3 ms·

A marketplace of algorithms and "pick your own algorithm" would be interesting. Is there any reason not to do this? Not sure why this is getting downvoted. It'

by memish 5y ago

A marketplace of algorithms and "pick your own algorithm" would be interesting. Is there any reason not to do this? Not sure why this is getting downvoted. It's not crazy. Even Jack is pushing for it: "The choice of which algorithm to use (or not) should be open to everyone" https://twitter.com/jack/status/1507146276416098307 https://twitter.com/jack/status/1507146276416098307

capableweb 5y ago
Hard to monetize when people can chose algorithms that doesn't prefer content that makes them stay at the platform longer, angry or happy. Public companies are all about making money.
- mountainriver 5y ago
  Subscriptions?
- hooande 5y ago
  who would choose to see unengaging content? "My algorithm is set to show bland agreement with last week's news"
klyrs 5y ago
Any reason to keep the general public from running arbitrary database queries? Yes, I think there is. For example, regular expressions are turing-complete so you can make a query arbitrarily slow without a single join. edit: complexity lol
- capableweb 5y ago
  You could also just limit a query to max N seconds, and abort it if so. I don't think "marketplace of algorithms" necessarily mean people need to be able to make arbitrary database queries either, but seems besides the point. Having their own DSL for sorting timelines can also solve that problem nicely and within whatever performance requirements they would end up with.
- systemvoltage 5y ago
  > are NP-complete You mean Turing-complete?
  - klyrs 5y ago
    I meant what I said, but I was wrong, thanks.
- fooker 5y ago
  That seems like a very unlikely scenario which can be handled with a timeout.
UncleMeat 5y ago
Several reasons. 1. Absolutely incredibly complex architecture. The recommendation engines are not just single libraries that can be easily open sourced or transformed to a plugin system. 2. Recommendation engines exist to push up relevant metrics. These are either clear wins for users (like reported satisfaction), mixed wins for users and businesses (content engagement), or clear wins for businesses (ad engagement). Most businesses aren't thrilled about subbing in systems that degrade metrics.
jmull 5y ago
I wonder about the feasibility of this. How does an algorithm "that attempts to prioritize nuanced conversations about important topics" work? (Or "to find mind-expanding threads", for "savage dunks" or "thirst traps of hot new snax"? -- other examples from the post.) I suppose you spend some time with existing tweets and ML and develop a model that can produce a score of some sort on these concepts for a tweet, and then run every tweet through the model and present the high-scoring ones. Of course, you can't just look at individual tweet, which in isolation don't mean much, but also at how it fits into a conversation. (For that matter, I'd be interested to see what a model of a conversation is.) Sounds expensive and quite possibly not accurate enough to be worthwhile. It also seems like a major strategic decision to give third parties every tweet by everyone. It seems like the business changes from being what twitter is now to a message routing backend, and these third parties become what twitter used to be. That's a fundamental shift, that probably devalues the company by an order of magnitude, since they would be dissolving the valuable thing they have -- the social network they have. Just doesn't make any sense to me.

by memish 5y ago

capableweb 5y ago

Hard to monetize when people can chose algorithms that doesn't prefer content that makes them stay at the platform longer, angry or happy. Public companies are all about making money.

mountainriver 5y ago
Subscriptions?
hooande 5y ago
who would choose to see unengaging content? "My algorithm is set to show bland agreement with last week's news"

klyrs 5y ago

Any reason to keep the general public from running arbitrary database queries? Yes, I think there is. For example, regular expressions are turing-complete so you can make a query arbitrarily slow without a single join. edit: complexity lol

capableweb 5y ago
You could also just limit a query to max N seconds, and abort it if so. I don't think "marketplace of algorithms" necessarily mean people need to be able to make arbitrary database queries either, but seems besides the point. Having their own DSL for sorting timelines can also solve that problem nicely and within whatever performance requirements they would end up with.
systemvoltage 5y ago
> are NP-complete You mean Turing-complete?
- klyrs 5y ago
  I meant what I said, but I was wrong, thanks.
fooker 5y ago
That seems like a very unlikely scenario which can be handled with a timeout.

UncleMeat 5y ago

Several reasons. 1. Absolutely incredibly complex architecture. The recommendation engines are not just single libraries that can be easily open sourced or transformed to a plugin system. 2. Recommendation engines exist to push up relevant metrics. These are either clear wins for users (like reported satisfaction), mixed wins for users and businesses (content engagement), or clear wins for businesses (ad engagement). Most businesses aren't thrilled about subbing in systems that degrade metrics.

jmull 5y ago

I wonder about the feasibility of this. How does an algorithm "that attempts to prioritize nuanced conversations about important topics" work? (Or "to find mind-expanding threads", for "savage dunks" or "thirst traps of hot new snax"? -- other examples from the post.) I suppose you spend some time with existing tweets and ML and develop a model that can produce a score of some sort on these concepts for a tweet, and then run every tweet through the model and present the high-scoring ones. Of course, you can't just look at individual tweet, which in isolation don't mean much, but also at how it fits into a conversation. (For that matter, I'd be interested to see what a model of a conversation is.) Sounds expensive and quite possibly not accurate enough to be worthwhile. It also seems like a major strategic decision to give third parties every tweet by everyone. It seems like the business changes from being what twitter is now to a message routing backend, and these third parties become what twitter used to be. That's a fundamental shift, that probably devalues the company by an order of magnitude, since they would be dissolving the valuable thing they have -- the social network they have. Just doesn't make any sense to me.