Hacker News Reader

Comment by danielheath | Hacker News Reader

We already had to beat "I made a spreadsheet", which continues to be pretty damn hard even for large teams of experienced engineers - ask your finance team sometime how many custom spreadsheets they use regularly.

A) Lots of useful apps aren't a great fit for a spreadsheet. AI seems to be opening many of those up the same way.

B) Lots of spreadsheets have bugs which cause then to give wildly inaccurate results, which are relied on to make crucial decisions. AI is also repeating this part of the pattern.

If you need it to work correctly all the time, there's still no substitute for expertise - but looking at the state of computing, clearly many people are willing to use things that have obvious, serious bugs.

View on HN

The problem is management and short-term stock-holder interest. If some spreadsheet cowboy convinces some C-level suit that this is the way to go, and it works now, not in 6 months time, IT and development will be made to bear the brunt. Same with AI. The most ardent supporters of its use seem to be the higher-ups, $$$ in their eyes. Convinced of their superior decision skills, they can wreck the place, or in their view, rightsize the help desk/IT/dev departments. Then they get a bonus for delivering cost reductions, and move to another job before being held accountable.

Exactly! This is what I always say when people claim "AI won't replace jobs because jobs still need expertise !" While the cause statement is absolutely true, the people with the power/money don't believe it to be true (or, in some cases, don't care if it's true) and can get a quick win for the shareholders.

I wonder if there's gonna be a big reckoning of these "AI-first" companies suddenly people to fix their shitty vibe coded stuff. Ooh, new job title: AI fixer, minimum salary, $1m/yr, with RSU.

We have an actual department that just supports spreadsheets and power bi reports for the c suite. That's all the poor bastards do.

At least they can put "data scientist" on their linkedin profile.

I've already seen (second hand, through my wife) a complete cycle of this:

- LLMs make writers (for the kind of formulaic crap we ghost-write) unnecessary. We're firing 75% of our writing + editing staff, mostly the writers. Editors will now use LLMs to do the whole thing.

- What do you mean even our best editor and best LLM-user is still taking 90% as long as the total writing + editing process before, and their parallel-work capabilities are down, so throughput is down to a trickle and also every remaining worker is pissed off about how intensely frustrating working with LLMs is so the whole team's grumbling about leaving? Of fuck, we committed (without asking you, naturally, and ignoring a ton of explicit warnings that this was a growing problem) to deadlines!

- Hires back nearly all the writers as contractors

Maybe I don't know enough about spreadsheets but two dimensions isn't enough for most applications. Maybe pivot tables? They are too hard to figure out. Need something like "SQLSheet" that takes a more complex data structure and presents viewing and editing it in a natural way with drill down and joins etc. AI should be able to help you design the DB and then create a tool to interact with it.

Microsoft Access?

If someone is going to invest the time in learning how to design a database and how to build a UI then they might as well do so on something more modern.

The people using these sheets have no idea what you’re talking about. They use multiple sheets to layer the dimensions and understand pivot tables perfectly.

Are people who understand relational databases and people who understand pivot tables disjoint sets?

I can look at someone’s finished pivot table and reproduce it from the data through other means, but any explanation of what a pivot table actually is and does reads like pure gibberish to me.

Probably. A pivot table is basically a way to turn on one of the dimensions of the sheet to make sense of the data. Like "show me all invoices, grouped by date and sum each group". It is effectively a query, in a way that makes sense for people working in spreadsheets.

It is certainly much easier to shown than to tell about pivot tables. I tried to think of a good form of words to explain them but couldn't.

The pivot tables don't require people to understand data normalization and software maintenance good practices. Outside of that, yeah, not really relational databases because those focus on having more than one table, but they do understand relations.

Check out the full version of Towards Scalable Dataframe Systems from VLDB 2020 [0]. They propose an algebra for dataframes and section 4.4's example succinctly describes the pivot operator.

[0] https://arxiv.org/abs/2001.00888

Reply on news.ycombinator.com