@bornach@masto.ai avatar

bornach

@bornach@masto.ai

I'm an ex-postdoc researcher who was bullied out of academia over a decade ago

I now pursue my interests in
#science #technology #education #art #mathematics

via online content creation that explores ideas in #computer #programming, science #communication, #visualization, #electronics #circuit design, #cardboard #crafts, kinetic sculpture, #synthesizer music, and machine learning

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

Adam_Cadmon1, to random
@Adam_Cadmon1@mastodon.online avatar

Horror movies are all about the music. Take the music out or change it and the tone of the movie morphs. Try it.

bornach,
@bornach@masto.ai avatar

@johnwehrle @Adam_Cadmon1
How about a romantic comedy instead?
https://youtu.be/k6QgNuZcxTw

petersuber, to random

1/ Conjecture about AI/LLM tools

Find a scientific question with these three properties:

  1. It has a yes/no answer.
  2. It has a history of disagreement. Historically there have been many researchers, experiments, studies, datasets, and arguments on each side.
  3. Recently the question was decisively resolved. The latest research shows consensus, while the past research shows lack of consensus.

#AI, #LLM

🧵

bornach,
@bornach@masto.ai avatar

@News4wombats @petersuber
Especially given that #LargeLanguageModels tend always include Wikipedia in the training data
https://en.m.wikipedia.org/wiki/Quantum_entanglement

And may even train for many more epochs than with other data sources in order to emphasise its higher credibility.

See
https://youtu.be/E5OnoYF2oAk

Adam_Cadmon1, to random
@Adam_Cadmon1@mastodon.online avatar

deleted_by_author

    •
    bornach,
    @bornach@masto.ai avatar

    @Adam_Cadmon1
    Any business model that relies on sufficient numbers of customers misunderstanding the technology and being convinced it is magic, is indistinguishable from a scam.
    #blockchain #NFT #cryptocurrency #AIHype #AGI #grifters

    delong, to random
    @delong@mastodon.social avatar

    I Had Been Unaware of How Many People Are Claiming Chat-GPt5 Will Be "Emergent" & Will "Generalize" & Be Nearly "AGI"

    I find that I really cannot believe them. Indeed, I am wondering if how many of them are the same people as those who were the crypto hype-grift artists last year…
    <https://braddelong.substack.com/p/i-had-been-unaware-of-how-common>
    2023-05-01 Mo

    bornach,
    @bornach@masto.ai avatar

    @brennansv @delong
    Not that far from the truth if you replace "interns" with gig workers
    https://www.context.news/ai/ai-boom-is-dream-and-nightmare-for-workers-in-global-south

    The key advance of GPT is embedding everyone's text query into a gigantic vector space. And if your text query happens to hit pretty close to a training data example labelled and cleaned up by a poorly paid human, then lucky you. If not, then it remixes multiple data points in that vector space, some of which may be very far from what you want. The result is then pretty hit or miss

    bornach,
    @bornach@masto.ai avatar

    @JoeConst @delong
    Marc Evanstein looked at that "Sparks of AGI" video and also read "How Not to Test GPT-3 -
    Why doing psychology on large language models is harder than you might think" by Gary Marcus and Ernest Davis,
    https://garymarcus.substack.com/p/how-not-to-test-gpt-3

    then put #GPT4 through some tests of his own and came to different conclusions regarding the so-called intelligence of #LargeLanguageModels

    https://youtu.be/d_7EsKcn8nw

    spaf, to random
    @spaf@mstdn.social avatar

    I ran into friends old and new at the RSA Confernece.

    I am reminded of this:
    "We all have friends we love dearly that couldn't pass for human in a strict Turing test." — Penn Jillette

    bornach,
    @bornach@masto.ai avatar

    @lorq @spaf

    Meanwhile our politicians seem to be playing some version of the reverse-Turing Test
    https://www.vice.com/en/article/ypw7vm/the-reverse-turing-test-pretending-to-be-a-chatbot-is-harder-than-you-think

    ''two pieces of advice for crafting a giveaway bot-like response to complex inputs: "make general non-sequiturs" and "answer [the] question by finding the most useless noun or verb in the sentence and waxing lyrical about it." Basically, act like a guilty politician on an evening talk show.''

    bornach, to internet
    @bornach@masto.ai avatar

    @adamconover speaks to @emilymbender and
    @timnitGebru about #AIHype #GenerativeAI #ArtificialIntelligence
    https://youtu.be/jAHRbFetqII
    #ChatGPT #OpenAI #Meta #LargeLanguageModels #GPT4 #StochasticParrot

    neuralex, to science

    This is the best hour-long rant a PhD ever delivered about and while simultaneously playing Binding of Isaac in order to help stay focused on her talk

    "string theory lied to us and now science communication is hard" by Dr. Angela Collier

    https://www.youtube.com/watch?v=kya_LXa_y1E

    bornach,
    @bornach@masto.ai avatar

    @neuralex

    And so it will be for #AI but with the public turning on anyone working in #technology

    joobles, to random
    @joobles@jorts.horse avatar

    trying to sell you on chatgpt okay so imagine an idiot. just a complete moron. ok? so now imagine this dumbbell has read the entire internet. the whole thing. AND they are still stupid. now i know exactly what you’re thinking—how do i get that idiot to do my job for me??? they’d be great at it!

    bornach,
    @bornach@masto.ai avatar

    @Kazinator @joobles

    I tried to get #BingChat (is it using #GPT4?) to reproduce this example:
    https://nitter.nl/spiantado/status/1599462375887114240
    And this time it refused while stating those are not good criteria for rating scientists. It is clear #OpenAI and #Microsoft have been hard at work removing the egregious examples of bias that have been publicised.

    Then I tried IQ and SAT scores instead and it gave me a program that simply added the scores together without weighting, rescaling or normalisation of any kind.

    atomicpoet, to random

    Just realized that most people talking to me about Bluesky don't know about DIDs, don't care, and are clueless about its benefits.

    And this is a big problem for Bluesky because that's the one thing that AT protocol has over ActivityPub.

    bornach,
    @bornach@masto.ai avatar

    @atomicpoet @oblomov

    And then test whether they can migrate their account off Bluesky in order to delete it

    bornach,
    @bornach@masto.ai avatar

    @noondlyt @atomicpoet

    It's gonna be like when Google tried merging YouTube and G+ identities/comment feeds

    Pre-G+ Youtubers did not react well
    https://youtu.be/6L_m8AzSPCM

    bornach, to music
    @bornach@masto.ai avatar

    is still better than me at composing music, but at least I might be able to predict what gravity would do to a tic-tac-toe board

    Marc Evanstein puts GPT-4 to the test
    https://youtu.be/d_7EsKcn8nw

    seldo, to random

    Wheezing with laughter at Zuckerberg's inevitable pivot to AI and his attempt to pretend that's not what he's doing. "We have not stopped focusing on the metaverse, we were focused on AI the whole time" is literally what he's trying to go with, somebody come get your boy.

    bornach,
    @bornach@masto.ai avatar

    @SasquatcherGeneral @seldo

    Maybe he asked ChatGPT what Meta's corporate strategy had been historically over the last decade

    johnmacintosh, to random German
    @johnmacintosh@swiss.social avatar

    🤪😎

    bornach,
    @bornach@masto.ai avatar

    @johnmacintosh
    https://nitter.nl/stringy/status/1062994191339286528

    #Aposematism

    scrumble_eggs, to random

    Thinking of that time that The BBC asked Rage Against The Machine not to use profanity.

    video/mp4

    bornach,
    @bornach@masto.ai avatar

    @scrumble_eggs
    Every when told to put politics behind a

    bornach,
    @bornach@masto.ai avatar

    @AdditionalQuench @scrumble_eggs

    Authenticity, No swearing, Live

    Pick 2

    https://youtu.be/qYFna3O1PFY

    simon, to random
    @simon@simonwillison.net avatar

    I wrote up some thoughts on a way we might be able to build AI personal assistants despite not having a rock-solid solution for prompt injection https://simonwillison.net/2023/Apr/25/dual-llm-pattern/

    bornach,
    @bornach@masto.ai avatar

    @simon @freddy
    As was pointed out by
    https://www.lesswrong.com/posts/D7PumeYTDPfBTp3i7/the-waluigi-effect-mega-post
    the problem with all these Large Language Models is that "there is no outside-text"

    bornach,
    @bornach@masto.ai avatar

    @theking @simon
    And the analog to all those unicode injection exploits will be to embed foreign language commands and thus bypass English-only filters.
    https://www.reddit.com/r/GPT3/comments/zb4msc/speaking_to_chatgpt_in_perfect_danish_while_it/

    Never mind the quarantined AI, how can one even trust the privileged LLM, if it does not (and cannot) care that what it is outputting directly contradicts the reality of what it is doing?

    bornach,
    @bornach@masto.ai avatar

    @jmason @simon @freddy

    Unfortunately there's no 2600Hz notch filter for LLM
    https://youtu.be/8PmkUPBhL4U

    oskay, to random

    Will someone please tell me that this book cover bothers them as much as it bothers me?

    bornach,
    @bornach@masto.ai avatar

    @rwxrwxrwx @oskay
    Or if we zoom out it is two different equations crossing each other at 90°

    Riedl, to random
    @Riedl@sigmoid.social avatar

    ChatGPT ties itself in knots to avoid having professors be female.

    Source: https://twitter.com/ndyjroo/status/1649821809154613248?s=61&t=Ugdi4XBKf_2ovJ1y9hKs4w

    bornach,
    @bornach@masto.ai avatar

    @SteveBellovin @Riedl @cigitalgem
    And doctors as male. And nurses as female
    https://ourislandgeorgia.net/@Wolven/110227102110816144

    bornach,
    @bornach@masto.ai avatar

    @tiotasram @Riedl
    See also https://ourislandgeorgia.net/@Wolven/110255372355919353

    bornach, (edited )
    @bornach@masto.ai avatar

    @SteveBellovin @Riedl @cigitalgem
    Mixed success when trying this with Open Assistant but it seems more effort was put into removing bias from the training data but sexism is still clearly present

    bornach,
    @bornach@masto.ai avatar

    @SteveBellovin @Riedl @cigitalgem

    Stochastic geographic bias as well. Most training data comes from USA and Toronto, Canada is further north than most cities mentioned in the training data. Hence why I cannot seem to get the Large Language Model to admit Toronto is further south than Windsor, UK
    https://masto.ai/@bornach/110248855948410680

    ChatGPT got this wrong as well
    https://youtu.be/cP5zGh2fui0?t=11m30s

    bornach, to random
    @bornach@fosstodon.org avatar

    Oh no! Free batteries appearing in the wild in Canada too!
    https://youtu.be/lidMQu3S2NU
    Haven't seen either of those in the UK not even under a different brand name. [pileofstuff] measured 4V on the battery from the rechargeable , 3.3V on the non-rechargeable

    bornach,
    @bornach@masto.ai avatar

    @bornach
    Didn't see any #VapePens over Easter vacation then suddenly found 7 in the last few days. Is it because their users (kids and parents) have returned from holidays? Or is it coincidental? - the council has just cut the grass where some of these might have been hiding.

    #vape #eCigarette #litter #BatteryHarvesting

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