What I wish someone had told me

One of the key things OpenAI does is very small scale model experiments. You can iterate a lot faster if your training run takes on the order of minutes. So vet ideas there, only allow proven ideas to be part of the longer training run.

ML model training has become the new "my code's compiling" except now it takes weeks in many cases. The train then chill mindset that some ML engineers have is not they way to make rapid progress.

"Nobody is laughing now."

—SamA from early 2023 interview, when asked about being the laughing-stock of AI circa 2018.

I think maybe your problem was starting the conversation with the form custodian. Their job is the form. They understand the form. They like the form.

I think you have to go up a level to have a chance of those conversations going anywhere. Someone with a little more perspective can go “oh yeah, shoot, good point that concern is OBE we really don’t need to be using that form.”

I often call bullshit out when I see it with little regard to the consequences, because it just goes against every fiber of my being - but when I call it out, I try to show evidence as to how it's bullshit, and if possible a better way to approach a problem said bullshit is trying to solve. That way I don't just tell people I don't like something, I also present evidence as to why and potentially a solution. Seems like people are more inclined to see your POV this way and not feel attacked.

Fighting is risky, it’s damaging, it’s hard and you’re gonna come out bruised.

It’s not a fight if someone just says yes.

You might not be properly armed. In that case the correct move might be to build your arsenal, recruit allies, etc before attacking

Well, the rub of that quote (seen in isolation at least) is that it's still less stupid.

I think with stuff like this, you can make an earnest attempt to understand why it's necessary, but it's on them to, if not convince you, at least convince you that they have reasons to believe it's necessary.

Everybody has their role to play, and these people are at least ostensibly hired so that you don't have to waste your energy understanding and problem solving certain things like compliance, process, etc. But as someone who's worked in admin for some very flat organisations, i.e., organisations with allergies to beaurocratic procedure, I've always seen it as important that the people subjected to my next brainchild at least have a feeling that I know what I'm doing.

I haven't seen the email thread, but failing to do that and responding negatively is a failing on their part. That said, admin can feel like hearding cats nost days, and sometimes a "polite enquiry into the necessity of a procedure" isn't just a polite request for clarification, but a threat to make the next 6 months of your work a living hell because people refuse to follow something you need to get your work done.

So I guess the prescription in those cases is cognitive empathy and finding the right tone. On both sides.

"a university"

Your problem is that you are an indeterminate number of iterations deep into people not fighting bureaucracy, and now it has completely metastasized. The number of administrators at universities is so ludicrously high that I have to imagine on some level the vast majority of them are aware of how useless they really are, and as such, the removal of bureaucracy is an existential threat to their job.

Consider this an object lesson in the importance of preventing this systematic failure from occurring as early and as often as possible.

In this particular case, yes, the correct choice is to bend like the willow and move on with life. Not only can you not defeat this process, the rewards for victory would be negligible. If the degree of fatalism that seems to entail is too much, consider changing jobs. Contrary to the loud people on the internet, there are jobs that are not just bureaucratic garbage, but you're not going to find them at a university. (They do have some jobs that are important by any measure, and if you value what you are doing because you have one of them, then by all means stay if you like, but they will pretty much all be encrusted with garbage as part of the price of entry.)

I think it’s important to understand the context you’re operating in. Universities are highly bureaucratic organizations. This is not an entirely bad thing; we don’t actually want them to operate like startups (for instance we expect them to prioritize longevity over risk-taking: we expect them to preserve capital, etc). It doesn’t mean you did a bad thing - perhaps you were correct, although it’s also possible that what seems unnecessary to you is necessary for some other reason you’re not aware of.

What concerns me is that everyone got angry including you. I might suggest reflecting on that a little bit. Is there a path where you didn’t get your way but you also didn’t get upset? And where you didn’t upset anyone else? If not, then accepting you work for a bureaucracy is one possibility, and if you dislike that, you could work for a nimble startup instead. The tradeoff is you won’t have a pension. ;)

Bureaucracy is very very difficult to fight if you are not in charge, and conversely wasteful bureaucracy is almost entirely in the domain of fault of the person in charge ( the CEO, the President, etc of a company).

If you are in charge, you can stop it. I do it all the time, and I rather enjoy when someone tells me I need to fill out a useless outdated form to do something, and I say 'So, if I don't, do I get fired?'. The answer is either 'yes, you will', or 'no... um.. I guess this might not really be needed.'

The message to eliminated wasted time and wasted process has to come from the very top, with such authority and clarity that it leaves no room for 'that is how we have always done it'.

Yeah, it's good advice if you're CEO, if you're not you better have nothing to lose and be willing to leave the job even if you think you are a good diplomat. I think your last sentence is often the only play, do just exactly what you need to do to appear to be all in and a model supporter of the "initiative" without putting any real effort in to it. But it's also a skill to know when you can, and when you can't change a course of action. Sometimes something silly is happening that no one cares about and you can just change it by speaking up and taking ownership of it.

I've personally decided to minimally comply. Sometimes I'll just skip steps and if no one calls me out on it then great. If they do... Well, I guess I have to waste more time. Easier to ask forgiveness and all that.

> > Fight bullshit and bureaucracy every time you see it and get other people to fight it too.

>I am curious about others' track record doing this.

I nearly lost my job fighting bullshit and bureaucracy at my current gig and have been relegated to a small corner of the business where I am unable to call out the incessant nonsense coming from the vast army of project managers and other professional meeting attenders that have infected this place. At 54 I’m past my sell-by date from new projects and customers, so my only option is to take the daily humiliation so I can keep paying the rent.

Sell the house and buy bitcoin I guess. This advice seems to assume that high conviction cannot be the result of cognitive biases.

Macroexpand

Just lok at the openai: they not only provide ML models, but also a chat experience, a way to save your preious interactions, an easy way to pay, fine-tunes, and that's all from day one

Steve Jobs was the same—here’s a clip of Ive talking about it

https://twitter.com/trungtphan/status/1545415644425752576?s=...

Instead of doing a lot small things, work on a few high value ones. The definition of "high value" is context dependent of course, for instance if the ones buying a software are not the ones who will using it, the value of UX efficiency when doing bulk operations will not be the same.

"Good, fast, or cheap, pick two"

I took it to mean, don't fight beurocratic battles that you won't win.

For example, fighting RTO at Amazon. Jassy has made it one of his top priorities, and no matter how much I disagree with it, I'm not going to shift the needle

Inertia: the energy required to change direction is proportional to the size of the organization.

The Mythical Man Month comes to mind as one example.

There are also lots of business plans that promise outrageous returns on investment... sometime down the line. But at the end of the day your business is never going to be more valuable than the amount of money your customers give you minus operating costs.

Attempting to operate in a space where 100x more money than you can muster is willing to burn itself operating in the same space.

Brook's Law - "Adding manpower to a late software project makes it later."

This applies to all projects, not just software in my experience. Everyone thinks you just need to find the right expert who will magically just slot right into the team and process and hit the ground running. In practice, the people worth hiring take 4-6 weeks to get up to speed with the help of an existing project member, and the people not worth keeping start spewing out code that is ultimately neither helpful nor correct the moment they have access to your repo.

Innovator’s dilemma. People you hire will grow the current business, rather than sacrifice the current business to find and build the next one

When I think of the laws of physics, I think of things which are reliable at some meaningful scale, and which can be used to make predictions about how things will go in the real world. So, to answer your question, no.

Focus.

What about this makes you say that?

In his defense, he's literally listing the things he says he didn't know.

obligatory https://xkcd.com/1827/

(i dont agree w the implication that sama is primarily a lottery winner, but i just like xkcd)

Actually, even if you understand the last item literally, it is impressive enough to gain reputation. Why did you put it as the last item, thinking it was nothing special?

As far as this particular post from Sam, who cares? Judge something by that something, not by who said it. If it is advice that is good, it is good. It isn't good because it comes from someone you like. It is good because it is good. Content is what matters, not the credentials of the author. Lots of people with great credentials say stupid things... and smart things...