#estimations — Public Fediverse posts

#AI #Research What I do in these situations is arrive at my #estimations for whatever the #situation is. And I run that one, and then I let AI do it without very many #guardrails. Just enough to stay in the playing surface. If they agree, within 10%-20% differentiation, I'm golden.

RE: https://bsky.app/profile/did:plc:hc7tndm7gduompba65aps75k/post/3mgygpshlw42d

https://www.europesays.com/fr/603288/ Les JO de Paris auraient finalement coûté deux fois moins que les premières estimations #Actualités #auraient #coûte #DE #deux #estimations #EU #europe #finalement #fois #FR #France #JO #les #moins #News #Paris #premières #que #RépubliqueFrançaise

Guerre en #Ukraine : "Plusieurs centaines de milliers", encore plus élevé que les #estimations #occidentales… le nombre de pertes #russes dévoilé par erreur par un #conseiller de #Poutine

https://www.lindependant.fr/2025/05/23/guerre-en-ukraine-plusieurs-centaines-de-milliers-encore-plus-eleve-que-les-estimations-occidentales-le-nombre-de-pertes-russes-devoile-par-erreur-par-12714160.php

Guerre en #Ukraine : "Plusieurs centaines de milliers", encore plus élevé que les #estimations #occidentales… le nombre de pertes #russes dévoilé par erreur par un #conseiller de #Poutine www.lindependant.fr/2025/05/23/g...

Guerre en Ukraine : "Plusieurs...

🎉BREAKING NEWS: Milan discovers the shocking truth that #estimations are wrong—because apparently, tech teams have been oblivious to this for years! 🙄 Meanwhile, managers are left scratching their heads wondering why their spreadsheets aren't psychic. 📉🔮
https://newsletter.techworld-with-milan.com/p/all-estimations-are-wrong-but-none #breakingnews #techteams #managementissues #dataanalysis #HackerNews #ngated

Now, #estimations are largely meaningless. What could we do instead?

Instead of dabbling in fortune telling (which as I said is what estimations are) we can instead decide how much time we’re willing to invest. It represents an upper boundary of time and therefore an upper boundary of how complex the solution may be. Shorter timeframe = less bells and whistles.

”ShapeUp” (a process framework by a company who shall not be named because they're tech bros, the bad kind) calls this "Appetite" but at the end of the day the name doesn’t really matter. What matters is that it’s meaningful (”this much time is what it’s worth to us”) and useful (”we can only do so much in that time”).

The problem with “Appetite” is that people aren’t used to it. It requires a change in how engineers think (and it requires them to actually think). But compared to #estimates it has at least the potential to be useful.

#SoftwareEngineering #Planning #Tech #TechLeadership #TechMyths #Bullshit

Everyone wants meaningful #estimations. Rarely are estimations meaningful. Estimations become more meaningful the further along the delivery one is. Early it’s a crapshoot and usually wrong. Estimations and fortune telling have more in common than you might expect.

So sure, we can do estimations. The questions are: when and with which level of information? How much confidence do you want in estimations? And let's ignore non-answers such as “as high as possible” because then the answer is that we do an estimation in the last week of delivery because that’s when confidence is the highest.

For me personally "meaningful" begins after technical planning. That is when the team sat down and actually planned out the work. It’s still _imagined work_ at that point - so dragons are still abundant - but it’s a good enough basis to give an educated guesstimate. Before that it’s reading tea leaves. Of course we can do it earlier but then the range (best vs worst case) will vary wildly. If that’s meaningful enough, great. If not, then that’s shit but nothing we can really do about it, unless someone on the team has a working crystal ball stashed away.

Estimations are one of these things were people love to confidently give wrong answers. It’s a self-perpetuating myth that’s largely done because it was always done and because it gives the illusion of control. Maybe that’s good enough? The illusion?

#SoftwareEngineering #Planning #Tech #TechLeadership #TechMyths #Bullshit

'Non-Asymptotic Guarantees for Robust Statistical Learning under Infinite Variance Assumption', by Lihu Xu, Fang Yao, Qiuran Yao, Huiming Zhang.

http://jmlr.org/papers/v24/22-0034.html

#estimators #estimator #estimations