cost: the data, compute, storage and talent resources needed for ML allow mostly large companies to benefit from it.
> utility: most ML content you see online are gimmicky tasks that can't extend into valuable real world applications.
> trust: ML needs to be engineered reliably and robustly.
This is a very honest take on ML / AI. I work in blockchain and what drives me crazy is that crypto / blockchain has delivered huge, tangible successes and value (measured easily and accurately via real-time prices) but the HN cognoscenti orgasms over ML and considers blockchain a scam.
IMO the situation is flipped - crypto is proving its success in a massive, revolutionary way while ML / AI is pretty stagnant at delivering world-changing products. It's cool that bots are sorting and picking at the warehouse but the value has been overstated by many orders of magnitude (thus far).
ML is a tool in a box right now, one of many that can potentially solve a class of problems, but it isn't (yet) a paradigm shift in how businesses operate.
- Rent seeking, such as by Visa and Mastercard that continually try and raise merchant fees
- Censorship, such as with OnlyFans and its attempted blacklisting by the banking system
- As a weapon of war, as with the treasury blocking entities via their monopoly of the global reserve currency and their licensing / penalty system of banks regardless of their country
- Debasement of money in general by the money printers
Blockchains, when properly implemented as closely as can be attempted using decentralized engineering techniques, are neutral plumbing for value transfer. Ethereum is owned by no one, thus competitors can use it, and fees can be optionally added by middlemen if their services justify the fees. The benefit to the world is choice in the world of finance, something society has never had until recently.
However, all those are specifically only useful for payments/currency. So, if blockchain is only nifty for distributed currency, that's a use case. But hardly lives up to the hype you bring forward. ML can solve problems I face. Blockchain cannot.
It seems to me that the listed issues are actually actions that have been taken by central authority systems. This isn't about trust, it's about historical behavior, and an incentive and power scheme that promotes those types of autocratic and exploitive actions.
People still buy crypto with dollars and they have shown an obsession with the dollar value
This is an unfair criticism. When I travel to a different country I think about their currency in terms of my home country's currency.
People are "obsessed" with the dollar value because crypto is not currently the main currency/asset/payment method/whatever you want to call it.
Proof of work crypto currencies seem like they are currently a net-negative because of the energy consumption. What’s the benefit to society that has already happened?
For AI better speech recognition is widely used and has a ton of benefits. For example nearly all radiologists use voice dictation in order to write faster/better/more reports.