@steely_glint@chaos.social
@steely_glint@chaos.social avatar

steely_glint

@steely_glint@chaos.social

I write secure opensource niche realtime social apps
and provide End-to-End IoT security for +2bn endpoints using #webRTC.
I read sci-fi, watch films and try to learn new things.
I Co-host of the “Distributed Future” podcast where we try and figure out what the future at the intersection of tech and society looks like by asking people who might know and listening to what they say.
I play the flute badly.
Find me in Berlin, Manchester, Nowhere
He/Him

This profile is from a federated server and may be incomplete. Browse more on the original instance.

penwing, to random
@penwing@wandering.shop avatar

Given up on Cries of Alaere (the #mcrsf book) and finally found the time to finish The Adventures of Amina Al-Sirafi which I thoroughly enjoyed...

Maybe I should move on with Earthsea again - I'm onto The Farthest Shore 8-)

steely_glint,
@steely_glint@chaos.social avatar

@penwing given up on, as in taking a break from, or not going to pick up again?

steely_glint,
@steely_glint@chaos.social avatar

@penwing 53% currently and taking a break.

steely_glint, to random
@steely_glint@chaos.social avatar

I am rather surprised to find @AnnLeckie ‘s short fiction book “the lake of souls” contains a Wooster-ish story where the Jeeves role is played by a god in the form of a sparrow.

danjenkins, to random
@danjenkins@fosstodon.org avatar

https://openai.com/careers/real-time-communications-engineer/

310-465k usd for an rtc engineer position at OpenAI.

That’s 244k gbp a year…

steely_glint,
@steely_glint@chaos.social avatar

@danjenkins Yeah, but the position will only exist for 6 months - is 120k worth selling your soul for?

steely_glint,
@steely_glint@chaos.social avatar

@danjenkins
"comfortable with ambiguity and rapidly changing conditions," aka morally bankrupt :-)

steely_glint,
@steely_glint@chaos.social avatar

@danjenkins funny thing is I had a really good idea for how to do non-text media driven AI.

steely_glint, to random
@steely_glint@chaos.social avatar

I'm proud to have played a small part in @sundogplanets 's campaign on the space junk pollution (both light and actual) that starlink and it's ilk produce .

https://distributedfutu.re/episode83.html

(We had a good track record of predicting future hot issues on that podcast - reality hasn't caught up with us yet)

steely_glint,
@steely_glint@chaos.social avatar

@danyork Unlikely - We both got busy for different reasons - but never say never.

steely_glint, to random
@steely_glint@chaos.social avatar

So now we can add "ask it a riddle" to the ways that LLM resemble pixies

Quinnypig, to random
@Quinnypig@awscommunity.social avatar

I'm sorry Slack, you're doing fucking WHAT with user DMs, messages, files, etc? I'm positive I'm not reading this correctly.

steely_glint,
@steely_glint@chaos.social avatar

@Viss @Quinnypig The best fun is it will invalidate patents - since 3rd parties will know about your design decisions before you register.

jon, to random
@jon@gruene.social avatar

DB ICE 731 Hamburg - Köln

A slowish route because all the infra is old. And it’s a ICE 1 from the 1980s, renovated more than a decade ago

But the ambience on board, the comfort of the seats, are still excellent. It’s like someone actually thought “what would passengers actually appreciate in a train interior?”

It doesn’t feel like packing as many people in as possible was the priority. And that’s so nice 😊

steely_glint,
@steely_glint@chaos.social avatar

@jon Yep - the space (and the woodwork) on the older ICE trains along with the quality of the ride and the wifi makes it possible to do a productive day's work while going at a reasonable clip across Germany. I fear how they will be replaced.

davidho, to random
@davidho@mastodon.world avatar

When you see the best selling vehicle in America on a French street, you realize how ridiculous it is that we let this infestation happen in the US.

📷: @briceperrin

steely_glint,
@steely_glint@chaos.social avatar

@amszmidt @davidho @briceperrin
https://www.vice.com/en/article/pkbxzg/american-cars-are-now-almost-as-big-as-the-tanks-that-won-wwii Is amusing and sad at the same time.

steely_glint, to random
@steely_glint@chaos.social avatar

Thanks to @saghul for the perfect illustration of the problems with chatGPT:

steely_glint,
@steely_glint@chaos.social avatar

@the_moep Kinda sure, since it replaces kg with kilograms and does the right thing with the plurals - unless there is an American meaning of kilograms that translates as 'falls faster' :-)

steely_glint,
@steely_glint@chaos.social avatar

@scribe @saghul ha, that is probably where the text came from.

1 is only a single bit away from 2 so does not change the textual best match.

Re-ordering the question does change the answer though.

steely_glint,
@steely_glint@chaos.social avatar

@solarisfire @saghul Which makes it a useful test I guess.

steely_glint,
@steely_glint@chaos.social avatar

@solarisfire @saghul could be a halfway house where they upped the weighting of numbers in questions that involve comparisons. - Essentially that is what is 'wrong' with 3.5 it allows the volume/placing of blather in the text to outweigh the difference between 1 and 2

(Ironic that it's a weighting problem ;-) )

steely_glint,
@steely_glint@chaos.social avatar

@quantensalat @saghul Yep, elsewhere in the comments someone pointed out that chatGPT4 gets it right (chatGPT 3 does not), we were discussing wether that's a fix for this riddle or a re-weighting of the importance of numbers in comparisons.

steely_glint,
@steely_glint@chaos.social avatar

@bjoernd @saghul yep, the order matters. Presumably because it matches more of the "feathers or lead" text out there, whereas "lead or feathers" has fewer matches and so the numbers take precedence.

steely_glint,
@steely_glint@chaos.social avatar

@nikolar @saghul Where can it have learned that ? ;-)

steely_glint,
@steely_glint@chaos.social avatar

@ariaflame @the_moep I think there may be special rules for synonyms (like kg and kilograms) - but yep, that isn't the same as understanding.

steely_glint,
@steely_glint@chaos.social avatar

@ami @saghul Which makes it an interesting test I think.

steely_glint,
@steely_glint@chaos.social avatar

@kravietz
I remember prolog - one of the research groups at Uni used it to model some British legislation. They struggled with counterfactuals. "If your grandfather had been alive in 1970, would he have been a British citizen?" required them to build a whole duplicate ruleset with a resurrected grandfather to answer the question, capture the result in the current rules then then discard the alternate universe before carrying on. ;-)

steely_glint,
@steely_glint@chaos.social avatar

@kravietz They also had a sign over their door saying "abandon all Hope ye who enter here" as a dig at the other group who were working on a transputer friendly functional language called Hope.

steely_glint,
@steely_glint@chaos.social avatar

@fedithom It isn't 'known' everywhere - Over on linkedin everyone is talking about how LLMs will revolutionise X,Y and Z . I felt it worth a single query to see if I could get the message across that LLMs are more of a party trick than a solution to any actual problem.

Also are you Canadian ? I've only ever heard a Canadian say "thank you kindly".

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • megavids
  • ngwrru68w68
  • thenastyranch
  • ethstaker
  • khanakhh
  • Youngstown
  • rosin
  • slotface
  • mdbf
  • vwfavf
  • everett
  • kavyap
  • cubers
  • DreamBathrooms
  • normalnudes
  • magazineikmin
  • Durango
  • cisconetworking
  • InstantRegret
  • osvaldo12
  • tacticalgear
  • tester
  • Leos
  • GTA5RPClips
  • modclub
  • provamag3
  • anitta
  • JUstTest
  • All magazines
    •