Scott Miller 🇺🇦 🇺🇸

I was reviewing a PR this evening at work for something urgent.

All of the ~30 new unit tests this person added to cover their edits failed. I looked at the tests and started figuring out what needed to be fixed, and provided the first fix. Then they failed for a different reason.

I asked the person if they’d run any of the tests before submitting the PR. “Well Copilot made them.”

1/x

#ScottProgramming #MicrosoftCopilot

RE: @seldo@alpaca.gold

“Tedious bastards” is a hell of a way to describe people that don’t want their content stolen for use in training LLMs.

It's been said that most people's taste in music gels by the time they are 25, and they stop liking new music.

Mmm.

I did not discover LCD Soundsystem until my 40s.

Don't stop liking new music.

https://song.link/y/V7QaGc2fmmE

#ScottMusic

1. All of this is predicated on knowing about how many tokens a particular task uses, which from what I've seen, the LLMs have deliberately made it nearly difficult to measure the tokens used by a particular session or project. This may change under pressure of business leaders now that they have suddenly become LLM cost-conscious.

5/5

1. Employees often don't know the value produced by their work. And that's often not just because of management is terrible at communicating priorities. For example, compliance work may not generate revenue today but will avoid consequences that may be hard to quantify. Another example is data exploration that might uncover a cost savings opportunity of unknown magnitude... or confirm existing processes are already efficient — how do you value that work before it starts?

2/x

1. If both tasks get done and on time, and both require the same amount of tokens, what difference does it make which one the employee uses the LLM on?

3/x

RE: @scottmiller42@mstdn.social

Remember the dotcom boom pricing model? Sell it at a loss, expand the user base, for years with expectation that economy of scale will lead to profit.

The fact that they are converting to a usage-based pricing model so quickly tells you they are losing their ass on this!!!

Also, for those prior dotcoms, the costs didn't grow so fast by usage (heavily driven by R&D and marketing). #GenAI #LLMs have a voracious appetite for silicon, electricity, and cash!

#GithubCopilot #Copilot

“We’ve always been at war with the Gotoro Empire.”

#StardewValley #GotoroEmpire #AltText

Today, I'm remembering that time on January 6, 2021 when Lindsey Graham indicated he might actually have a backbone by speaking out against the illegal coup attempt and said that Biden legitimately won the election.

It didn't stick though. He returned to his cowardly support of Turnip leading up to the 2024 election.

https://www.youtube.com/watch?v=5SmAfeziSoo&t=124s

#LindseyGraham

1. Imagine two tasks that consume the same amount of LLM tokens, where the LLM cuts the time in half. Task A is tedious, takes 40 hours, and is "low value". Task B is interesting, invigorating, takes 8 hours, and is "high value". Both tasks will be completed on time.

Guidance would be to use LLM to complete Task B, but using the LLM on Task A would save more hours of work, work that is unenjoyable. The employee's mind would be fresher and able to do other work sooner.

4/x

Without intending to do so, the orange shit bag gave us a gift today. Worried about how you might be criticized for your post about a certain special person’s death? Save this message as the template you can use. If you follow this wording exactly, not a single soul can object to your message.

#POTUS47