robinhoustonMay 8
User cachecrab on Twitter noticed that Google has a rather unusual interpretation of the query “20% of £1k”.
Show threadSimon TathamMay 8

@robinhouston reminds me of a friend who once asked Wolfram Alpha for a graph of "tan tan x", and got a graph of tan x, drawn in a light brown colour.

I think they fixed it, though. (Disappointingly.)

Show threadrobinhoustonMay 8
@simontatham Wolfram Alpha used to have some utterly hilarious misinterpretations! I wish I could find examples.
Mos2d ago
Show thread𝓙. 𝓜.
@robinhouston @simontatham this is the one I remember quite well:
May 8 at 5:10pmWeb
Show threadrobinhoustonMay 8
@tpfto That was a classic! Thanks for digging it up.
Show thread𝓙. 𝓜.May 8
@robinhouston I sometimes get the feeling that the whole of Alpha is a collection of perfectly okay mathematical methods, held together by duct tape and baling wire. That said, I find that when they announced that ChatGPT can interface with Alpha some time back, it seemed like LLMs were the long-missing front end for Alpha.
Show threadPozorvlakMay 8
@tpfto @robinhouston I'd like to know how Alpha worked before LLMs! Probably, as you say, duct tape and baling wire.
Show threadjan Ki | 奇  2d ago
@pozorvlak @tpfto @robinhouston
classic NLUI, much more powerful than any LLM could ever dream to be
Show thread𝓙. 𝓜.2d ago
@ki well... what I *am allowed to* say about how the business was implemented is that there's a whole lot of pattern recognition going behind the scenes. Since you mention NLUIs: deciding what to do for ambiguous input was part of why I joked about the thing being "held together by duct tape and baling wire" (there's a fair bit of weighting involved, but I am not going to say anything further). (cc. @pozorvlak @robinhouston)
Show threadjan Ki | 奇  2d ago
@tpfto @pozorvlak @robinhouston
can't be as bad as what I'm used to in my day job :') *cope*
Show threadEmber is so tired of people. ​2d ago
@tpfto @robinhouston better something that always succeeds on some queries and always fails on others than something that fails most of the time and succeeds sometimes, randomly
Show thread0xC0DEC0DE07EA1d ago
@tpfto @robinhouston yeah, one could imagine taking wordvec inputs to your expert system to help disambiguate things (or enumerate possible interpretations and have a secondary prompt about what you meant). The LLM chatbots though, blech.
Show threadSundewMay 13
@tpfto @robinhouston @simontatham Who doesn't like to count in base 20 on their fingers and toes?
Show threadgebastel 2d ago
@tpfto @robinhouston @simontatham
https://www.google.com/intl/de/help/currency_disclaimer.html
Google-Hilfe: Haftungsausschluss bei Währungsumrechnung

Show threadSigma2d ago
@tpfto @robinhouston @simontatham 🤣 That just made my day! ^^
Show threadLuna Ruby Artemis 💛🤍💜🖤2d ago
@tpfto @robinhouston @simontatham Dwarf Fortress when a dwarf tries to punch and kick a megabeast.
Show thread0xC0DEC0DE07EA1d ago
@tpfto @robinhouston @simontatham less hilarious when the all-powerful supercomputer sends you a box full of those digits taken from (also random) people.