4 ms·

I would welcome DOGE focusing on how much velocity it will increase and how many lines of bad policy it will remove. Likewise, cutting inefficient people should

I would welcome DOGE focusing on how much velocity it will increase and how many lines of bad policy it will remove. Likewise, cutting inefficient people should instead be replacing them. It is so hard to make progress within gov, so focusing on firing and demoralizing makes the environment even more afraid to move, and inefficiency even more of a self-fulfilling prophecy. As a counterpoint, I'm happy for the pro-AI discussions as these systems have been difficult to move forward. Many senior people are working to figure out how to legally do it, so dictates cutting red tape for them, not headcount, is welcome, and would let them focus on the project phases they actually want to do. We work with multiple groups at the federal level, and while some people our peers interact with I do wish were out of the way, those are a minority, and more of a symptom. The ultimate problem is the procedures being tasked, the underlying mandates & regulations driving them, and lack of ability to succeed despite these. Imagine if your IT department was largely run by lawyers who fear jail time for any sign of malpractice because that's what the law says and they've built up defensive practices over the years as different situations came up. Of course the default is 'no'! The solution is not to stop having an IT department, but to fix the policy problems and rotate leadership for the new culture.

brokenmachine 2y ago
>I'm happy for the pro-AI discussions as these systems have been difficult to move forward. What needs does government have that you think would be better performed by AI?
- lmeyerov 2y ago
  At the federal level, most things are big, underfunded, understaffed, underpaid, and over regulated, so programs are low quality, behind, and slow. I stay away from state level, and expect there is less data and $, and even more problems. Think fairly basic data extraction, data cleaning, workflow automation, moving to the cloud, getting basic modern security.
  - brokenmachine 2y ago
    I'm not trying to be troublesome, just really trying to understand the opportunities... Basic data extraction and data cleaning? You can't trust AI to do anything without hallucinating or ignoring possibly important data. Workflow automation, in what way? Once again, can't be trusted. Moving to the cloud? How does AI help with that? Getting basic modern security. Again, how does AI help with that?
    - lmeyerov 2y ago
      It may help to think about baselines. The status quo is being late or bad, understaffed, low budget, etc. It is not comparing to perfect code written by top CS grads with all the time in the world. Developers need help writing basic code, and for fancier tasks, help writing models better than their bad rules & bad ML. Also, data is frustrating in these contexts. For example, consider data cleaning, where AI is a pretty clear win for spotting bad input errors, performing entity resolution, extraction of dirty data, etc. Federal level is by definition pretty big, and data comes from all sorts of terrible systems, often staples together from a hodgepodge of them. The shift to ML/AI here has been going for 10+ years, eg, Trifacta. A little more out there, a lot of forms and data sources dealing with people is text, not just nicely structured data, so being able to more easily work with that is going to keep a lot of people busy for awhile. Related to the last point, and I stay away from this side, but gov is also ultimately very much a service industry. Automating citizen interactions, and cheapening the cost to do it well, is a big deal.

brokenmachine 2y ago

>I'm happy for the pro-AI discussions as these systems have been difficult to move forward. What needs does government have that you think would be better performed by AI?

lmeyerov 2y ago
At the federal level, most things are big, underfunded, understaffed, underpaid, and over regulated, so programs are low quality, behind, and slow. I stay away from state level, and expect there is less data and $, and even more problems. Think fairly basic data extraction, data cleaning, workflow automation, moving to the cloud, getting basic modern security.
- brokenmachine 2y ago
  I'm not trying to be troublesome, just really trying to understand the opportunities... Basic data extraction and data cleaning? You can't trust AI to do anything without hallucinating or ignoring possibly important data. Workflow automation, in what way? Once again, can't be trusted. Moving to the cloud? How does AI help with that? Getting basic modern security. Again, how does AI help with that?
  - lmeyerov 2y ago
    It may help to think about baselines. The status quo is being late or bad, understaffed, low budget, etc. It is not comparing to perfect code written by top CS grads with all the time in the world. Developers need help writing basic code, and for fancier tasks, help writing models better than their bad rules & bad ML. Also, data is frustrating in these contexts. For example, consider data cleaning, where AI is a pretty clear win for spotting bad input errors, performing entity resolution, extraction of dirty data, etc. Federal level is by definition pretty big, and data comes from all sorts of terrible systems, often staples together from a hodgepodge of them. The shift to ML/AI here has been going for 10+ years, eg, Trifacta. A little more out there, a lot of forms and data sources dealing with people is text, not just nicely structured data, so being able to more easily work with that is going to keep a lot of people busy for awhile. Related to the last point, and I stay away from this side, but gov is also ultimately very much a service industry. Automating citizen interactions, and cheapening the cost to do it well, is a big deal.