Raghubansh Mani

Meet GitHub Copilot - your AI pair programmer. copilot.github.com

@rob_england AlphaZero trains against itself and finds new moves without any human input. Humans that wrote the software are not capable of making 3300 elo moves. Pretty simplistic to compare it with a telephone, no?

@rob_england If some system figured out a new move after studying the position and evaluating options, without any outside help during runtime, I would call it intelligent, whether it be human or software. If a kid did it, we would call them a prodigy, I don't see much difference.

@rob_england The way I see it, moves that AlphaZero comes up with are new knowledge, not something humans could conjure with their intellect. If some delta is added to human knowledge, and humans didn't come up with it directly, I would attribute some learning attributes to the AI.

@rob_england Since no human has breached 3000 elo, they can't teach it to AlphaZero. It has to learn on its own. An argument could be made though that AlphaZero doesn't really know anything, it just evaluates options fast.

@rob_england AlphaZero does play chess against itself and learn. The best human players can't analyze the moves sometimes, let alone teach it to AlphaZero. It's similar to how a sports coach can teach stuff but the disciple can do stuff coach can't, because of sheer natural ability.

Is code the best way to enforce principles?

@rob_england @scotthannen @KentBeck Lol. I admit it's a second hand story. A friend of mine said it happened to them. New Relic was the vendor in question. They couldn't prevent the outage because no one believed it. So, can't really speak to it's overall effectiveness.

OH: “You can’t manage what you can’t measure.” “Really? Why are you employing incompetent managers?”

@rob_england @KentBeck @scotthannen I agree. I am not asking for an unsupervised manager robot. I am asking for a manager's assistant. All decisions should definitely be attributable to the manager, not the robot. Start with simple tasks like Standups, on to maybe Sprint planning, prepare code review reports, etc.

@scotthannen @KentBeck @rob_england These days there are systems that successfully predict the next service outage. I don't think anyone would feel bad about heeding that prediction. I believe similar systems should exist for management too. That's all. Help me be a better manager without me investing as much time.

@raagmani @KentBeck @rob_england The danger is either taking human decision-making out of the loop or if humans yield their authority and follow whatever the software recommends, which has the same result.

@scotthannen @KentBeck @rob_england That does sound horrible. It's a risk with all AI systems. The outlier events evoke disgust. People using it have to be trained to be able to interpret and apply themselves where needed. However, the software must have been creating some value most of the times?

@raagmani @KentBeck @rob_england I read an experience some time back about a customer service phone rep. Customers loved her personality, but software judged that her voice lacked emotion. Computers judged a human's humanity. I find that obscene.

@scotthannen @KentBeck @rob_england How do you feel about people managing people with some help from a computer? I feel it will be a good thing. Instead of me keeping track of the nature of my code review comments for a particular dev, if my assistant kept track of it, maybe prepared a report, it would be awesome.

@KentBeck @raagmani @rob_england Automating management isn't a terrible idea - it's a terrible reality. Amazon employees are managed by robot overlords. I joke about it but I fear that it is the future. It's creeping in slowly enough that we accept it, but IMO computers managing people is obscene.

@KentBeck @rob_england @scotthannen I offer a different perspective - complex tools can remove the need for "management activities" and free up that time for more mentoring. I can think of a lot of scenarios where having a computational assistant can make my job easier and make me a better manager.

@raagmani @rob_england @scotthannen That sounds like a terrible idea. Simple tools are sufficient for good managers. Complex tools make bad managers worse by insulating them from humanity—theirs and that of their reports and customers.

@scotthannen It would be amazing if it turns out to be any good. Machine estimation can be treated as a recommendation, a good starting value, if not the final value. What all do you think needs to go in as input? A good PRD would be essential. Can the model output Universal Story Points?

I wonder if Microsoft could make Azure DevOps guess the number of points in stories using machine learning. I think it would be as accurate as humans and make the same guesses so we could spend less time estimating.

Realizing that a more sensible definition of "low-code" may actually be "you do get to write some code but you don't ever have to learn git"

@scotthannen @KentBeck This is very interesting. Have you seen any progress made on this front? As a dev manager, I would surely love to automate certain parts of my job, and get more time to focus on code and users. Project management activities are first in line.

@KentBeck Management that depends on measurement can be automated. I hate the idea of software managing people but I would love the irony of their jobs getting absorbed into Jira.

@nyootron @kunalb11 Though startups differ from all board games on one very crucial aspect - both parties don't start with equal resources. It's more like an odds match. A startup founder starts with one king and one queen perhaps, and has to play like @GMHikaru to beat his opponent.

@nyootron @kunalb11 Chess teaches these things too. The one with more space on board has more control and better coordination of pieces, usually resulting in better positions. Making connections, battery or Alekhine's gun, is a powerful tool.

Ambitious founders have limitless TAM if they learn the craft of sequencing and chess moves.

Nirvana.Work: Put your Sprints on Autopilot betalist.com/startups/nirva…